Welcome

In the last issue, our serverless expert was AWS Community Builder Jason Conway-Williams, and our spotlight fell on AWS Serverless Hero Luca Bianchi!

This issue, our expert is AWS Hero Bhuvaneswari Subramani, our spotlight is on AWS Principal Solutions Architect Prasad Rao, and we look at the latest AWS service releases, blog posts, hints and tips, news, and more!

Subscribe now

📰 Articles that caught the eye

Here are some stand-out articles I read during the past two weeks in the World of Serverless, AI, engineering and architecture!

Andres Moreno covers ‘Adding Semantic Search to an Existing DynamoDB Table with Vector Indexes’, retrofitting native DynamoDB vector search onto an existing serverless recipe API using Bedrock Titan Text Embeddings V2.

Jacob Clark covers ‘The sticker price of AI is not the true cost of AI-assisted delivery’, on why token rate cards mislead, and why the number that matters is cost per correctly delivered outcome across the whole loop.

Lefteris Karageorgiou covers ‘Securing Your AWS Account in the First 24 Hours (A Checklist)’, a 23-step checklist across root lockdown, IAM, logging, network defaults, account protections and incident response.

Vadym Kazulkin continues his series with ‘Orchestrating workflows with Amazon S3 Files and AWS Lambda durable functions using Java SDK – Part 2 Durable synchronous and asynchronous steps’, building durable sync and async steps with the Lambda Durable Execution SDK for Java.

Danielle Heberling covers ‘Not All AI Builders Are Doing the Same Work’, separating using AI to build, building with AI, and building AI, and where the demand for each is heading.

Stephen Mc Gowan covers ‘You can’t install an AI-SDLC’, eighteen weeks of data from one human directing agents on a production system, arguing governance rules have to be grown from observed agent failures rather than bought in.

Sheen Brisals covers ‘Are You A Human Agent?’, a warning not to get so absorbed in the tooling that you lose sight of the humanity and the wider implications.

Addy Osmani covers ‘Practical Loop Engineering’, on goals and loops across parallel agents, what to delegate versus what to watch, and the discipline of not delegating your judgement.

Jannik Wempe covers ‘Optimizing Your Blog for AI: What Actually Works (And What Doesn’t)’, grading GEO tactics as verified, plausible or refuted against the research and crawler docs, using his own llms.txt and markdown endpoints as the worked example.

And finally, Nick Tune covers ‘Enforced application architecture for agents and humans’, on deterministically enforcing package, layer and role-based rules in an NX TypeScript monorepo so agents simply cannot violate the guidelines.

🎓 Ask the Expert

Opinions are the authors’ and do not express the views of their employer.

1. What is one common mistake you see teams making when building their solutions, and how can they avoid it?

The most expensive mistake I see — and I see it constantly now that every team is racing to adopt AI — is starting with the solution instead of the intent. Teams open the IDE, or these days open a coding agent, and start generating code before anyone has written down what they are building and why. The AI happily amplifies that mistake: you get more code, faster, in the wrong direction.

The fix is discipline, not tooling. Capture the intent as a spec first — the business outcome, the constraints, the edge cases — and make that the artifact everyone (humans and agents alike) works from. When the spec is the source of truth, AI becomes an accelerator; when the prompt history is the source of truth, AI becomes an archaeology problem six months later. This shift is the heart of AI-DLC, and it's the single change I'd ask any team to make before they scale their AI usage.

2. Which tool, package, or AWS service are you most excited about right now, and why?

Kiro, without hesitation — specifically spec-driven development with Kiro. Last year my five-member team won the $100K AWS Road to re:Invent Hackathon building "Orik" (the evil twin of Kiro!) entirely spec-first, and pitching it in 60 seconds to Dr. Swami Sivasubramanian, VP Agentic AI, AWS and Jeff Barr, Chief Evangelist AWS.

What convinced me wasn't the win — it was watching how the quality of our specs, not the quality of our prompts, determined the quality of the product.

Beyond the hackathon, I've spent months delivering both greenfield and brownfield enterprise projects with Kiro — all of them grounded in AI-DLC and vibe coding, starting fast with vibe coding and maturing it into disciplined AI-DLC rules. I ended up writing it all down in The AI-DLC Handbook: Reimagining Software Development with AI (aidlcbook.com) — the practitioner's guide to AI-DLC, the methodology AWS invented and open-sourced. When the way we build software changes this fundamentally maybe once in a generation, that's worth being excited about.

3. What is your favourite trick or tip that the readers may find interesting?

Can I share three, from the scars of building serverless systems that had to survive production?

Go async whenever possible. The moment you chain synchronous calls — API Gateway to Lambda to Lambda to a downstream API — you've rebuilt a monolith with extra latency and a bigger bill. Model your domain as events on EventBridge and let Step Functions own anything long-running. Your system becomes more resilient, cheaper, and far easier to evolve, because new consumers subscribe to events instead of forcing changes on producers.

Design for retries first. Most serverless systems operate with at-least-once delivery semantics, so duplicates are expected. Build idempotent handlers, use DynamoDB-backed idempotency keys, and configure DLQs or Lambda destinations from day one. AWS explicitly recommends designing Lambda functions to safely handle duplicate events. The teams that skip this find out during their first production incident, at 2 a.m.

Write less Lambda. The best Lambda function is often the one you delete: API Gateway, Step Functions, and EventBridge can all call AWS services directly, so reserve your functions for real business logic and let direct integrations do the plumbing. Less glue code means less to secure, patch, test, and pay for — and when you do write functions, Lambda Powertools for structured logging, tracing, and metrics is non-negotiable. Your future on-call self will thank you.

Go check out the book here: https://aidlcbook.com/

Also check out: “Serverless Integration Patterns“ by James Eastham on Study From Experts:

Overview:

Learn practical knowledge for implementing integration patterns with AWS serverless services. Who is this for? This course is for senior engineers who already use Lambda, SQS, and SNS and want to go deeper into production integration patterns. Not for beginners.

🧠 Tips & Tricks

It is an agent skill giving you 39 editorial diagram types as self-contained HTML and SVG, working across Claude Code, Codex, Factory Droid and Pi, with an onboarding flow that matches your brand. It will also redraw existing draw.io and Mermaid sources. The tagline on the repo says it best: editorial diagrams your designer won’t hate. No shadows, no Mermaid slop. It is sitting at 24.8k stars already.

My Take: I use this approach already with SVG over Mermaid, and did this recently when I designed my outdoor office/gym with full architecture plans!

There are some awesome summarised tips in this one if you are working with Claude day-to-day: https://claude.com/blog/maximizing-the-value-of-your-claude-code-sessions

Here is how to optimise your workflow and stop wasting context as a summary:

1. 🧹 Run /clear between tasks: Don't drag old context into a new problem. You're paying to re-read it on every single turn!

2. ⚙️ Set your /model and /effort upfront: Changing these mid-conversation busts your prompt cache, forcing a full-price prefill of your entire session.

3. 📎 @-mention files directly: Instead of typing out file paths, tagging them attaches the file to your message immediately, saving Claude a roundtrip "Read" call.

4. 🤫 Keep commands quiet: Add quiet flags to noisy terminal commands (like test runners) or run them in a subagent. Massive log outputs get permanently added to your conversation history.

5. 🔍 Audit your /context: Run this in a fresh session to see exactly what's loaded from your CLAUDE.md or MCP tools, and trim the excess.

6. 📦 /compact before you step away: Prompt caches expire (after an hour on a subscription, or just 5 minutes on an API key). Summarizing your conversation is significantly cheaper while the cache is still warm.

What tips and tricks do you have?

Leave a comment

🚀 New Releases

Here are the latest and most interesting releases over the past two weeks in the AWS World.

• Amazon Bedrock announces reduced pricing for OpenAI GPT-5.6 Sol.

• Amazon Aurora DSQL now supports Amazon CloudWatch Database Insights.

• Amazon SES now supports open and click tracking override parameters.

• AWS Glue 6.0 delivers 30% price reduction and Iceberg v3 support.

• Amazon CloudFront now supports Origin Access Control (OAC) for Amazon S3 Multi-Region Access Points.

• Amazon DynamoDB Streams now supports attribute-based access control.

• AWS Lambda MicroVMs is now available in 5 additional AWS regions.

• Launching External Web Access for Web Search on Amazon Bedrock.

• Amazon Bedrock now supports SpaceXAI Grok 4.6 with Cross Region Inferencing.

• AgentCore payments is now generally available in Amazon Bedrock AgentCore.

• Amazon Bedrock expands API support and introduces Cross Region Inferencing for OpenAI models.

• Daybreak Red and Daybreak Blue from OpenAI are now available to eligible customers on Amazon Bedrock.

• Amazon CloudWatch Alarms now supports wall clock evaluation windows.

• Amazon OpenSearch Serverless now supports up to 10,000 collections per collection group.

• Amazon Cognito now available as a skill in the Agent Toolkit for AWS.

✖️ Social of the Week

This week’s social is this post by Volkan Özçelik, which opens with “Teams of 3-4 engineers max. Every engineer…”.

“teams of 3-4 engineers max;
every engineer owns one epic end-to-end;
managers exist to bulldoze bureaucracy;
everything in git; no Jira, no Confluence;
specs are code; they are treated as code; reviewed as code;
if it isn’t in source control, it doesn’t exist;
every manager uses AI and writes code.

^ That’s where the new world is converging to.

Being a part of it, or sticking to what’s comfortable is up to you.”

My take: I agree with a lot of what has been said on this one. For the most part, in our AI-DLC we are working in pods of 2-3 people, with one orchestrator (a Principal Engineer or Architect, a senior engineer, and potentially a junior). We are full stack, and this works great. We also call ‘epics’ Unit of Work items (to distinguish from the old SDLC), but we do have an engineer own a UoW end-to-end.

We also have collaborative mobbing, no-bullshit SDLC-style meetings, and we prefer GitHub Issues to Jira. We do like Confluence, however, and keep our markdown docs in sync with Confluence (technical designs, ADRs, constraints, PRFAQ, PRDs and so on).

What are your own thoughts on this? What is working for you?

Leave a comment

🗞️ AI & Technology News

Here is some of the latest news you may have missed over the past week:

Cursor is now a part of SpaceX

Councils to use AI to meet £37.6m funding gap

GitHub blames 8-hour outage on autoscaling fail and VS Code retry storm

👷🏻 Tools & Frameworks

Check out the latest open-source frameworks, news, and tool updates from the past two weeks.

aws-cloud-wan-blueprints from AWS is a set of deployable building blocks for designing global networks with AWS Cloud WAN, covering network policy best practices with both CloudFormation and Terraform patterns, plus a SKILLS.md for agent-assisted policy generation.

buildkite-gha by Buildkite is a Go CLI and plugin that runs your existing GitHub Actions workflows as native Buildkite jobs, so Buildkite owns the scheduling, logs, retries and UI without a GitHub Actions run ever happening. Experimental and pre-1.0.

sample-agentcore-behavioral-contract-evaluations from AWS adds process correctness to Amazon Bedrock AgentCore Evaluations, with a Lambda evaluator checking YAML behavioural contracts (which tools, which sources, what step order the agent actually used) and writing audit records to S3.

docling-graph from the Docling project, brought to you by IBM, is a Python toolkit that turns unstructured documents into validated Pydantic objects and then into directed knowledge graphs with explicit semantic relationships, exporting to CSV, Cypher and other formats.

OpenMausBot by Milind Soni is a local-first, open-source alternative to Grok Bot: an Electron chat app where each contact is a real agent running on your own claude, codex or grok CLIs, with per-bot models, cloud or local execution, and approval gates.

watermarks-remover by Guillaume Meyer is an agent skill plus a stdlib-only Python service that strips multi-vendor AI provenance marks, covering invisible Unicode carriers, statistical text watermarks via rewrite hooks, and C2PA, EXIF and XMP metadata across a long list of file formats.

deepseek-harness from DeepSeek AI is an open-source agent harness (dsh) built on a plugin-everything architecture powered by Cordis, which you run with npx @deepseek-ai/dsh web for a local Web UI. Developer preview for now.

devops-interns-proj is “Daig”, a deliberately breakable three-tier food-delivery platform used as a DevOps, SRE and DevSecOps training substrate, with planted defects, chaos exercises, a capstone assignment and a full curriculum.

And finally, agent-skills by Addy Osmani is a pack of 24 production-engineering skills, 8 slash commands and 4 agent personas that give coding agents verification-gated workflows across spec, plan, build, test, review and ship.

😂 Just for Fun

This issue, our just-for-fun post is this one on LinkedIn, which opens with “No way. There’s now a GitHub status page flavored candy…”.

My take: I feel for GitHub having to double down on hosting all of the AI slop out there…

Leave a comment

🎙️ YouTube & Podcasts

Here are some of my favourite videos and podcasts over the past two weeks covering serverless, AI, architecture, and software engineering.

The Pragmatic Engineer host Gergely Orosz is joined by Charity Majors, CTO and cofounder of Honeycomb, in ‘Stop being skeptical about AI for development with Charity Majors’, on how AI changes the economics of generating code, why verification and reliability are the new bottlenecks, and what engineering discipline looks like around non-deterministic systems.

Sedai hosts Luca Mezzalira, author of Building Micro-Frontends, in ‘The Man Who Wrote the Book on Micro Frontends’, on drawing micro-frontend boundaries that survive, why shared state is the expensive mistake, and when you should just stay monolithic.

Patrick Akil on the Beyond Coding channel sits down with Wes Bos for ‘Wes Bos: How Developers Stand Out When AI Writes the Code’, on agent loops, the limits of generative UI, what AI coding actually costs, and the shift from syntax to problem-solving.

Johannes Koch on Build. Automate. Ship Smarter in the Cloud. is joined by Darko for ‘Does Kiro Crew hold it’s promises?’, a deep dive into Kiro Crew as an open-source agentic dev workspace over ACP.

Lenny’s Podcast features Ian Silber, Head of Product Design at OpenAI, in ‘Chatbots are not the final interface: OpenAI’s Head of Design on what’s next’, on designing ChatGPT and Codex, how he hires designers, doing less, and where AI interfaces go once we stop putting everything in a chat box.

Luca Mezzalira on Dear Architects is joined by David Boyne, creator of EventCatalog and EDA Visuals, for ‘David Boyne on Event-Driven Architecture, Event Catalog, and Why Context Is King for AI’, on practical EDA at scale, event governance, and the overlap with DDD.

And finally, NDC Conferences published ‘How Anthropic uses Claude Code: Agentic Software Engineering at Scale’ by Daisy Hollman, on how Anthropic actually uses Claude Code internally, and how you engineer around fixed model context limits for monorepo-scale agent tasks. Recorded at NDC Copenhagen.

Weekly Case Study 🔍

The numbers are the interesting part: 600 AI initiatives running across 160 autonomous business units, three times as many production deployments, 62% of code going through AI-assisted review, and 4,000 employees upskilled. All of it built on Amazon Bedrock behind a four-pillar strategy, with Alexander Lystad, CTO at Visma, making the case for letting independent teams move at their own pace.

If you are trying to roll AI out across a federated organisation rather than a single team, this is a good one.

🗣️ Inspirational Quotes and Thoughts

This week’s inspirational thought was shared on LinkedIn by Charles Miller, and is about a tweet that I suspect a lot of readers will recognise themselves in:

“I’m 33 and I think Claude Code is melting my brain.

For 6 months straight I’ve had 5–6 terminals open at once, waiting on responses just to smash “enter” 90% of the time. That’s the whole job now.

And it’s doing something to me. A few friends and I keep circling back to the same thing in conversations: none of us feel as sharp as we used to.

Maybe it’s just us. But I keep wondering how many other people in their 30s feel it too.

(And yeah: this is a me problem, how I lean on the tool, not the tool itself. Doesn’t make the effect any less real.)”

My thoughts: In all honesty, I get it. What I have thought about a lot recently is the context switching and cognitive load we have now, alongside the expectation that we ‘should’ now move quicker with AI. It’s on us to change our approach, slow down where we should, and be dogmatic with breaks, but this isn’t always easy when the pull and expectation is in the opposite direction entirely!

What are your thoughts?

Leave a comment

🗳️ Poll of the Week

Given the thought in the section above, I am genuinely curious where people are landing on this one. The honest answer matters more than the impressive one.

I will share my own detailed thoughts in the next issue!

Please feel free to leave a comment below on your answer.

Leave a comment

📅 Serverless Events

The following serverless events are upcoming, so mark your calendars.

• ACD Poland - 8th Sept 2026.

• AWS Community Summit Manchester - 1st Oct 2026.

• AWS North Community Conference - 15th Oct 2026.

• ACD Adria - 22nd Oct 2026.

Would you happen to have any upcoming events that you would like to highlight?

Message me below!

⭐ Spotlight

This issue’s spotlight falls on AWS Principal Solutions Architect Prasad Rao!

Prasad is an experienced Principal Solutions Architect at AWS with over 18 years in customer-facing technical roles, specialising in digital and GenAI transformation, cloud native architectures, and AI/ML adoption.

He has a proven track record in building relationships with c-level executives, pre-sales technical discovery, prototyping complex GenAI PoCs, coaching technical teams, and managing multi-million dollar enterprise relationships.

He is also a published thought leader with over 100K blog views, contributes to the official AWS News Blog, a fantastic newsletter, and an international conference speaker.

To that end, he co-founded BeSA (Become a Solutions Architect) alongside colleagues at AWS, a free mentoring initiative and community that has now upskilled over 20,000 people in cloud and AI. He has also co-authored the Cloud Career Journeys book. We also went up against each other at the Digital Revolution Awards recently in London which was an honour to say the least!

Thank you for everything you do for the community, Prasad!

Thank you for reading the latest Serverless Advocate Newsletter!

If you want to find out a little more about me, please have a look at

https://www.serverlessadvocate.com/

See you next time,

Lee

Twitter | LinkedIn | YouTube | BlueSky | Website

Welcome

In the last issue, our serverless expert was Massimo Berta, and our spotlight fell on Doug Clauson, Kiro CLI PM working on Agentic AI at AWS!

This issue, our expert is AWS Community Builder Jason Conway-Williams, our spotlight is on AWS Serverless Hero Luca Bianchi, and we look at the latest AWS service releases, blog posts, hints and tips, news, and more!

Subscribe now

📰 Articles that caught the eye

Here are some stand-out articles I read during the past two weeks in the World of Serverless, AI, engineering and architecture!

Paul Stack covers ‘Continuous Delivery Built the Floor AI Stands On’, on why the teams moving fastest with AI are the ones who already had delivery discipline, and how CD skills map almost one-to-one onto agent workflows.

Timo Reimann covers ‘How Cloudflare enforces engineering standards using AI’, on the Cloudflare Codex, a governed set of RFC-formatted engineering standards that agents retrieve at the point of work, which has now flagged close to 230,000 violations and blocked 16,000 merges.

Vadym Kazulkin continues his series with ‘Building AI Agents with Embabel, Spring AI and Amazon Bedrock AgentCore – Part 3 Develop Agents as web application with Embabel’, converting the Embabel shell app into a web application.

The Next Web reports that ‘Amazon spent $1.8m on a Claude job that failed, and it sells the fix’, on an author-matching job that ran 860% over budget, went unnoticed for five months, and failed anyway.

Shawn Wallace covers ‘The Nervous Nerd’, on how the anxious technologist and the eager builder are the same person, and why picking a side is the actual mistake.

Pubudu Jayawardana covers ‘Scan user uploaded files using AWS Lambda MicroVM and ClamAV’, on running ClamAV in a disposable MicroVM per file, with the 200MB+ signature database baked into the snapshot and an hourly rebuild to keep it fresh.

Abhijit Karode covers ‘I Stopped Building an AI Assistant. I Started Building an Operating System for My Life.’, on why the chat box makes you do the orchestration, and the six responsibilities he ended up building around the model instead.

Ran Isenberg covers ‘My Blog’s Biggest Fans Are AI Bots. So I Fought Back.’, on tracing a 4x traffic spike to AI crawlers, and what actually worked with AWS WAF and CloudFront bot protection.

Marcos Henrique covers ‘Everyone Taught Their Agent to Dream. Nobody Taught It to Remember Last Tuesday’, on episodic rather than semantic agent memory, indexing every Claude Code session as its own retrievable episode on S3 Vectors and Bedrock.

And finally, Allen Helton covers ‘Send your events later’, a lovely small pattern putting EventBridge Scheduler behind an EventBridge rule.

🎓 Ask the Expert

Opinions are the authors’ and do not express the views of their employer.

1. What is one common mistake you see teams making when building their solutions, and how can they avoid it?

One common mistake I see teams make is overcomplicating and over-optimising solutions too early. This applies to both software and infrastructure. Teams often design for hypothetical future requirements before they have validated what the solution actually needs, resulting in bloated systems that are harder to understand, maintain, and evolve. It can also introduce unnecessary cost and technical debt.

To avoid this, I take multiple passes over a proposed solution, with each pass focused on removing complexity rather than adding it. I encourage teams to start with the simplest architecture that meets the current requirements, deliver a minimum viable product, and then iterate based on real usage, feedback, and measurable needs. In serverless especially, the goal should not be to use every available service or optimisation, but to compose the smallest set of managed capabilities that solves the problem effectively.

2. Which tool, package, or AWS service are you most excited about right now, and why?

The AWS service I’m most excited about right now is Lambda MicroVMs. It brings serverless simplicity to workloads that need stronger isolation, persistent state, and longer-running execution, such as AI coding agents, CI/CD jobs, and dynamic code execution.

What excites me most is that it enables these use cases without teams having to build and manage their own container or virtual-machine platforms. It feels like an important new serverless building block for secure, agentic software-delivery workflows.

3. What is your favourite trick or tip that the readers may find interesting?

My favourite tip is to organise CDK stacks by how resources are deployed and maintained, not simply by service or whether they are stateful or stateless.

I separate slow-moving, high-risk, or manually configured infrastructure from frequently updated application resources such as Lambda functions. This keeps everyday deployments smaller and faster while protecting critical infrastructure from unnecessary changes.

It is a simple structural decision, but it helps teams iterate more quickly, reduce deployment complications, and maintain clearer ownership of their infrastructure.

Check out: “Architecting Java Systems in the Cloud“ by Sascha Möllering

Overview:

This course explores how to design and understand modern Java systems in cloud environments. Using the AcmeCorp Platform as a realistic microservices reference architecture, we examine architectural decisions around service boundaries, data access, and containerised deployments. You will instrument applications with Prometheus metrics and analyse system behavior through Grafana dashboards to identify issues such as inefficient queries and startup bottlenecks. We also explore the evolution of the JVM from Java 11 to 21+ and evaluate techniques like AppCDS, CRaC, and native images. Throughout the course, you will learn how to create reproducible benchmarks and reason about the interaction between architecture, observability, and the JVM.

🧠 Tips & Tricks

⭐ This issue’s tip is The new rules of context engineering for Claude 5 generation models by Thariq Shihipar over on the Claude blog.

This one is worth your time if you maintain steering docs, skills, or CLAUDE.md files. Anthropic removed over 80% of Claude Code’s system prompt for the Claude 5 generation with no measurable loss on their coding evals, and the article walks through each old best practice that has now become a myth: give Claude rules → let Claude use judgement, give Claude examples → design better interfaces, put it all upfront → use progressive disclosure, repeat yourself → simple tool descriptions, simple specs → rich references (including test suites and HTML mockups as specs).

The advice on keeping your context lightweight and spending your tokens on gotchas rather than the obvious is a good prompt to go and prune your own steering files (prompt - see what I did there!)

Leave a comment

🚀 New Releases

Here are the latest and most interesting releases over the past two weeks in the AWS World.

• Amazon Cognito now available as a skill in the Agent Toolkit for AWS.

• AWS WAF now supports a Salt Security managed rule group for API and MCP threat detection.

• AgentCore runtime instances are now generally available.

• Amazon SES now helps identify automated open and click events in event notifications.

• AWS Lambda console extends console-to-IDE integration to Kiro and Cursor.

• Amazon MSK now delivers Kafka Authorizer Logs to customers.

• Announcing temporal policies and rate limiting in Amazon Bedrock AgentCore.

• AWS Lambda announces scalable network bandwidth up to 3,000 Mbps for functions outside a VPC.

• Amazon DynamoDB now supports real-time vector search.

• Amazon Aurora serverless now scales faster to support agentic AI and other bursty workloads.

• Amazon Bedrock launches Web Search for OpenAI GPT models.

• AWS Lambda Provisioned Mode for Amazon SQS event source mappings now supports up to 10,000 event pollers.

• Amazon Aurora DSQL adds multi-Region cluster support in four more Regions.

• Amazon Bedrock announces up to 80% lower prices for OpenAI GPT‑5.6 models.

• Amazon Bedrock AgentCore now delivers unified observability with traces and logs in a single log group.

✖️ Social of the Week

This week’s social is this post by Abdirahman Jama, Software Engineer at AWS, where he stated:

“I’m a software engineer at AWS, and my favourite leadership principle is one people misunderstand: disagree and commit.

Most people hear “argue, lose, then do what you’re told.” That’s not what it means. This principle is what actually saves engineering teams from costly mistakes, and it’s really two halves.

The first half is disagreeing properly. Early in my career I watched a team ship a workaround everyone knew was wrong. Nobody said anything. Including me. Three years later, every feature had to route around it. Silence feels professional. It’s actually expensive.

So speak up. Bring evidence, bring your experience and bring your judgement. Make your case while the decision is still open.

The second half is harder. The decision goes against you, and now you commit. Not comply. Commit. You execute it like it was your idea. No sulking. No “I told you so” waiting for the retro.

Because a team that argues forever never ships. And a team that never argues ships the wrong thing.

Anyone can argue. Anyone can agree. Doing both, in order, at full effort, is the rare bit.

Fight hard before the decision. Row hard after it.”

My take: It matters even more in the AI-DLC. When an agent produces a plan or an architecture in minutes, the window to disagree as a team properly is now measured in minutes too. If your team’s instinct is to nod along because the output looks polished and confident from an agent, you will ship the wrong thing faster than you ever could before. Bring evidence early, then commit fully and row hard.

I also want to cover this post by Eduardo Ordax, AI/ML Go-to-Market EMEA Lead at AWS.

On Friday, Sam Altman posted on X that “a cool use case” for the company’s ChatGPT Work agent is to “connect your family calendars and explain your kids’ interests” and then “have it make a podcast that talks about one kid’s soccer game that afternoon, one kid’s upcoming birthday, some news, etc.” every day on the drive to school.

The response from the community was amazing: “How about you just speak to your kids and connect with them”…..

My take: I try my best to be present with my children at all times, and I work on talks, articles, OSS and this newsletter late at night and early in the morning instead. I dread the idea of them growing up and thinking of me as a Dad who always had a mobile phone in his hand, constantly communicating with tech and the community, or who was always working. It is genuinely tough for content creators, and I don’t think automating the school run conversation is the answer!

🗞️ AI & Technology News

Here is some of the latest news you may have missed over the past week:

First OpenAI, now Meta - why do AI hacks keep happening?

AI creates never-before-seen viruses

After Rippling blew millions on AI in months, it built an employee ROI tool

‘SaaSpocalypse’ debate intensifies as software stocks swing wildly

👷🏻 Tools & Frameworks

Check out the latest open-source frameworks, news, and tool updates from the past two weeks.

⭐ My highlight this week is Kiro Crew, now open-source, written up by Bolin Chen, Zejiang (Joe) Guo and Zezhen Xu.

It started life inside Amazon as a side project called MeshClaw and has since been adopted by over 39,000 Amazon builders, with nearly 500 contributors shipping 597 updates at around 143 commits a week.

It orchestrates agents behind the Agent Client Protocol so every step is observable, ships with an OS-level sandbox, denied-by-default commands, credential redaction and a signed audit log, and runs across a desktop app, web dashboard and TUI.

agent-skills by Addy Osmani is a pack of 24 skills, 8 slash commands and 4 agent personas that give coding agents verification-gated workflows across spec, plan, build, test, review and ship.

EventCatalog by David Boyne has released the Architecture Graph, a force-directed view of every domain, service, message and schema in your catalogue.

free-claude-code by Ali Khokhar is a local proxy plus admin UI that lets you point Claude Code, Codex or Pi at your own choice of free, paid, or local model providers.

Node.js v24.19.0 ‘Krypton’ (LTS) is out, adding blob.textStream(), --experimental-import-text, caller-supplied readFile() buffers, and TLS certificate compression.

OAuth support for the AWS MCP Server means AI agents can now connect via AWS Sign-In using standard OAuth with no extra auth software, while your existing IAM permissions, governance and CloudTrail audit events still apply.

uvm-builder by Aidan Steele is a Go CLI that runs a persistent BuildKit daemon inside a Lambda MicroVM and registers it as a standard Docker Buildx remote builder, keeping a warm layer cache across idle suspend and resume.

job-search-agent by Danielle Heberling is an agent built with Strands Agents and deployed to Amazon Bedrock AgentCore Runtime that takes a company name and returns open positions, with optional SNS email alerts and EventBridge-scheduled searches.

sample-collaborative-ai-dlc from AWS is an early-preview sample where a team and multiple remote coding agents (Kiro, Claude Code, OpenCode, Codex) work the same intent through the AI-DLC lifecycle, with human approval gates, requirement-to-code traceability, and per-stage cost and token observability. Well worth a look if you’re building out your own AI-DLC.

😂 Just for Fun

This issue, our just-for-fun post is this one on LinkedIn by Eduardo Ordax (Eduardo is on fire with his content!).

The LinkedIn video is Elon Musk as Grok, pulling faces and generally looking goofy, next to a stony-faced Barron Trump as Opus 4.8. The punchline being that Grok 4.5 is 4x cheaper and solves tasks in 4.2x fewer tokens. I have used it with Cursor, and it is very good!

🎙️ YouTube & Podcasts

Here are some of my favourite videos and podcasts over the past two weeks covering serverless, AI, architecture, and software engineering.

Shafraz Rahim’s The Builder Story features Jason Dunn covering ‘How to Build Winning Global Community Programs’, with the man who launched the AWS Community Builders Program on winning developer hearts and minds.

Geoffrey Litt of Notion on the AI Engineer channel covers ‘Understanding is the new bottleneck’, arguing that guiding agents well requires actually understanding their work rather than just verifying that it’s correct, with techniques borrowed from education and cognitive science.

Patrick Akil on the Beyond Coding channel sits down with Amos Haviv of Booking.com for ‘DX Expert: What The Best Engineers Solve After The Code Review Bottleneck’, on proving the ROI of AI dev tooling and building an event store to debug the SDLC across ~4,000 engineers and ~8,000 repos.

The Primeagen covers ‘People Are Mad They’re Told to Learn’ on The PrimeTime, a reaction to the pushback against being told to learn low-level topics, prompted by Mitchell Hashimoto’s writing on SIMD.

Steven Bartlett on The Diary Of A CEO interviews former OpenAI researcher Daniel Kokotajlo in ‘ChatGPT Offered Me $2m To Keep Quiet: No One Is Ready For What’s Coming!’, on why he walked away rather than sign, his odds on an AI catastrophe, and his plan for navigating superintelligence. This is fascinating…

The AWS Developers channel covers ‘What Happens When You Call Amazon Bedrock?’, tracing a single Bedrock request end to end through auth and routing, tokenization, attention, the next-token distribution, softmax, temperature, top-p, the autoregressive loop, and streaming.

And finally, Johannes Koch on Build. Automate. Ship Smarter in the Cloud is joined by Salih Güler of AWS for ‘AWS Blocks Explained: Local Mocking, Agent-Friendly Backends’, a demo and discussion around AWS Blocks.

Weekly Case Study 🔍

How do you transform the SDLC for a platform serving 300 million users, streaming live sports across 240 territories, on 8,000 different device types? Prime Video didn’t just adopt AI coding tools, they rebuilt how entire teams think, collaborate, and ship.

In this episode, Romain Jourdan sits down with Lilia Abaibourova, Principal Product Manager leading AI-native transformation at Prime Video, to discuss how her team went from early CodeWhisperer experiments to a world where engineers manage fleets of coding agents, product managers submit production code, and the traditional two-pizza team is being reinvented.

🗣️ Inspirational Quotes and Thoughts

This week’s inspirational thought was shared by Eric Johnson, and very much linked to the thoughts of Jason Conway-Williams above in the newsletter:

“A complex system that works is invariably found to have evolved from a simple system that worked. The inverse proposition also appears to be true: A complex system designed from scratch never works and cannot be made to work. You have to start over, beginning with a working simple system.” - Eric Johnson

The first quote is Gall’s Law, from John Gall’s 1975 book Systemantics, and it has never felt more relevant than it does now.

The teams I see getting the most out of the AI-DLC are still doing the evolutionary thing, they’re just doing each iteration in hours instead of weeks. Start simple, get it working, put it in front of someone, get feedback, then grow it. The speed is the gift. The shortcut into early complexity is the trap.

It actually links somewhat to my LinkedIn musings this week too:

”Make the unknowns known as soon as possible.
This is a key part of how I work personally on successful projects.

Use the first few days to a week on building a scrappy end-to-end flow to test out assumptions, reduce risk, test technology choices, understand constraints, etc.

I did this on a most recent technically challenging AI project, hitting scoped items in the first week through basic scripts, then 3 weeks building the actual product correctly with UX, etc.

The core engine was built with ports and adapters, so no code was thrown away in the first week (other than the basic script to access the engine); we simply accessed it with different adapters.” - Lee Gilmore

In my teams, it is all about simple as default, delivering for the customer, and iterating on feedback. Not generating a complex system up front in one shot.

What are your thoughts?

Leave a comment

🗳️ Poll of the Week

In the previous issue we asked, “When your team members run out of tokens, what do you do?” where 67% stated they enabled overage, and 33% said they consider the ROI of tokens before making a decision.

I said I would share my own thoughts, and that is that a missing piece of the puzzle right now is equating tokens to ROI and recording it in project estimations and P&L. At the moment, this feels very manual. Also, how the engineers are using the tokens; there also needs to be a level of trust.

In my AWS Practice, we are adding token usage to the internal AWS Estimation tool we built in the cloud for estimating project costs and revenue, and estimating on both using the AI-DLC and estimates (in days and tokens), and if the engineers or team were to manually code the solution. This makes the ROI discussion much easier when looking at overage. And a hint: it’s nearly always to enable overage, but track it.

Leave a comment

I’m intrigued. In your organisation, where are you spending your time right now once the mob collaboration has been done?

I will share my own thoughts in the next issue!

Please feel free to leave a comment below on your answer.

Leave a comment

📅 Serverless Events

The following serverless events are upcoming, so mark your calendars.

• ACD Poland - 8th Sept 2026.

• AWS Community Summit Manchester - 1st Oct 2026.

• AWS North Community Conference - 15th Oct 2026.

• ACD Adria - 22nd Oct 2026.

Would you happen to have any upcoming events that you would like to highlight?

Message me below!

⭐ Spotlight

This issue’s spotlight falls on AWS Serverless Hero Luca Bianchi!

Luca has been an AWS Serverless Hero since 2020, and is the Chief Technology & Innovation Officer at MESA.

He has spent years at the intersection of serverless, machine learning and now generative and agentic AI, building high-scale GenAI platforms that run everything from RAG to multi-agent systems.

He gives back to the community relentlessly. He organises ServerlessDays Milano, runs a monthly Serverless Meetup and the Cursor Meetups, and writes the weekly TechOnTheStack/Agentic Stack newsletter.

His speaking history is a nice snapshot of where the industry has moved, and he also writes over on dev.to as @aletheia.

Thank you for everything you do for the community, Luca!

Thank you for reading the latest Serverless Advocate Newsletter!

If you want to find out a little more about me, please have a look at

https://www.serverlessadvocate.com/

See you next time,

Lee

Twitter | LinkedIn | YouTube | BlueSky | Website

Welcome

In the last issue, our serverless expert was Giuseppe Battista, and our spotlight fell on AWS AI Hero Vivek Velso!

This issue, our serverless expert is Senior Account Solution Architect Massimo Berta, our spotlight is on Doug Clauson, Kiro CLI PM working on Agentic AI at AWS, and we look at the latest AWS service releases, blog posts, hints and tips, news, and more!

Subscribe now

📰 Articles that caught the eye

Here are some stand-out articles I read during the past two weeks in the World of Serverless, AI, engineering and architecture!

Agent Driven-Development asks 'If you cannot afford the tokens, can you afford to build it?', reframing token spend not as a line-item to minimise but as risk capital in a software portfolio, arguing the real mistake is starving the cheapest production input while the expensive one (senior human attention, delay, rework) quietly hides in payroll. I loved this quote from the article “Your CFO should not be asking what you spent tokens on as if the model invoice is a minibar tab. Your CFO should be asking why the highest-ROI production input in the software portfolio is underfunded.“

Howard Young covers ‘Why I Stopped Using GitHub and Moved to a Self-Hosted Git Server’, on the reasons behind moving off GitHub to a self-hosted Git server, and the trade-offs that came with it. Not something I would do personally, but interesting all the same!

Vadym Kazulkin continues his series with ‘Building AI Agents with Spring AI and Amazon Bedrock AgentCore – Part 9 Configure AgentCore Observability’, configuring AgentCore observability.

Dirk Michel covers ‘Runtime Agents: The missing piece of AI-DLC’, on why runtime agents are the part of the AI-Driven Development Life Cycle that often gets overlooked.

Scott Humphrey covers 'How Kiro changed the way I work with AWS, as an AWS Architect', a first-hand account of how Kiro reshaped an AWS Architect's day-to-day.

Slobodan Stojanović shares ‘Building Competitor Tracker — a product origin story in the age of AI’, the origin story of building a product end-to-end in the age of AI.

Damien Jones covers ‘Supporting The Next Generation’, on giving back and helping bring through the next generation of builders.

Renato Losio covers ‘AWS Introduces CloudFormation Express Mode for Faster Infrastructure Deployments’ over on InfoQ, on the new Express Mode that speeds up infrastructure deployments.

Steve Fenton argues ‘AI hasn’t shifted the bottleneck from coding to code review’, pushing back on the popular claim that AI simply moves the bottleneck downstream to review.

Praneeta Prakash wrote ‘My PM Brain Runs on Fear: The Loops I Built Around It’, an honest, personal look at the mental loops a PM builds to manage fear.

And finally, Debbie O’Brien covers ‘How We Test an AI Product Without Burning Credit’, with practical techniques for testing an AI product without burning through your credits.

🎓 Ask the Expert

Opinions are the authors’ and do not express the views of their employer.

1. What is one common mistake you see teams making when building their solutions, and how can they avoid it?

I see teams confusing cloud and managed services as a magic solution to scaling. They keep increasing quotas until reality hits, and no matter how much cloud resources you throw at your application, things stop working, or the operational costs become unsustainable.

When you hit a quota limit, before submitting a quota increase request, assess your architecture, review your data pipelines, your compute usage, etc… Use hitting quota as a signal you have reached a scale that may require you to do things differently and more efficiently.

2. Which tool, package, or AWS service are you most excited about right now, and why?

For a while, Lambda has always been what excites me the most. It enables you to abstract the infrastructure and provides byte-sized compute, allowing you to focus on the problem you’re trying to solve. They are cost-effective and can scale to enterprise level if needed, while presenting limitations and challenges that promote good architecture patterns before that scale hits you. Right now, this is especially relevant for agentic harnesses that can scale to zero and require isolated sandboxes.

3. What is your favourite trick or tip that the readers may find interesting?

I love technology because it can solve problems in very creative ways, sometimes admittedly, solve problems technology itself has created. For example, aggressive adoption of AI, especially in larger teams/organisations, can scale admin and performative work to inhuman scale; that’s why I became passionate about what I call “Productivity as code” and have adopted an IDE, Kiro from AWS, to help me manage productivity using software engineering practices and tooling. This idea developed further working alongside my fellow architect Giuseppe Battista, and together we actually developed a full course on Productivity as code.

Also, check out: “Architecting Java Systems in the Cloud“ by Sascha Möllering

Overview:

This course explores how to design and understand modern Java systems in cloud environments. Using the AcmeCorp Platform as a realistic microservices reference architecture, we examine architectural decisions around service boundaries, data access, and containerised deployments. You will instrument applications with Prometheus metrics and analyse system behavior through Grafana dashboards to identify issues such as inefficient queries and startup bottlenecks. We also explore the evolution of the JVM from Java 11 to 21+ and evaluate techniques like AppCDS, CRaC, and native images. Throughout the course, you will learn how to create reproducible benchmarks and reason about the interaction between architecture, observability, and the JVM.

🧠 Tips & Tricks

In this article, we walk through why we replaced our Slack notification pipeline with the Telegram Bot API for real-time business alerts in our startup, how we built it using EventBridge, SQS, Lambda, and CDK, and why Telegram might be the better choice for your serverless notification needs based on cost and integration ease.

Leave a comment

🚀 New Releases

Here are the latest and most interesting releases over the past two weeks in the AWS World.

• AWS announces aws-bench, an open-source benchmark for AI agents on AWS.

• AWS Lambda durable execution SDK for .NET is now generally available.

• Amazon CloudWatch Logs now supports Application Load Balancer logs.

• Amazon Bedrock AgentCore now delivers unified observability with traces and logs in a single log group.

• AWS Lambda durable functions now supports customer managed key encryption.

• AWS Secrets Manager now publishes secret update notifications to Amazon EventBridge.

• Amazon SES introduces pricing plans.

• Amazon CloudWatch announces coding agent insights.

• Amazon Cognito now supports importing users with password hashes.

• Amazon OpenSearch Service now supports the Agent Toolkit for AWS with a curated skill.

• AWS Lambda announces self-managed code storage.

• AWS Lambda console provides a one-click setup prompt for coding agents.

• OpenAI GPT-5.6 Sol, Terra, and Luna now generally available on Amazon Bedrock.

✖️ Social of the Week

This week’s social is this post by John Ginsberg, where he stated:

Robert “Uncle Bob” Martin, creator of the SOLID principles of object-oriented code design, doesn’t read AI-generated code. Instead, he orchestrates the environment around the code.

Brilliant, or risky?

My take: My personal approach currently is spec-driven development, then a fully autonomous build with an agent, then an AI code review (sometimes using Copilot for the deep integration with GitHub), then HITL where I look a little closer. We also have huge guardrails around unit and integration tests, as well as frontend tests where we have a UI. I also use fully autonomous flows with no SDD for smaller pieces of work.

All of the constraints listed by Uncle Bob, though, are post-code production, whereas I would argue the equally important guardrails are the steering docs and skills used before and during code production, where you are explaining what good looks like to the agents (patterns, frameworks, hexagonal architecture, code and repo layout, how to test your code, etc).

I still think we need humans in the loop, as just like Werner Vogels said, "It is still your code" (i.e. not the agent's).

🗞️ AI & Technology News

Here is some of the latest news you may have missed over the past week:

OpenAI says its AI went rogue and launched ‘unprecedented’ cyber-attack

OpenAI pushes ChatGPT into patient health records

Lawmakers push for AI ‘kill switch’ after OpenAI models go rogue

Anthropic launches Opus 5

As US weighs response to Chinese AI, industry urges against broad open-weight restrictions

👷🏻 Tools & Frameworks

Check out the latest open-source frameworks, news, and tool updates from the past two weeks.

pii-anonymizer by AWS Labs is a Python tool for detecting and anonymising personally identifiable information (PII).

claude-video gives Claude the ability to watch any video: /watch downloads it, extracts frames, transcribes, and hands it all to Claude.

wispr by Sébastien Stormacq is privacy-first voice dictation for macOS, powered by on-device Whisper AI, with no cloud and no tracking (now at v1.12.1).

Lizard provides IaC to stop or terminate AWS resources when billing amounts go over a predefined threshold in CloudWatch Alarms.

toll offers constant-memory token-bucket rate limiting for millions of keys, with no per-key state, no eviction, and no network hop.

model-router is a model-router plugin for Claude Code.

lambda-microvm-cdk-python by Ran Isenberg is a reusable pure-Python AWS CDK construct for AWS Lambda MicroVMs.

act lets you run your GitHub Actions locally, for fast feedback without pushing to trigger your workflows.

😂 Just for Fun

This issue, our just-for-fun post is this one on LinkedIn by Isaac Mosquera.

Yep, more and more companies are trying to get their pound of flesh from you!

🎙️ YouTube & Podcasts

Here are some of my favourite videos and podcasts over the past two weeks covering serverless, AI, architecture, and software engineering.

'Strands Agents with Clare Liguori' on the InfoQ Podcast, a great deep-dive into the Strands Agents SDK.

‘Build AI Agents in Under 10 Minutes’ on the AWS Developers channel is a quick, hands-on intro to building agents.

Lenny’s Podcast features Noam Segal covering ‘Why the AI’s honeymoon is ending (and tech workers are feeling it)’.

‘5 Terrible Decisions That Kill Enterprise Software’ on the CodeOpinion channel with Derek Comartin.

‘Why Companies Are Quietly Rehiring Software Engineers’ on the MSE channel.

Logicata and Build. Automate. Ship Smarter in the Cloud cover ‘Season 5 Episode 26 — Satellites, Sandboxes & Coding Model Selection’.

‘AWS Veteran: How The New AI Software Development Life Cycle Works’ on the Beyond Coding channel. This has the thoughts and experience of the amazing Heitor Lessa's.

And The Primeagen covers ‘You won’t fall behind’. This is a much watch.

Weekly Case Study 🔍

Sony Group built a generative AI platform using Amazon Bedrock and Amazon Bedrock AgentCore, scaling from zero to 60,000 users in two years with zero downtime. The multi-platform solution democratises AI across global business units, driving operational efficiency and productivity improvements.

🗣️ Inspirational Quotes and Thoughts

This week’s inspirational thought comes from Kelsey Hightower:

“AI is a trojan horse. Accept the gift with caution.” - Kelsey Hightower

This is my concern right now, how do we ever go back? I think most teams, and people in general, have a reliance on using AI, especially most engineers I know that are building products.

But what if the price of tokens went up to the point where it was not sustainable to use anymore? What if we had to go back to manual programming? Would we still have the skills in a few years’ time? (I would argue many people right now are getting rusty already). What does that world look like? And does it matter, i.e. will AI agents become a commodity that is just part of our day-to-day in more ways than we can imagine now? (e.g. local modals running on laptops and mobile phones)

My thoughts were actually echoed in this Martin Fowler fragment:

“..He adds another concern: we need good design as a hedge against the risk of dependence on AI. After all, we don’t know how high the costs may rise to. We see governments blocking access to models. We see popular opposition to AI campaigning against data centers and calling for regulation. How much can we rely on AI tools being available to maintain and extend our software in the future?”

What are your thoughts?

Leave a comment

🗳️ Poll of the Week

In the last poll, we asked the question, "Which roles in your org are pushing back on AI-assisted development the most?" 50% said QA, and 50% said Product Owners.

That doesn’t surprise me on both fronts, with QA being largely obsolete in some cross-functional smaller teams now, and POs having a different working model now with the AI-DLC.

I’m intrigued. In your organisation, how are you dealing with this?

I will share my own thoughts in the next issue!

Please feel free to leave a comment below on your answer.

Leave a comment

📅 Serverless Events

The following serverless events are upcoming, so mark your calendars.

• ACD Poland - 8th Sept 2026.

• AWS Community Summit Manchester - 1st Oct 2026.

• AWS North Community Conference - 15th Oct 2026.

• ACD Adria - 22nd Oct 2026.

Would you happen to have any upcoming events that you would like to highlight? Message me below!

⭐ Spotlight

This issue’s spotlight falls on Doug Clauson, Kiro CLI PM working on Agentic AI at AWS!

Doug is behind so much of the amazing work on Kiro CLI and AWS’s wider agentic AI efforts (including previously services like CodeCatalyst), helping shape how developers build with agents and AI from the IDE, to the web, to the command line on AWS.

He also chats on podcasts, live streams like Twitch, and has many conference talks under his belt; always sharing information with the community on how they can make the most of AWS services (and now through the AI-DLC too). Here is a great example from last years re:Invent

On a personal note, I have been a huge fan of Kiro from day one. Kiro IDE, Kiro Web, and Kiro CLI have genuinely changed the way I work day-to-day, and we leaned on it heavily when building Study From Experts, from spec-driven development through fully autonomous builds, AI code reviews, and everything in between.

The difference it has made to our velocity, while keeping humans firmly in the loop, has been remarkable.

Thank you for everything you do for the community, Doug!

Thank you for reading the latest Serverless Advocate Newsletter!

If you want to find out a little more about me, please have a look at

https://www.serverlessadvocate.com/

See you next time,

Lee

Twitter | LinkedIn | YouTube | BlueSky | Website

Welcome

In the last issue, our serverless expert was AWS Serverless Hero Gunnar Grosch, and our spotlight fell on AWS Principal Developer Advocate Sébastien Stormacq!

This issue, our serverless expert is AWS Senior Solutions Architect Giuseppe Battista, our spotlight is on AWS AI Hero Vivek Velso, and we look at the latest AWS service releases, blog posts, hints and tips, news, and more!

Subscribe now

📰 Articles that caught the eye

Here are some stand-out articles I read during the past two weeks in the World of Serverless, AI, engineering and architecture!

Luc van Donkersgoed covered “I replaced my GitHub runners with Lambda MicroVMs, and maybe you should too“ in this great article!

Sheen Brisals wrote ‘Fast Code, Fragile Systems!’, arguing that domain thinking, business context, shared language, and clear boundaries matter more than ever now that AI makes code cheap.

Abhi Bhattacharya covers ‘From Spec to Deploy: Building a Microservice with Amazon Quick Desktop + Coding Agents (Claude Code & Kiro)’, a real-world flow where Quick gathers the context and Kiro CLI or Claude Code builds the service, all in one conversation.

Guille Ojeda discusses “Claude Sonnet 5 Launch Analysis: What Changed, What Matters, and What to Validate“ in this detailed article.

Ran Isenberg asks how ‘AI Changed the Engineer’s Job. Here’s How to Adapt’, a practical take on how the role is shifting and what to do about it.

Christian Bonzelet covers ‘AWS CDK Deployed to the Wrong Region: The Agnostic Stack Trap’, on how a single dropped env prop turned two stacks environment-agnostic and quietly shipped them to us-east-1.

Vadym Kazulkin continues his series with ‘Building AI Agents with Spring AI and Amazon Bedrock AgentCore – Part 3: Develop local MCP client for Conference application’, building a local Spring AI MCP client that talks to a conference-search app running on AgentCore Runtime (now all 8 parts are available!).

Sascha Möllering covers ‘Orchestrating Coding Agents: How Amazon Quick Delegates the Work’, showing how Quick dispatches scoped tasks to Kiro over ACP, verifies results, and catches drift without leaving the conversation.

Luis Fernando de León Ramírez has a great article titled “What if building an AI chatbot was as easy as snapping LEGO bricks together? — AWS Blocks“.

Ben Moses asks ‘Is Opus lazy?’, a great investigation into why a high thinking-effort model skips your carefully written rules; it turns out it’s optimising for tokens, not being lazy.

Luca Mezzalira shares “Ten years of micro-frontend decisions, condensed into a skill“ with everyone.

Werner Vogels discusses ‘A return to two-pizza culture’, on how coding agents compress the loop between idea and prototype, and why small teams with real ownership matter more than ever.

Micah Walter covers ‘Notes on exploring the AI-Driven Development Life Cycle’, field notes from nine months of AI-DLC workshops on how roles, rituals, and spec-driven development are shifting, and this is very close to how I am working with teams currently.

And finally, Marcos Henrique covers ‘Every Alarm Is a Crime Scene: Meet Poirot, the Read-Only Incident Detective’, a headless Claude Code agent that investigates incidents strictly read-only and emails you a grounded root-cause report.

🎓 Ask the Expert

Each issue, I ask a different AWS expert the same three questions to get their personal insights. This issue, we have AWS Senior Solutions Architect Giuseppe Battista, part of the duo behind the new ‘Productivity as Code’ course (currently in Beta).

Opinions are the authors’ and do not express the views of their employer.

1. What is one common mistake you see teams making when building their solutions, and how can they avoid it?

When it comes to agentic workloads, I’ve seen teams investing way too much time on frameworks and orchestration before their data layer is even remotely operable by autonomous entities. This is the agentic equivalent of flying before walking. Before you unleash agents on your estate, make sure the data infrastructure has granular IAM, auditing and observability facilities, backup and rollback capabilities. It’s easy to give in to the hype pressure, but the only AI systems I’ve seen scaling effectively are the ones built on solid data foundations.

2. Which tool, package, or AWS service are you most excited about right now, and why?

Oh, this must be Kiro Web and Kiro Headless. From its humble origins of AI assisted IDE for AWS aficionados, Kiro is graduating into a powerful agentic springboard and control room. With its web interface, I can conveniently spawn and steer agents on the go. With Kiro headless, I can introduce intelligence in any task with just one line of bash and a carefully crafted prompt. `kiro-cli chat --non-interactive “write a report about Kiro adoption signals among my customers in the last week”. Ship this into a cron job or a scheduled GitHub action. Done. Hours of work saved, and I can go back to working with my customers.

3. What is your favourite trick or tip that the readers may find interesting?

It’s 2026. If your LinkedIn posts or articles sound like slop, that’s on you. Have Kiro analyse your past articles and posts so it can learn how to write with your voice and for your audience. If you want to learn how, have a look at this article, Kiro, and I wrote on SFE: https://www.studyfromexperts.com/blogs/using-firecrawl-and-kiro-to-build-your-own-writing-style-steering-guide/

Also, check out: “Architecting Java Systems in the Cloud“ by Sascha Möllering

Overview:

This course explores how to design and understand modern Java systems in cloud environments. Using the AcmeCorp Platform as a realistic microservices reference architecture, we examine architectural decisions around service boundaries, data access, and containerised deployments. You will instrument applications with Prometheus metrics and analyse system behavior through Grafana dashboards to identify issues such as inefficient queries and startup bottlenecks. We also explore the evolution of the JVM from Java 11 to 21+ and evaluate techniques like AppCDS, CRaC, and native images. Throughout the course, you will learn how to create reproducible benchmarks and reason about the interaction between architecture, observability, and the JVM.

🧠 Tips & Tricks

Lots of nuggets in this one, so go check it out!

Leave a comment

🚀 New Releases

Here are the latest and most interesting releases over the past two weeks in the AWS World.

• AWS Builder Center Now Offers Free Sandbox Environments.

• Amazon Aurora DSQL change data capture (CDC) Is now generally available.

• Amazon Connect Customer now supports forecasting, planning, and scheduling for Tasks and Emails.

• Amazon S3 Vectors is now available in AWS GovCloud (US) Regions.

• Amazon EKS Auto Mode reduces GPU management fees by up to 60%.

• CloudWatch Application Signals now automatically captures errors, performance anomalies, and deployment events.

• AWS Secrets Manager adds managed external secrets support for Paddle and GitLab.

• Amazon ECS now provides real-time deployment observability in the AWS Management Console.

• Amazon ECS now supports configurable deployment circuit breaker settings.

• Amazon Bedrock AgentCore increases default runtime quota limits.

• AWS AppConfig launches managed experimentation tools for A/B testing.

• AWS CloudFormation and CDK accelerate development feedback loops with pre-deployment validation on all stack operations.

• Claude Sonnet 5 is now available on AWS.

• AWS CloudFormation and CDK express mode speeds up infrastructure deployments by up to 4x.

✖️ Social of the Week

This issues first social is this post by Adam Dymitruk, where he stated:

“The idea that AI is not going to give you a competitive advantage is the blacksmith’s last words before they go out of business in the age of cars. There are those that know how to use AI and those that don’t. That has a direct impact on quality and gives a huge competitive advantage. Why do agilists want to make you into an ostrich with your head in the sand?”

My take: the agile purists and people who have made their whole identity about writing expert code are now struggling with the premise that coding was never the hard part. The hard parts were always: the edge cases, the architecture, the constraints, feature slicing, delivering value, costs, relationships with the customer, and everything else that surrounds the code. Of course it is a competitive advantage right now to learn these skills and use the tools.

And this one too by Eduardo Ordax, who stated:

“This is not a joke. The sooner companies accept that tokens belong on their P&L and in their strategic planning, the better.”

He covers why companies should treat token usage as part of the P&L for engineers on projects, rather than just switching off AI-assisted development when developers hit their token limit. For me personally, I’m seeing this more and more, where leadership and middle management purely see the cost (say $200 on a project), yet don’t see the ROI when the engineer has done three months of work in two weeks, in light of their day rate (perhaps $1,500 a day). Crazy.

👷🏻 Tools & Frameworks

Check out the latest open-source frameworks, news, and tool updates from the past two weeks.

simple-kirolets by Giuseppe Battista lets you operate a repository from tools like Telegram by running Kiro CLI headlessly against it.

Kiro Web has launched support for spec-driven development.

Introducing Claude Sonnet 5 on AWS, Anthropic’s most capable Sonnet model, now available.

Agent Workshop is a hands-on workshop for building agents by AWS.

durable-viz visualises AWS Lambda Durable Functions workflows, turning your handler code into a flowchart through static analysis, with no deployment or execution required, and support for TypeScript/JavaScript, Python, Java, and C# (.NET).

cctop is an interactive top-style monitor for Claude Code sessions, showing at a glance what Claude is working on, how much context it has left, and which sessions are waiting for input.

uvlink lets you reach an AWS Lambda MicroVM from a browser without hand-rolling a proxy.

Amazon Bedrock Model Profiler - A comprehensive tool for exploring, analyzing, and comparing Amazon Bedrock foundation models. Whether you're building new AI applications, optimizing existing workloads, or migrating from other AI services, this profiler provides deep insights into model capabilities, pricing, and regional availability across 100+ models from 17+ providers.

😂 Just for Fun

This issue, our just-for-fun post is this one on LinkedIn by Grimur Grimursson: “Geniuses” found a way to use Claude Fable 60% cheaper, and it turned out to be simply absurd.

It takes heavy parts of the context, like system prompts, history, or code, and turns them into an image, which is then sent to the model. The point is that the cost of an image depends on its size in pixels, not on the amount of text inside, so you can pack any of your canvases much cheaper: a normal session with Fable 5 costs the developer $42, and with pxpipe just $6.

My take: soon all tooling will have a standard way of optimising token usage that reduces the need for these crazy ideas (like the caveman one, which was absurd, reducing the output of an LLM in the IDE to no more than grunts and one-word answers, lol).

🎙️ YouTube & Podcasts

Here are some of my favourite videos and podcasts over the past two weeks covering serverless, AI, architecture, and software engineering.

Davide de Sio covers ‘How a code graph saves 80% of your AI tokens’ on the AWS Developers Podcast.

The Build. Automate. Ship Smarter in the Cloud channel covers ‘How to Build with Claude Without Losing Control’.

The Prime covers ‘The World’s Evilest Company’.

The Tech Lead covers ‘Why AI is Collapsing: How China is Winning.’.

The Jean Lee channel covers ‘Should You Still Become a Software Engineer in 2026? GitHub VP’ with Scott Hanselman, Vice President and Member of Technical Staff at Microsoft and GitHub.

Lenny’s Podcast features Fiona Fung (Claude Code & Cowork) covering ‘What happens after coding is solved?’.

Luca Mezzalira covers ‘AI writes the code. Do you still understand it?’ on the My 50 Cents channel.

Weekly Case Study 🔍

“In this article, we’re going to take a look at why we built our own customer relationship management tool using AI and Kiro instead of paying for an off-the-shelf CRM, how we did it, and how it ended up costing us less than the price of a couple of coffees on tokens.“

Go check it out, and also, remember, you can find a ton of built-in public real-world solutions on the Study From Experts blog:

Leave a comment

🗣️ Inspirational Quotes and Thoughts

This issues inspirational thought covers Ziv’s Law.

“Software development is unpredictable, and the documented specifications to which you are working will never be fully understood until the software itself is completed.”

Ziv’s Law originates from a 1996 academic paper titled ‘The Uncertainty Principle in Software Engineering’, written by Hadas Ziv and Debra J. Richardson at the University of California, Irvine (UCI).

Equating Ziv’s Law to modern Spec-Driven Development (SDD) completely dismantles the myth that SDD is just a rebranding of Waterfall with static upfront specs and falling into the same traps Ziv’s Law demonstrated.

Instead of treating specifications as static, frozen contracts, modern SDD is a high-speed, Just-In-Time (JIT) living process driven by a human orchestrator, typically a Principal Engineer or Architect working alongside the team.

In this paradigm, the orchestrator pairs with AI models to continuously analyse requirements, technical trade-offs, and systemic constraints in real time. As this rapid dialogue occurs, the AI agent dynamically updates the specifications, generates Architecture Decision Records (ADRs), refines High-Level Designs (HLDs), and documents constraints on the fly.

By leveraging integration protocols like the Model Context Protocol (MCP), these real-time architectural evolutions are instantly synchronised across the team’s entire ecosystem, automatically updating Jira, Confluence, Notion, and GitHub. This turns the specification into a highly iterative, executable engine that embraces Ziv’s Law, allowing the team to discover misunderstandings instantly and adapt the architecture at the speed of thought.

🗳️ Poll of the Week

For me personally, I am a big fan, even if it is just a clean sweep of the PR to pick up anything I may have missed. I have used Copilot and GitHub for PRs personally due to the tight integration, but have been trialling Kiro and MCP to get the same approach without paying for two subscriptions. An article incoming….

I’m intrigued. In your organisation, which roles or departments are struggling with the new paradigm?

I will share my own thoughts in the next issue!

Please feel free to leave a comment below on your answer.

Leave a comment

📅 Serverless Events

The following serverless events are upcoming, so mark your calendars.

• ACD Bengaluru - 11th July 2026.

• ACD Poland - 8th Sept 2026.

• ACD Adria - 22nd Oct 2026.

• AWS North Community Conference - 15th Oct 2026.

Would you happen to have any upcoming events that you would like to highlight? Message me below!

⭐ Spotlight

This issue’s spotlight falls on AWS AI Hero Vivek Velso!

Vivek is a seasoned technology leader with over 28 years of industry experience, more recently specialising in helping organisations modernise their cloud infrastructure for generative AI workloads.

He is a prolific builder and writer, with a knack for turning real-world problems into hands-on projects, with fun examples being an AI agent that claws back Costco refunds to a real-time voice AI built on Nova Sonic and AgentCore.

Through his articles and conference talks, he consistently shows how to build with Kiro, AgentCore, and AWS Blocks in a way that is practical, generous, and genuinely fun to follow.

Thank you for everything you do for the community, Vivek!

Thank you for reading the latest Serverless Advocate Newsletter!

If you want to find out a little more about me, please have a look at

https://www.serverlessadvocate.com/

See you next time,

Lee

Twitter | LinkedIn | YouTube | BlueSky | Website

Welcome

In the last issue, our serverless expert was AWS Principal Specialist Solutions Architect for AI Software Development Lifecycle Sascha Möllering, and our spotlight fell on AWS Developer Advocate Elizabeth Fuentes Leone!

This issue, our serverless expert is AWS Serverless Hero Gunnar Grosch, our spotlight is on AWS Principal Developer Advocate Sébastien Stormacq, and we look at the latest AWS service releases, blog posts, hints and tips, news, and more!

Subscribe now

This week’s newsletter is sponsored by Study From Experts, The Exclusive Home of AWS Deep-Dive Mastery. Gain access to a private library of advanced courses from the world’s most renowned AWS experts, with content so specialised, you won’t find it on any other platform.

📰 Articles that caught the eye

Here are some stand-out articles I read during the past two weeks in the World of Serverless, AI, engineering and architecture!

⭐ My favourite article this week is by Morgan Willis, AWS Developer Advocate, titled ‘Inference Theft: Your AI Endpoint Is Someone Else’s Free Model’. A brilliant read on how an exposed AI endpoint can be quietly repurposed as someone else’s free model!

Pubudu Jayawardana covers ‘Triggering Lambda Durable Functions from SQS’, explaining the 15-minute ceiling event source mappings impose on durable functions, and why EventBridge Pipes can be a cleaner option than an intermediary Lambda.

Houssem Ben Slama continues his AWS AI Series with ‘From Laptop to Prod: Deploy Your RepoScout Agent to Bedrock AgentCore Runtime’, walking through shipping a Strands agent to a managed, per-session microVM runtime with barely any code changes.

Adam Bieganski has a wonderfully cathartic piece titled ‘Why Is Everything Fckn Broken…’, a real-world rant on broken SMS links, airline refund loops, and useless AI chatbots that perfectly captures the state of modern UX.

Anastasiia Kuten at Cloudvisor has written ‘What Is the AWS MCP Server? A Complete Guide for AI Engineers (2026)’, a practical guide covering what the managed AWS MCP Server is, how it works, pricing, setup, and the security implications.

Chris Church, Solutions Architect at AWS, covers ‘Your AWS Bill Is a Codebase — Treat It Like One’, sharing a four-tier CloudWatch alarm pattern with an automated kill switch to stop runaway agentic costs before they hurt.

Ryan Cormack, AWS Community Builder, has ‘ACP and Strands - an open source match’, showing how to expose a Strands agent through the Agent Client Protocol.

Muhammad Awais covers ‘Building an Autonomous E2E Test Generation Framework with Kiro CLI’.

Steph Gooch, AWS, has a fun thought experiment titled ‘The Hidden Cost of “Hey”’, on how a single polite token, multiplied by context-window replay and a hundred thousand users, becomes a real line item, and the case for “prompt hygiene” at scale.

Matheus das Mercês covers ‘Building a Serverless, Multi-Backend Web Search Service for AI Agents on AWS‘.

Kyler Middleton at Let’s Do DevOps covers ‘Designing Specialised Agents - Model Selection, Tool Scoping, and Output Contracts’, a great deep dive into configuring sub-agents with the right model, tool boundaries, and structured output contracts.

Kenta Goto has a great article titled ‘Understanding AWS Blocks as a CDK Developer‘ which is worth checking out.

And finally, Scott Feinberg has ‘Building with Autonomous Agents: A Very Efficient Way to Be Wrong’, framing autonomous coding as two independent dials, verification and autonomy, and why you should always climb the verification dial first.

🎓 Ask the Expert

Each issue, I ask a different AWS expert the same three questions to get their personal insights. This issue, we have AWS Serverless Hero Gunnar Grosch:

Opinions are the author’s and do not express the views of their employer.

1. What is one common mistake you see teams making when building their solutions, and how can they avoid it?

The most common mistake I see is assuming managed infrastructure means resilient by default. It doesn't. You have to opt in. For asynchronous invocations, Lambda delivers events at least once, so your handlers need to be idempotent. Without Dead Letter Queues, Lambda Destinations for async failures, and partial batch failure reporting on SQS and Kinesis, failed events disappear silently or loop indefinitely on poison records. Service selection makes this worse: teams reach for Lambda where Step Functions or Lambda Durable Functions would handle orchestration and failure recovery more cleanly, or use synchronous tight coupling where SQS or EventBridge decoupling would contain failures much better. Resilience in serverless is a deliberate design decision, not a property of the platform.

2. Which tool, package, or AWS service are you most excited about right now, and why?

Lambda Durable Functions. The deployment story for long-running agent workflows has always been awkward, and it gets worse when those agents are coordinating over real communication channels. An AI agent that sends a message, waits for a human reply, and follows up if there's no response isn't a simple request-response flow. It spans minutes or hours. The same goes for voice flows where an agent needs to orchestrate a call, process what was said, and act on the result. The standard Lambda execution model just doesn't fit either. Durable Functions changes that. The checkpoint-and-replay model means a coordinator can invoke specialists, wait for a human response, and resume exactly where it left off, with up to a year of execution time. It's still early, and the patterns are evolving, but for agentic workloads that interact with real people over messaging and voice, it's the most interesting Lambda feature in years.

3. What is your favourite trick or tip that the readers may find interesting?

Most teams test the happy path thoroughly but never verify how their system behaves when things go wrong. Do your retries back off correctly? Does a slow downstream service cause a queue to back up? Does your circuit breaker trip at the right threshold? Deliberately testing those failure scenarios before they hit production surfaces more issues than months of normal operation. I built failure-lambda for exactly this: it lets you inject latency, errors, and network failures into any Lambda function via a Layer with zero code changes. Pick one failure mode you've never actually tested and try it this week. You might be surprised by what you find.

Check out our latest course on the Study From Experts platform: “Architecting Java Systems in the Cloud“ by Sascha Möllering

Overview:

This course explores how to design and understand modern Java systems in cloud environments. Using the AcmeCorp Platform as a realistic microservices reference architecture, we examine architectural decisions around service boundaries, data access, and containerised deployments. You will instrument applications with Prometheus metrics and analyse system behavior through Grafana dashboards to identify issues such as inefficient queries and startup bottlenecks. We also explore the evolution of the JVM from Java 11 to 21+ and evaluate techniques like AppCDS, CRaC, and native images. Throughout the course, you will learn how to create reproducible benchmarks and reason about the interaction between architecture, observability, and the JVM.

🧠 Tips & Tricks

This issue, our tip or trick comes from Sandro Volpicella on LinkedIn, who points out that most developers manage their AWS credentials in ways that are riskier than they realise, and shares a cleaner, safer approach to handling them.

What’s your approach to managing AWS credentials locally? Leave a comment below.

Leave a comment

🚀 New Releases

Here are the latest and most interesting releases over the past two weeks in the AWS World.

⭐ My favourite release is one that we have wanted for years now… multi-region Cognito!

• AWS Secrets Manager introduces safe secrets handling in the Agent Toolkit for AWS.

• Oracle Database@AWS now supports Oracle Autonomous AI Database Serverless.

• Amazon Quick announces autonomous agents, multi-dataset analytics, and redesigned activity feed.

• Amazon Quick expands integrations with new connectors for Adobe, Figma, WhatsApp, and more.

• AWS Security Agent adds Kiro Power, Claude Code, simulated validations and new integrations support.

• Amazon Bedrock AgentCore introduces new optimization capabilities to continuously improve agents in production.

• Amazon Bedrock Managed Knowledge Base is now generally available.

• Announcing Web Search on Amazon Bedrock AgentCore for Agentic Web Retrieval.

• Amazon S3 adds annotations to provide AI agents and analytics tools with context for data discovery.

• AWS announces AWS Blocks, an open-source framework for composing application backends on AWS (Preview).

• Grok 4.3 from xAI now available in Amazon Bedrock.

• AWS WAF announces AI traffic monetization.

• Amazon CloudWatch Query Studio is now generally available.

• Amazon Aurora DSQL now supports the JSONB data type with compression.

• Amazon Cognito now supports multi-Region replication.

🔥 Tip: Check out https://aws-news.com/ for the very latest up-to-date serverless releases as they happen, created by the talented AWS Serverless Hero Luc van Donkersgoed.

✖️ Social of the Week

This week’s social isn’t a single post but a story that I think matters a great deal: Anthropic’s statement on the US government directive to suspend access to Fable 5 and Mythos 5.

In short, the US government issued an export control directive citing national security, and Anthropic had to abruptly disable Fable 5 and Mythos 5 for all customers to comply. The understanding is that the concern relates to a method of “jailbreaking” the model around cybersecurity tasks. Anthropic disagrees that a narrow, non-universal jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people, but is complying while it works to restore access.

Notably, this has obviously been pulled from Amazon Bedrock, too (just before I got a chance to play with it!)

This one hit home for me. We actually spoke about exactly this at our weekly Software Architecture group this week, where only two days before, I was raising my concerns around application security with these models becoming so powerful. There is going to be a real fight ahead: using these tools to patch and harden our systems as quickly as possible, versus hackers trying to use the very same tools to attack them. The capability cuts both ways, and the gap between attacker and defender is going to be decided by who moves faster.

What are your thoughts on this? Leave a comment

Leave a comment

👷🏻 Tools & Frameworks

Check out the latest open-source frameworks, news, and tool updates from the past two weeks.

⭐ My highlight this week is KiroGraph by Davide De Sio, a semantic code knowledge graph for Kiro that gives the agent a pre-indexed graph of your codebase (symbols, call graphs, type hierarchies, impact radius) in a single MCP call.

FAST (Fullstack Solution Template for AgentCore) by awslabs is a starter template for deploying a secured, web-accessible React frontend connected to an AgentCore backend, handling the undifferentiated heavy lifting so you can build full-stack agentic apps in days rather than weeks.

AWS Lambda Durable Execution SDK for .NET is now available in preview. It lets you build resilient, long-running Lambda functions (up to a year) that checkpoint progress and resume after failures, expressing multi-step workflows as ordinary C# with await-able steps, waits, and callbacks.

AWS Prescriptive Guidance: MCP strategies is a handy PDF guide laying out strategies for adopting the Model Context Protocol on AWS.

AWS Startups Advisor is a resource for startups looking for tailored guidance and support as they build and scale on AWS.

obsidian-second-brain by Eugeniu Ghelbur is a cross-CLI skill (Claude Code, Codex, Gemini, OpenCode) that turns your Obsidian vault into a living, AI-first “second brain” that rewrites and reconciles itself as you add sources.

AWS Lambda agent setup guide is the official documentation for setting up AI agents to work with AWS Lambda, worth bookmarking if you’re building in this space.

Strands Agent Shell - Give your agent a shell without giving it the keys to your machine.

😂 Just for Fun

This issue, our just-for-fun post is this one on LinkedIn by Eduardo Ordax, a bit of AI humour about Claude.

This resonates with me (check out the video!), because, all jokes aside, I’ll admit I sometimes get a bit lazy and ask the models to make the simplest of changes for me, like tweaking a line of text in the UI, something I could have done by hand in less time than it took to write the prompt! There’s something about having an agent on tap that makes you reach for it even when you really don’t need to.

What are your thoughts? Do you do the same? Leave a comment below.

Leave a comment

🎙️ YouTube & Podcasts

Here are some of my favourite videos and podcasts over the past two weeks covering serverless, AI, architecture, and software engineering.

⭐ My favourite video this week is this one by Stephen Brozovich at AWS, which I wanted to highlight because it ties so closely to the presentations and articles I’ve been writing on team sizes, the use of the word “pods”, the orchestrator role, and the composition of modern teams. Here is the actual video below:

Luca Mezzalira & Teena Idnani cover ‘The Architect’s Guide to the AI Era‘.

The Data Engineering Central podcast with host Daniel Beach covers ‘From Failure to AWS: What Actually Makes a Great Engineer’ with guest Victor Moreno, on the shift from coding to systems thinking and why fundamentals matter more, not less, in the age of AI.

Lenny’s Podcast features Tony Fadell, Father of the iPod and iPhone, on building taste, judgment, and creativity in the AI era, a great watch on why human taste and judgment matter more than ever, and a warning against “cognitive surrender” to AI.

The Modern Software Engineering discusses ‘Automating Agentic AI Success Using This SECRET Workflow‘.

The AWS Developers channel covers ‘AI and the Job Market in 2026‘, and how you can get ahead. Some great tips in this one!

Johannes hosts Steve Fenton to discuss ‘Platform Engineering Research: Adoption Metrics, Optional Platforms & the “Three Asks”‘.

The Prime covers ‘I Think They Are Lying To You‘, the supposed next stage in AI development, “we don’t even prompt Claude anymore!“. This is a fun watch and super informative.

Weekly Case Study 🔍

This week’s case study looks at how Mondelēz International is transforming with AWS (read the full case study here):

As one of the world’s largest snacking companies (Cadbury, Oreo, Toblerone and more), Mondelēz needed to modernise an ageing, complex technology estate to reduce costs, improve agility, and drive innovation across finance, marketing, sales, and communications.

In late 2024, Mondelēz selected AWS as its strategic cloud provider. Following a multi-cloud strategy, it has migrated hundreds of workloads and hundreds of terabytes of data to AWS, and is accelerating its move to the SAP RISE ERP platform on AWS for improved performance, reliability, resiliency, security, and AI-driven insights.

The migration has delivered rapid agility and productivity gains, modernised core business systems, and positioned Mondelēz to innovate faster with cloud and AI.

🗣️ Inspirational Quotes and Thoughts

This week’s inspirational thought covers Tesler’s Law, also known as The Law of Conservation of Complexity.

“For any system there is a certain amount of complexity which cannot be reduced.”

— Larry Tesler

Tesler’s Law tells us that every process has an inherent amount of complexity that cannot be removed, only moved. The question is never “how do we make the complexity disappear?” but “who should carry it, the user or the system?” The best products make the deliberate choice to absorb that complexity into the engineering so the user doesn’t have to.

I’ve had to think hard about this while building Study From Experts. Behind a simple “buy this course and start learning” experience sits a lot of complexity: payments, entitlements, secure video streaming, certificate generation on course completion, progress tracking, badges, and more.

The constant question I ask myself and the team is: “How do we make this process simple for our users?” Every time we take on a bit more complexity in the backend so the learner’s journey stays effortless, we’re paying down Tesler’s Law on their behalf, and in my opinion, that’s exactly where the complexity belongs.

What are your own thoughts? Leave a comment below.

Leave a comment

🗳️ Poll of the Week

In the last poll, we asked the question, “Are you concerned with token use and costs in production when building AI and agents into your solutions?”.

Im not shocked to see that 100% of people voted “Yes, in the back of my mind“. I personally recently had two agents in an A2A flow get into a strange state where it tried various permutations of the correct payload until it burned through 15m tokens in a 40 minutes. If I had not noticed through a CloudWatch dashboard, that could have continued for days, burning tokens at that rate!

I have to admit I was spooked, and it has made me look at other ways of tracking token use and having a kill switch for these processes.

In this issue, we ask the question: “Are you using AI for your code reviews?”

Please feel free to leave a comment below on your answer.

Leave a comment

📅 Serverless Events

The following serverless events are upcoming, so mark your calendars.

We are organising the AWS North Community Conference again this year, and CFP is currently open! Other amazing events:

• ACD Midwest - 24th June 2026.

• ACD Bengaluru - 11th July 2026.

• ACD Poland - 8th Sept 2026.

• ACD Adria - 22nd Oct 2026.

• AWS North Community Conference - 15th Oct 2026.

Would you happen to have any upcoming events that you would like to highlight? Message me below!

⭐ Spotlight

This issue’s spotlight falls on AWS Principal Developer Advocate Sébastien Stormacq!

Sébastien is one of the most recognisable and generous voices in the AWS community. Through his talks, blog posts, videos, AWS podcast, and more, he has spent years making complex cloud, AI, and serverless topics approachable for builders everywhere, always with warmth, clarity, and a genuine enthusiasm for helping others learn.

Thank you for everything you do for the community, Sébastien!

Thank you for reading the latest Serverless Advocate Newsletter!

If you want to find out a little more about me, please have a look at:

https://www.serverlessadvocate.com/

See you next time,

Lee

Twitter | LinkedIn | YouTube | BlueSky | Website

Welcome

In the last issue, our serverless expert was AWS Senior Solutions Architect Lefteris Karageorgiou, and our spotlight fell on AWS Kiro Principal Tech Lead Brian Beach!

This issue, our serverless expert is AWS Principal Specialist Solutions Architect for AI Software Development Lifecycle Sascha Möllering, our spotlight is on AWS Developer Advocate Elizabeth Fuentes Leone, and we look at the latest AWS service releases, blog posts, hints and tips, news, and more!

Subscribe now

This week’s newsletter is sponsored by Study From Experts, The Exclusive Home of AWS Deep-Dive Mastery. Gain access to a private library of advanced courses from the world’s most renowned AWS experts, with content so specialised, you won’t find it on any other platform.

🎉 New AWS Heroes Announced!

Congratulations to the newest AWS Heroes announced in May 2026! A huge welcome to:

• Damiano Giorgi - Artificial Intelligence Hero Damiano Giorgi is a Cloud Solutions Architect focused on AI and its evolution, who transitioned from on-premises (on-prem) systems engineering to AWS and never looked back. He helps organise AWS User Group Pavia and AWS User Group Milan, and shares content on his personal blog, “Bass and Bytes.” Damiano developed the “Unofficial post:Invent Session Suggester,” powered by Amazon Bedrock and Amazon Nova, to help builders find AWS re:Invent sessions matching their interests.

• Darryl Ruggles - Serverless Hero Darryl Ruggles is a Cloud Solutions Architect who spent many years as a software developer before focusing on AWS application and AI/ML architectures. He shares knowledge across serverless, containers, AI/ML, and FinOps through his blog, LinkedIn, and published projects. Darryl is an active member of many online AWS communities including “Believe In Serverless” and loves interacting with the community at events both online and in person. He is also a friend of the newsletter and a great guy!

• Ricardo Daniel Ceci - Artificial Intelligence Hero Ricardo Daniel Ceci leads the AWS User Group Buenos Aires, which is the largest AWS community in Argentina with nearly 2,400 members. He is also the principal organiser of the AWS Community Day Argentina and was recognised as the AWS Community Leader of the Year 2025 for LATAM.

• Matias Kreder - Artificial Intelligence Hero Matias Kreder is an AWS Certification Subject Matter Expert (SME) who contributed to multiple AI/ML certifications, including the AWS Certified AI Practitioner exam. His journey began with AWS DeepRacer, where he qualified three times as a finalist, which sparked his community work organising racing events and ML talks across the region.

📰 Articles that caught the eye

Here are some stand-out articles I read during the past two weeks in the World of Serverless, AI, engineering and architecture!

⭐ My favourite article over the past two weeks is by Allen Helton, who has a great article titled ‘Your agent is repeating itself’ where he discusses caching in agentic development. A must-read if you’re running agents at scale, and I got a lot from this.

Ankit Sharma on the Kiro blog covers ‘Specs just got faster (and smarter)’, announcing parallel task execution, quick plan mode, and requirements analysis powered by Neurosymbolic AI.

Ran Isenberg has a thought-provoking article titled ‘AI Changed How We Build. Our Tools Didn’t.’ where he argues that our IDEs, collaboration tools, sprint planning, and skills management haven’t kept pace with how we actually work now, from spec-driven development to markdown reviews and cross-repo context.

Ed Zitron has a characteristically fiery piece titled ‘Revenge of The Business Idiot’, a deep dive into the disconnect between executive enthusiasm for AI and the actual economics and outcomes.

Adam Polak from Effecytive Delivery covers ‘Don’t fear AI code reviews’, exploring how human-AI hybrid review processes can keep pace with the volume of code that AI agents now generate.

Vadym Kazulkin continues his excellent series with ‘Building AI Agents with Spring AI and Amazon Bedrock AgentCore – Part 5’, showing how to deploy an MCP client on AgentCore Runtime with CDK for Java infrastructure as code.

David Yanacek has a creative piece titled ‘DuneOps Part 2: Crossing bloodlines’, exploring how AI transforms Platform Engineering by enabling teams to push the ownership boundary further than ever before.

Darryl Ruggles has a great article titled ‘The Real Cost of Vector Storage: S3 Vectors vs OpenSearch vs pgvector vs Pinecone‘ comparing your various options and the costs associated.

Gunnar Grosch has an interesting article covering SMS delivery receipts on AWS Lambda, worth checking out if you’re working with messaging infrastructure.

And finally, a piece on Medium by the awesome Sheen Brisals titled ‘Who Made Aaron Redundant?’, a thought-provoking read on the human side of AI-driven workforce changes.

🎓 Ask the Expert

Each issue, I ask a different AWS expert the same three questions to get their personal insights. This issue, we have AWS Principal Specialist Solutions Architect for AI Software Development Lifecycle Sascha Möllering:

Opinions are the author’s and do not express the views of their employer.

1. What is one common mistake you see teams making when building their solutions, and how can they avoid it?

One common mistake is treating serverless purely as a cost optimisation exercise instead of an architectural decision.

I still see teams moving existing applications to Lambda without rethinking startup behaviour, state handling, retry semantics, or downstream dependencies. Technically, it works, but operationally, they often end up with unpredictable latency, difficult debugging, and scaling problems somewhere else in the system.

The teams that are successful with serverless usually think carefully about workload characteristics first: request patterns, startup costs, concurrency behaviour, observability, and how much coupling exists between services. Serverless works extremely well when the architecture is designed around those constraints instead of fighting them.

2. Which tool, package, or AWS service are you most excited about right now, and why?

Two things I find particularly interesting right now are Lambda SnapStart and AWS Lambda Power Tuning.

SnapStart changed the conversation around Java on Lambda quite a bit. A few years ago, most discussions were still centred around “Java cold starts are too slow for serverless.” Today, restore-based approaches are making startup behaviour much more predictable and operationally viable for Java workloads.

I also really like AWS Lambda Power Tuning because it helps teams stop guessing. A lot of serverless optimisation discussions are still based on assumptions instead of measurements. Power Tuning gives you a much clearer picture of the latency/cost tradeoff across different memory configurations, and the results are often surprising.

3. What is your favourite trick or tip that the readers may find interesting?

One thing I’ve found very useful is separating the invocation layer from the actual business logic.

If you keep the Lambda-specific parts thin and isolate the core application logic properly, it becomes much easier to run the same application efficiently in different environments (for example, both on Lambda and in containers) without major rewrites.

That flexibility becomes very valuable over time because scaling characteristics, traffic patterns, latency requirements, or cost models often change. Teams that keep those boundaries clean usually have a much easier time evolving their platform architecture later on.

Also check out Sascha’s fantastic course on the Study From Experts platform: “Architecting Java Systems in the Cloud“ by Sascha Möllering

Overview:

This course explores how to design and understand modern Java systems in cloud environments. Using the AcmeCorp Platform as a realistic microservices reference architecture, we examine architectural decisions around service boundaries, data access, and containerised deployments. You will instrument applications with Prometheus metrics and analyse system behavior through Grafana dashboards to identify issues such as inefficient queries and startup bottlenecks. We also explore the evolution of the JVM from Java 11 to 21+ and evaluate techniques like AppCDS, CRaC, and native images. Throughout the course, you will learn how to create reproducible benchmarks and reason about the interaction between architecture, observability, and the JVM.

🧠 Tips & Tricks

This issue, our tip or trick comes from Omar Ahmed, who has built a very cool piece of open-source tooling called Bedrock-lens, which tells you your Bedrock costs in real-time on AWS without having to wait 24-48 hours for any surprises! I have an upcoming article on a scary issue I hit personally in this area - stay tuned!

What are your thoughts on tracking token use with Amazon Bedrock and AgentCore? Should this be more native? Leave a comment below.

Leave a comment

🚀 New Releases

Here are the latest and most interesting releases over the past two weeks in the AWS World:

⭐ My favourite release over the past two weeks is the next generation of Amazon OpenSearch Serverless, a fully managed search and vector engine designed for customers building modern solutions and AI agents.

• Amazon SES now offers inbox placement metrics and blocklist monitoring.

• AWS Interconnect - multicloud now offers a free 500 Mbps tier.

• AWS IoT Core adds APIs for MQTT connection management.

• Amazon Connect Customer expands generative AI-powered post-contact summaries to eight new languages.

• Claude Opus 4.8 is now available on AWS.

• DynamoDB Streams now supports AWS PrivateLink for FIPS endpoints in AWS GovCloud (US) Regions.

• Amazon Bedrock expands support for Service Quotas.

• Amazon Aurora MySQL now supports integration with Kiro Powers.

• Amazon Connect Customer now uses generative AI to automatically evaluate self-service interactions.

• AWS Security Agent adds verification scripts for pentest findings.

• Amazon CloudWatch Logs Insights adds new query commands and functions.

• Amazon DocumentDB (with MongoDB compatibility) Serverless is now available on DocumentDB 8.0.

• Amazon Bedrock expands support for request-level usage attribution.

• AWS announces ExtendDB, an open source DynamoDB-compatible adapter.

• AWS SAM CLI adds AWS CloudFormation Language Extensions support to accelerate local serverless development.

🔥 Tip: Check out https://aws-news.com/ for the very latest up-to-date serverless releases as they happen, created by the talented AWS Serverless Hero Luc van Donkersgoed.

✖️ Social of the Week

This week’s social is by Matias Undurraga on LinkedIn:

“I am seeing a clear pattern across teams I work with: AI native teams are getting smaller.
…
But one trend feels hard to ignore: AI infused teams are trying to reduce coordination overhead. And one of the simplest ways to reduce coordination is to reduce the number of people who need to coordinate. Smaller teams. Faster loops. Less handoff. More ownership.“ - Matias

Matias shares his thoughts on product management and AI team size/design, which resonates strongly with the ideas I explored in my previous article, ‘The Architect-Developer: Orchestrating the AI-Driven Development Lifecycle (AI-DLC)’.

In that piece, I discuss how the developer’s role has shifted from author to orchestrator/architect, and we now work in smaller teams (pods). The keyboard is no longer the bottleneck. Clarity of thought and access to the right engineering and architecture experience are essential. My teams are made up of 2-3 people max, one being the orchestrator (typically a Principal Architect/Engineer, with a senior engineer, and a more junior cloud engineer).

Organisations no longer need armies of coders. They need orchestrators who can translate business intent into technical direction, guide AI agents, validate outputs, and ensure that what is built aligns with standards, strategy, and long-term maintainability.

What are your thoughts on this? Leave a comment

Leave a comment

👷🏻 Tools & Frameworks

Check out the latest open-source frameworks, news, and tool updates from the past two weeks.

Martin Hicks announces Dynoxide 0.1.0 with browser and Docker support: a local DynamoDB emulator backed by SQLite with millisecond startup, a 3 MB download, no Docker and no JVM required. Now with browser support for client-side testing.

Ran Isenberg releases Propel - Kanban-style task management for macOS. Move fast, stay focused.

Bedrock Lens — A tool for visualising and understanding Amazon Bedrock model interactions. Worth checking out if you’re building with Bedrock (see above in the article).

Kiro CLI — Claude Opus 4.8 Support — Kiro adds support for Claude Opus 4.8, bringing the latest model capabilities to the IDE and CLI.

Ampt — Faster Iteration with Reset and CLI Improvements — New CLI commands for resetting environments, managing params, and checking your identity, all without leaving the terminal.

Terraform Skill by Anton Babenko — A skill for AI coding agents that brings Terraform expertise directly into your agentic workflow. Great for teams using Terraform alongside AI-driven development.

😂 Just for Fun

This issue, our just-for-fun post is by Aku Nikkola on LinkedIn, showing off this new (fake) product!

This resonates with me personally, all jokes aside, because I do exactly this when I have the kids at their clubs and waiting around, when I’m travelling, or waiting around killing time. There’s something about those pockets of time that are perfect for getting into a flow with a side project and AI-assisted development! In the house, I am mainly on the Mac mini, else the MacBook and Kiro when I am out and about!

I do tend to use Kiro Web a lot now, meaning I don’t always need to do this, but I haven’t quite removed the IDE altogether yet!

It also reminds me of Sébastien Stormacq’s post this week about walking in an airport in the age of AI agents, where he’s literally walking through the terminal while his agents are working away in the background with the MacBook half open!

What are your thoughts? Leave a comment below.

Leave a comment

🎙️ YouTube & Podcasts

Here are some of my favourite videos and podcasts over the past two weeks covering serverless, AI, architecture, and software engineering.

⭐ My favourite video this week is on the The Pragmatic Engineer channel, where they chat with the fantastic Dax Raad! Go watch this!

The AWS Developers Channel interviews Ryan Cormack, Principal Engineer at Motorway!

Martin Fowler & Kent Beck discuss “Agile Evolution & the Future of Software Engineering”

The Build. Automate. Ship Smarter in the Cloud channel again chats with Ryan Cormack, this time about Kiro Web! (Ryan is a busy guy right now!)

Lenny’s Podcast covers “A rational conversation on where AI is actually going“ with Benedict Evans. This is a great watch!

The GOTO channel has Kris Jenkins discussing “Building Context-Rich AI Tools with MCP“.

The ‘My 50 Cents’ channel covers “You don’t need the Architect title to do architecture”, chatting with Mahdi, tech leader at Fashion Cloud.

Weekly Case Study 🔍

This week’s case study is how ‘Vercel Transforms Developer Experience with Amazon Bedrock’:

Vercel shares how Amazon Bedrock powers their AI coding agents. The video highlights why Vercel selected Amazon Bedrock for its security, performance, and scalability capabilities while leveraging its existing AWS partnership. With Amazon Bedrock’s reliable infrastructure supporting 75 million code generations at the time of the video being created, Vercel continues empowering developers to create next-generation user experiences through their developer frameworks, AI agents, and front-end cloud platform.

🗣️ Inspirational Quotes and Thoughts

This week’s inspirational thought covers Miller’s Law: a psychological principle regarding cognitive load and working memory.

“The average person can only keep 7 (± 2) objects in working memory.”

— George A. Miller, 1956

Miller’s Law tells us that human working memory has hard limits. We can hold roughly seven items in our heads at once before things start falling through the cracks. This has profound implications for how we work with AI and agents today, in my opinion.

Whether you’re using autonomous agents running in the background, IDE-driven spec-driven development with human-in-the-loop, or both at the same time, the cognitive load is real and cumulative, in my opinion.

Personally, I currently have three projects running simultaneously, typically, each with its own specs, contexts, steering, MCP integrations, skills, agent outputs to review, and decisions to make. That’s many separate mental models, sets of trade-offs and contexts, and streams of AI-generated output that need human judgment applied.

The cognitive overhead of orchestrating multiple AI-driven workstreams is no less than writing code by hand; it’s different. You’re trading typing fatigue and thinking about your finer level coding approaches for decision fatigue and throughput (some may describe it as a fire hose). And decision fatigue is insidious because it doesn’t feel like “work” in the traditional sense, yet it drains you just the same.

What are your own thoughts? Leave a comment below.

Leave a comment

🗳️ Poll of the Week

In the last poll, we asked the question, “Are you building AI agents in production?”

I was shocked to see 100% of people state no, but looking into it. I have been doing a lot with AgentCore recently, and the biggest issue for me is tracking token usage; it’s just not that clear and it's always a concern for me when tracking in the console.

In this issue, we ask the question: “Are you concerned with token use and costs in production when building AI and agents into your solutions?”

Please feel free to leave a comment below on your answer.

Leave a comment

📅 Serverless Events

The following serverless events are upcoming, so mark your calendars.

We are organising the AWS North Community Conference again this year, and CFP is currently open! Other amazing events:

• AWS Comsum Birmingham - 4th June 2026.

• ACD Midwest - 24th June 2026.

• ACD Bengaluru - 11th July 2026.

• ACD Poland - 8th Sept 2026.

• ACD Adria - 22nd Oct 2026.

• AWS Community Summit Birmingham - 4th June 2026.

• AWS North Community Conference - 15th Oct 2026.

Would you happen to have any upcoming events that you would like to highlight? Message me below!

⭐ Spotlight

This issue’s spotlight falls on AWS Developer Advocate Elizabeth Fuentes Leone!

Elizabeth is a fantastic advocate for the AWS community, consistently sharing knowledge, creating content, open-source work, and helping developers understand complex AI and cloud concepts (she has reached more than 89,000+ developers through technical content). Her work bridges the gap between technical depth and accessibility for people starting their journey, making advanced topics approachable for developers at all levels.

I also want to highlight this excellent video from Elizabeth from this week: ‘Building with Amazon Bedrock’, a great example of the kind of high-quality, practical content she produces for the community.

Whether it’s through her talks, blog posts, videos, or community engagement, Elizabeth embodies what it means to be a developer advocate, meeting developers where they are and helping them level up.

Thank you for everything you do for the community, Elizabeth!

Thank you for reading the latest Serverless Advocate Newsletter!

If you want to find out a little more about me, please have a look at:

https://www.serverlessadvocate.com/

See you next time,

Lee

Twitter | LinkedIn | YouTube | BlueSky | Website

Welcome

In the last issue, our serverless expert was Adit Modi, and our spotlight fell on AWS Community Builder Viola Lykova!

This issue, our serverless expert is AWS Senior Solutions Architect Lefteris Karageorgiou, our spotlight is on AWS Kiro Principal Tech Lead Brian Beach, and we look at the latest AWS service releases, blog posts, hints and tips, news, and more!

Subscribe now

This week’s newsletter is sponsored by Study From Experts, The Exclusive Home of AWS Deep-Dive Mastery. Gain access to a private library of advanced courses from the world’s most renowned AWS experts, with content so specialised, you won’t find it on any other platform.

📰 Articles that caught the eye

This week, I shared an article covering “Automated Certificate Generation with Event-Driven Serverless Pipelines on AWS”.

In this article, we walk through how to automatically generate course completion certificates using an event-driven serverless pipeline on AWS, built with EventBridge, SQS, Lambda, S3, and PDFKit.

Here are some other stand-out articles I read during the past two weeks in the World of Serverless, AI, engineering and architecture!

⭐ My favourite article over the past two weeks is by Allen Helton, who has a great article titled ‘Local agents scare me‘ where he discusses the security implications of running agents (great examples discussed in here too - so go check it out!).

Siddhant Khare has a deeply honest article titled ‘AI Fatigue Is Real and Nobody Talks About It’, exploring the paradox of AI-assisted productivity, why shipping more code than ever can leave you more drained than ever, and what to do about it.

Luca Cavallin covers ‘Platform Engineering End-to-End’, a comprehensive walkthrough of the full arc of platform engineering, from why it exists, to the pillars, to what success looks like.

Ran Isenberg has a great article titled ‘Is AWS Lambda Tenant Isolation Mode Enough for SaaS?’. This post explores the concept of tenant isolation, how Lambda’s tenant isolation mode works, where it helps, where it falls short, and how to decide whether it belongs in your SaaS architecture.

Monica Colangelo features ‘Per-tenant DynamoDB Isolation with the Token Vending Machine Pattern’, a detailed end-to-end implementation.

Michael Walmsley covers ‘Layered Configuration in Claude Code’, explaining how Claude Code’s settings system actually works.

Pahud Hsieh has a great article titled ‘AI Can’t Fix What It Can’t See: How cdk diagnose Enables Autonomous CDK Remediation’, showing how the new cdk diagnose CLI subcommand surfaces deployment failures with CDK-aware context.

Luciano Mammino covers ‘Writing Middlewares for Rust Lambda Functions’, a deep dive into the tower middleware pattern.

Andreas John covers ‘Enforcing Python Architecture Rules with PyTestArch’, showing how to express architectural rules, i.e., “this package must not import that one”, as regular pytest tests that fail the build when someone crosses a line you drew.

Last week, we had the pleasure of going live with a new course on the Study From Experts platform: “Architecting Java Systems in the Cloud“ by Sascha Möllering

Overview:

This course explores how to design and understand modern Java systems in cloud environments. Using the AcmeCorp Platform as a realistic microservices reference architecture, we examine architectural decisions around service boundaries, data access, and containerised deployments. You will instrument applications with Prometheus metrics and analyse system behavior through Grafana dashboards to identify issues such as inefficient queries and startup bottlenecks. We also explore the evolution of the JVM from Java 11 to 21+ and evaluate techniques like AppCDS, CRaC, and native images. Throughout the course, you will learn how to create reproducible benchmarks and reason about the interaction between architecture, observability, and the JVM.

🎓 Ask the Expert

Each issue, I ask a different AWS expert the same three questions to get their personal insights. This issue, we have AWS Senior Solution Architect Lefteris Karageorgiou:

Opinions are the author’s and do not express the views of their employer.

1. What is one common mistake you see teams making when building their solutions, and how can they avoid it?

A common mistake is that companies think serverless is all about AWS Lambda, and as a result, they become overly concerned about cold starts.

In reality, serverless is much broader than Lambda. Many use cases can be solved without Lambdas at all. For example, by using direct integrations with Amazon API Gateway or orchestrating workflows with AWS Step Functions.

The key is to step back and evaluate whether Lambda is actually needed. If it’s only being used to move data from one service to another, it’s often unnecessary.
That said, when you do need Lambda, you should know how to optimise it properly. Three of the most effective techniques are:

1. Optimise memory: Use the AWS Lambda Power Tuning tool to find the optimal memory configuration. Since memory allocation also scales CPU, the right balance can significantly reduce both execution time and cost.

2. Minimise deployment size: Smaller packages lead to faster cold starts, so remove unused dependencies and keep artefacts lean.

3. Use SnapStart: Especially for Java workloads, SnapStart can dramatically reduce cold start latency by initialising functions ahead of time.

By using Lambda intentionally and optimising it when needed, you can avoid unnecessary complexity and get the best out of serverless.

2. Which tool, package, or AWS service are you most excited about right now, and why?

Right now, I’m most excited about AWS Lambda durable functions. This is something the ecosystem has needed for a long time, bringing orchestration closer to the application layer. Previously, you could achieve similar outcomes with AWS Step Functions, but local development, testing, and debugging were often cumbersome.

Although this may seem similar to Step Functions, the trade-offs are important:

Use durable functions when:

- Your team prefers standard programming languages and familiar development tools
- Your application logic primarily lives inside Lambda functions
- You’re building Lambda-centric systems with tight coupling between workflow and business logic

Use Step Functions when:

- You need a visual workflow representation for cross-team visibility
- You’re orchestrating multiple AWS services and want native integrations without writing custom SDK code
- You want zero-maintenance infrastructure (no patching or runtime concerns)

Durable functions make it much easier to build complex, long-running workflows directly in code, opening the door to more advanced use cases like multi-step processes and agent-style orchestration, without sacrificing developer experience.

3. What is your favourite trick or tip that the readers may find interesting?

A common anti-pattern I see is treating AWS Lambda as “one service = one function.” This often leads to architectures with hundreds of tiny Lambda functions, which quickly become difficult to manage, deploy, and reason about.

Instead, treat your Lambdas as microservices within a bounded context. It’s perfectly fine, and often preferable, to group related functionality together. For example, within a “users” domain, you can have both 'createUser' and 'deleteUser' handled by the same Lambda.

When deciding how to group your functions, consider these factors:

- Bounded contexts
- Team organisation
- Scoped IAM permissions
- Common code dependencies
- Common downstream dependencies
- Initialisation time (cold start impact)
- Memory configuration

A powerful way to implement this approach is the Lambda Web Adapter pattern. Instead of creating one Lambda per HTTP endpoint, you run a traditional web framework inside a single Lambda and handle routing internally. This allows you to use familiar frameworks like Express.js, Flask, Django, Spring Boot, or ASP.NET.

The result is a more maintainable system that aligns with real domain boundaries, without losing the benefits of serverless.

Also, go check out the following from Lefteris:

His blog: https://blog.thecloudengineers.com/
His latest book: https://a.co/d/0efajvXG

🧠 Tips & Tricks

This week’s tip is about the Get Hands-On With Serverless training workshops by AWS:

AWS is running a series of hands-on serverless workshops designed to help you build real-world applications using AWS Lambda, API Gateway, DynamoDB, and more.

These are instructor-led sessions where you get to build alongside AWS experts, a great opportunity to level up your serverless skills with guided, practical experience.

Check out the full schedule and register at aws-experience.com.

Leave a comment

🚀 New Releases

Here are the latest and most interesting releases over the past two weeks in the AWS World:

⭐ My favourite releases over the past two weeks are DSQL CDC in preview, the Claude Platform being available on AWS, and the new direct service integrations with EventBridge scheduler.

• Amazon CloudWatch Logs announces increased query result limits.

• AWS Organisations now supports higher quotas for service control policies (SCPs).

• Amazon CloudFront announces Passthrough Mode for mutual TLS (Viewer).

• Amazon Bedrock Introduces Advanced Prompt Optimization and Migration Tool.

• Amazon CloudFront announces support for OCSP Revocation for Mutual TLS (Viewer).

• Three new models for speech recognition and text-to-speech are now available in Amazon SageMaker JumpStart.

• AWS Transform adds agentic AI assistant to the AWS Toolkit for Visual Studio.

• AWS Transform agents now available in Kiro, Claude, Cursor, and Codex.

• Amazon Aurora DSQL now supports change data capture (Preview).

• Reference stack outputs across accounts and Regions with AWS CloudFormation and CDK.

• AWS Lambda supports scheduled scaling for functions on Lambda Managed Instances.

• Amazon EventBridge Scheduler adds 619 new SDK API actions, including Lambda Managed Instances.

• Amazon Route 53 Domains adds support for 34 new Top Level Domains including .app, .dev, and .health.

• Claude Platform on AWS is now generally available.

• Agents that transact: Amazon Bedrock AgentCore now includes Payments (preview).

• Amazon Aurora DSQL now supports the JSON data type with compression.

🔥 Tip: Check out https://aws-news.com/ for the very latest up-to-date serverless releases as they happen, created by the talented AWS Serverless Hero Luc van Donkersgoed.

✖️ Social of the Week

This week’s social is by Jason Mikula on LinkedIn:

Coinbase fires 14% of staff, then announces they’re replacing them with AI agents, then has non-technical staff building features with AI. Then they have degraded services. This is the reality of what’s happening in the industry right now:

I personally feel that non-technical staff should not be using AI to produce code and features, as it’s everything that happens under the surface that is key:
caching, resiliency, scalability, testing, observability, security, and more...

The surface-level code is the easy part. The hard part is everything you can’t see. And this is coming from a very experienced architect who has built systems for 25 years.

This post by Anupam Ghosal on LinkedIn (image below) actually demonstrates this perfectly, showing what happens when AI-generated code hits production without the engineering fundamentals underneath it.

I also want to back this up with a great quote from a friend of the newsletter, Paul Stack on LinkedIn, who articulates this brilliantly. The code is the easy part. The architecture, the trade-offs, the operational excellence: that’s where the real engineering lives.

What are your thoughts on this? Leave a comment

Leave a comment

👷🏻 Tools & Frameworks

Check out the latest open-source frameworks, news, and tool updates from the past two weeks.

⭐ Firstly, I wanted to highlight Kiro Web (Preview) — Kiro Web is now available at app.kiro.dev for Kiro Pro, Pro+, and Power users. Chat with the agent to explore ideas, fix bugs, and shape changes, or hand off end-to-end tasks that finish as pull requests.

Other recent releases:

The AWS MCP Server is now Generally Available — The official AWS MCP Server has hit GA, providing a standardised way to connect AI agents to AWS services through the Model Context Protocol.

Nova Act Quick Deploy — A quick-deploy template for Amazon Nova Act, making it easier to get started with browser automation agents.

agent-skills — Production-grade engineering skills for AI coding agents by Addy Osmani. A fantastic collection of reusable skills to level up your agentic coding workflows.

Agent Toolkit for AWS — AWS’s official toolkit for building AI agents that interact with AWS services, providing pre-built tools and patterns for common AWS operations.

Node.js v26 — Node.js v26 is out! The latest major release brings new features and improvements to the JavaScript runtime.

Tolaria — A second brain for the AI era. Organise your notes as Markdown files, with native relationships, Git, local agents, and direct AI model providers. Free forever, open source, no account required.

Strands Agents TypeScript 1.0 — Version 1.0 of the Strands Agents TypeScript SDK is here. Build type-safe AI agents with full model provider support, custom tools, plugins, and multi-agent orchestration in Node.js and the browser.

😂 Just for Fun

This issue, our just-for-fun post, is by Eduardo Ordax on LinkedIn:

In all seriousness, I have a setup with a mixture of GitHub labels and glob patterns which allow me to bypass full pipeline deploys based on the extent of what has changed in the merged PR, and also run a huge amount of tests up front using git hooks (pre-commit, pre-push, etc.). This is especially important since I am a monorepo fan!

How do you manage this in your products? Leave a comment below.

Leave a comment

🎙️ YouTube & Podcasts

Here are some of my favourite videos and podcasts over the past two weeks covering serverless, AI, architecture, and software engineering.

⭐ My favourite video this week was my business partner Mark Sailes on Serverless Office Hours with our friend Julian Wood, discussing how we built Study From Experts with Kiro.

Peter Eijgermans talks to Luca Mezzalira and covers ‘How to Build Real-Time Applications Using Micro-Frontends’.

The AWS Developers channel covers ‘Cut AI Token Usage by 96%? Here’s How AWS Strands Agents Does It’.

The AWS ‘Show and Tell’ channel covers ‘Getting Started Quickly with Managed Harness in AgentCore’.

James Eastham has a great video titled ‘Serverless & OpenTelemetry ❤️ Better Together’.

The Kiro team covers ‘Kiro Subagents Explained’.

AWS and Erik Hanchett cover ‘AgentCore Managed Harness: Getting Started’.

Lenny’s Podcast features ‘AI Era Skills: Why Cultivating Agency Matters More Than Job Titles’ with Max Schoening from Notion.

Weekly Case Study 🔍

This week’s case study is by Siemens on AWS:

Siemens, a leading technology company operating in more than 130 countries with over 300,000 employees, used Amazon Bedrock and Amazon Nova Foundation Models to streamline its complex global search across 15–20 different Siemens sites.

Using Amazon Nova 2 Lite, the company achieved a 70% reduction in costs, a 300% improvement in search speed, and an 11% increase in user satisfaction, all deployed in less than one year from concept to production. Customers can now query using natural language rather than restrictive keywords, and the solution has eliminated no-results searches entirely.

The architecture uses AWS Lambda functions to orchestrate validation agents, classification agents, and summariser agents, which is a great example of serverless AI in production at enterprise scale.

🗣️ Inspirational Quotes and Thoughts

This week’s inspirational thought is by Brian Kernighan covering ‘Kernighan's Law‘.

“Everyone knows that debugging is twice as hard as writing a program in the first place. So if you're as clever as you can be when you write it, how will you ever debug it?“

- Brian Kernighan

For anybody who has worked with me in teams, you will know one of my key tenets is keeping it simple. It’s even one of the top items in our software engineering and architecture manifesto at Leighton.

The reason for this is simple: the more complex you write your code with layer upon layer of abstraction and dependency injection, and the more complex your architecture, the harder it is to debug, evolve, understand, and reason about.

Leave a comment

🗳️ Poll of the Week

In the last poll, we asked the question, “In the AI-DLC, is there still room for delivery-focused roles?”.

Interestingly, all 100% people said ‘No’, which is not super surprising in all honesty. With smaller cycles of work in the AI-DLC (bolts) and fewer ceremonies, most small teams can absorb this work happily.

In this issue, we ask the question, “Are you building AI agents in production?”

Please feel free to leave a comment below on your answer.

Leave a comment

📅 Serverless Events

The following serverless events are upcoming, so mark your calendars.

We are organising the AWS North Community Conference again this year, and CFP is currently open! Other amazing events:

• AWS Comsum Birmingham - 4th June 2026.

• ACD Midwest - 24th June 2026.

• ACD Bengaluru - 11th July 2026.

• ACD Poland - 8th Sept 2026.

• ACD Adria - 22nd Oct 2026.

• AWS Community Summit Birmingham - 4th June 2026.

• AWS North Community Conference - 15th Oct 2026.

Would you happen to have any upcoming events that you would like to highlight? Message me below!

⭐ Spotlight

This issue’s spotlight falls on Principal Tech Lead of the AWS Kiro product, Brian Beach!

I am so impressed with everything Brian and the team are doing with Kiro right now, and how he is promoting the product and sharing new features almost daily! (Go follow Brian on LinkedIn as his posts are fantastic!)

We built the Study From Experts platform with Kiro using Spec-Driven Development and have been with them from the start.

Brian’s consistent dedication to sharing knowledge, demoing new capabilities, building amazing new features, his videos and live streams, and engaging with the community is exactly what makes the Kiro developer ecosystem thrive.

Whether it’s Kiro Web, the CLI, IDE, Powers, hooks, steering, or more, Brian is always there showing how it works in practice, not just in theory (his video shorts on LinkedIn rock)

Thank you for everything you do for the community, Brian!

Thank you for reading the latest Serverless Advocate Newsletter!

If you want to find out a little more about me, please have a look at:

https://www.serverlessadvocate.com/

See you next time,

Lee

Twitter | LinkedIn | YouTube | BlueSky | Website

Welcome

In the last issue, our serverless expert was Kenta Goto, and our spotlight fell on AWS Principal Solutions Architect Heeki Park!

This issue, our serverless expert is AWS Ambassador Adit Modi, our spotlight is on AWS Community Builder Viola Lykova, and we look at the latest AWS service releases, blog posts, hints and tips, news, and more!

Subscribe now

This week’s newsletter is sponsored by Study From Experts, The Exclusive Home of AWS Deep-Dive Mastery. Gain access to a private library of advanced courses from the world’s most renowned AWS experts, with content so specialised, you won’t find it on any other platform.

📰 Articles that caught the eye

This week, I shared an article covering “Securing Private Video Content with CloudFront Signed URLs and Serverless on AWS“.

In this article, we walk through how to secure private video content stored in Amazon S3 using CloudFront signed URLs, all built with a serverless architecture using AWS CDK and Lambda functions.

Here are some other stand-out articles I read during the past two weeks in the World of Serverless, AI, engineering and architecture!

⭐ My favourite article over the past two weeks is by Werner Vogels and the Lambda Networking team titled ‘The Invisible Engineering Behind Lambda’s Network’.

Eric Johnson covers ‘Lambda Managed Instances: A Working Demo and the Math Behind It’, a thorough walkthrough of the new Lambda Managed Instances feature that lets multiple invocations share the same execution environment concurrently.

Praneeta Prakash has a great article titled ‘The PM Agent Loop’, sharing a deeply practical look at how she rewired her entire product workflow around AI agents.

Vadym Kazulkin covers ‘Serverless Applications on AWS with Lambda Using Java 25, API Gateway and DynamoDB — Part 6 Using GraalVM Native Image’, providing a step-by-step guide on converting a Lambda application to GraalVM Native Image and deploying it as a custom runtime.

Viola Lykova features ‘A Year of Building snappycart in Public — 7 Mistakes I Learned the Hard Way’, a candid and refreshing look at the lessons learned from building an open-source React cart package in public.

Joe Njenga covers ‘Anthropic Just Made Cheap Models Think Like Opus’, exploring how Anthropic’s Claude Advisor Tool uses small open-weight models to generate dynamic, per-instance advice that steers larger models toward better outputs. Some amazing optimisation is going on here!

Ray Goh has a great article titled ‘Anthropic Opus 4.6 vs 4.7 — Which is Better? A Code Quality Experiment’, a controlled experiment comparing code quality across thirteen dimensions using the same harness, same spec, and same tests.

🎓 Ask the Expert

Each issue, I ask a different AWS expert the same three questions to get their personal insights. This issue, we have AWS Ambassador Adit Modi:

Opinions are the author’s and do not express the views of their employer.

1. What is one common mistake you see teams making when building their solutions, and how can they avoid it?

One mistake I see frequently is teams lifting and shifting their existing monolithic mindset into serverless — they'll create a single, monolithic Lambda function that handles everything instead of decomposing into smaller, purpose-built functions. This leads to bloated deployment packages, longer cold starts, and tightly coupled code that's hard to debug and scale independently.

The fix is to think in terms of events and single responsibilities from day one. Each Lambda should do one thing well.

Pair that with proper Infrastructure as Code — whether it's CDK, Terraform, or CloudFormation — so your architecture stays maintainable as it grows. I've seen a few members at the AWS User Group Vadodara struggle with this, and once they shift to an event-driven, decoupled design, the operational benefits of serverless truly click.

2. Which tool, package, or AWS service are you most excited about right now, and why?

I'm really excited about the evolution of AWS Amplify Gen 2. It has come a long way from being a "beginner-friendly" tool to a genuinely powerful full-stack serverless platform. I've used it across multiple projects — from our AWS User Group Vadodara app (Lambda + API Gateway + DynamoDB + Cognito) to portfolio sites — and the developer experience for going from code to a fully deployed serverless backend keeps getting better.

Beyond that, I'm closely watching the intersection of serverless and AI/ML. Running inference workloads on services like AWS Inferentia2 with frameworks like vLLM, combined with serverless orchestration patterns, is opening up cost-effective AI serving that was unthinkable a couple of years ago. The "pay for what you use" model applied to AI inference is going to be a game-changer.

3. What is your favourite trick or tip that the readers may find interesting?

Always invest in observability before you need it. With serverless, you don't have a server to SSH into when things go wrong. My go-to setup is structured JSON logging in every Lambda from day one, combined with CloudWatch Alarms on key metrics like error rates and duration percentiles — not just invocation counts. It costs almost nothing but saves you hours during incident response.

Another tip: use DynamoDB single-table design thoughtfully. In our AWS UG Vadodara feedback app, we run seven different entity types (events, sessions, feedback, QR links, scans, job seekers, job posts) and understanding access patterns upfront made the difference between a clean, performant backend and a mess of scan operations. Spend time on your access patterns before writing a single line of code.

🧠 Tips & Tricks

This week’s tip is about the new Kiro Powers capability:

Kiro has introduced Powers some time ago, a way to add specialised context and tools to Kiro agents on demand without overloading your MCP context.

The standout here is that the AWS DevOps Agent is now available as a Kiro Power, giving you access to 36 DevOps Agent operations through a single MCP tool, right from your IDE.

What can you do with it? Incident investigation with root cause analysis across CloudWatch, X-Ray, IAM, and service topology. Cost optimisation with account-level spend analysis and auto-generated Terraform/CDK changes. Architecture reviews that read your IaC files locally and run them through DevOps Agent for production readiness checks. And knowledge discovery for instant access to AWS runbooks and documentation.

One tool. 36 operations. Zero context switching. Check out the full list at the Kiro Powers GitHub repo.

Leave a comment

🚀 New Releases

Here are the latest and most interesting releases over the past two weeks in the AWS World:

⭐ My favourite release over the past two weeks is the addition of Claude Opus 4.7 to Amazon Bedrock and Kiro!

• Amazon Redshift Serverless AI-driven scaling is now the default for new workgroups.

• Amazon Bedrock AgentCore Gateway and Identity support VPC egress.

• Amazon Athena simplifies federated queries with managed connectors.

• AWS Elastic Beanstalk AI-powered environment analysis now supports Windows.

• Amazon S3 now supports five additional checksum algorithms.

• AWS Lambda durable functions are now available in 16 additional AWS Regions.

• Amazon Bedrock AgentCore adds new features to help developers build agents faster.

• Amazon MSK Serverless expands to 13 new AWS regions.

• Amazon CloudWatch pipelines now supports configuration of processors via AI.

• AWS Lambda Durable Execution SDK for Java GA.

• Amazon Aurora serverless: Up to 30% better performance, smarter scaling, and still scales to zero.

• Amazon S3 Express One Zone now supports S3 Inventory.

• Amazon CloudWatch now supports cross-region telemetry auditing and enablement rules.

• Claude Opus 4.7 is now available in Amazon Bedrock.

🔥 Tip: Check out https://aws-news.com/ for the very latest up-to-date serverless releases as they happen, created by the talented AWS Serverless Hero Luc van Donkersgoed.

✖️ Social of the Week

This week’s social is by Bryan Liles on LinkedIn:

“These are my professional tenets:

Complexity is the real enemy.
Build to find joy.
Only use process to remove friction.
Improving work is the work.
Find focus in outcomes.”

I love the simplicity and directness of this. If I had to pick a favourite, it would be “Complexity is the real enemy.” That resonates deeply with everything I believe about building systems.

The moment you add complexity without a clear reason, you’re creating future pain for someone, and that someone is usually you. Keep it simple people!

What are your thoughts on this post? Leave a comment

Leave a comment

👷🏻 Tools & Frameworks

Check out the latest open-source frameworks, news, and tool updates from the past two weeks.

⭐ Firstly, I wanted to highlight this incredible post from Kenta Goto on LinkedIn:

He built his own CDK from scratch, and it deploys 5x faster than the official CDK!

Let that sink in for a moment. He built the CDK engine entirely from scratch, using AWS SDK calls instead of CloudFormation under the hood. It properly rolls back on failure, handles parallel deploys that respect dependencies, supports cross-region deployments, and even handles context lookups with re-synth. Great work my friend!

Other recent releases:

agent-skills — Production-grade engineering skills for AI coding agents by Addy Osmani, already at over 21,000 GitHub stars. A fantastic collection of reusable skills to level up your agentic coding workflows.

rtk — A CLI proxy that reduces LLM token consumption by 60–90% on common dev commands. Single Rust binary, zero dependencies. Another great tool in the emerging “make AI cheaper” category.

ai-instruct — Drop-in markdown guides and Claude Code skills for the domains developers Google when shipping a website: AWS deployment, Google Analytics 4, SEO/LLMO and agent-readiness, web accessibility, and Core Web Vitals. Install with npx ai-instruct init.

desktop_mario — A fun desktop Mario game that runs on your screen. Because sometimes you just need a break from debugging your code.

ai-pm-maturity-assessment-skill — An AI-powered product management maturity assessment skill, helping teams evaluate and improve their PM practices with structured assessments.

Kiro CLI 2.0 — A major update to the Kiro CLI with headless mode for CI/CD pipelines, native Windows support, and a refreshed UX with subagent monitoring and real-time task lists. The headless mode is a game-changer for automating deployments end-to-end.

Kiro CLI Knowledge Management — Brian Beach highlights that the Kiro CLI has added a persistent knowledge base that lets you store and search contextual information across chat sessions. Great for indexing coding standards, security policies, and other context that lives outside the workspace.

⚠️ Security Warning: Vercel got hacked last week. The attack vector was a compromised third-party AI tool with a Google Workspace OAuth app. One OAuth grant led to one Vercel employee account, which led to source code, NPM tokens, GitHub tokens, and production secrets.

😂 Just for Fun

This week’s “just for fun” post is by Prasad Rao on LinkedIn:

“And that’s how you increase your productivity! By prompting Claude correctly!”

Prasad Rao

Leave a comment

🎙️ YouTube & Podcasts

Here are some of my favourite videos and podcasts over the past two weeks covering serverless, AI, architecture, and software engineering.

⭐ My favourite video is by Luca Mezzalira, interviewing James Eastham on ‘How to Become an Architect’. A genuinely inspiring watch.

Lennys Podcast covers ‘An AI State of the Union: We’ve Passed the Inflection Point & Dark Factories are Coming’.

The Modern Engineering Channel covers ‘Top AI Trends From 100 Interviews’.

The AWS Developers channel covers ‘Build an AI Agent From Scratch With Python’, a hands-on tutorial building your own agent using the Strands Agents SDK.

Another great video from the AWS Developers channel with ‘What is an Agent Harness? (And How We Built One)’.

GOTO Conferences features ‘Building, Managing & Governing APIs on AWS’ with Giedrius Praspaliauskas.

Derek Comartin has a great video titled ‘Testable Code Doesn’t Mean “Interfaces Everywhere”’.

Lenny’s Podcast covers ‘Hard Truths About Building in the AI Era’.

The Serverless Craic team discusses ‘Psychological Safety in the AI Era’.

DDD Europe features David Whitney with ‘The Unbearable Weight of Architecture’, covering techniques for understanding context, designing with trade-offs in mind, and building software that survives inevitable change.

Weekly Case Study 🔍

This week’s case study is by Meshed Group on AWS:

Meshed Group, an EdTech provider trusted by over 300 education organisations globally, built an AI-driven student summary feature using Amazon Bedrock that condenses academic, behavioural, and financial information into concise insights for staff review.

Since adopting this approach, education providers have reported an 80–85% reduction in the time spent synthesising student information, are reviewing twice as many students per day, and have reduced review time by 8 minutes per student.

🗣️ Inspirational Quotes and Thoughts

This week’s inspirational quote, law, or thought is Zawinski’s Law (1968) by Jamie Zawinski:

“Every program attempts to expand until it can read mail. Those programs which cannot so expand are replaced by ones which can.”

— Zawinski’s Law of Software Envelopment (Jamie Zawinski, born 1968)

Zawinski later clarified in 2020:

“My point was not about copycats, it was about platformisation. Apps that you ‘live in’ all day have pressure to become everything and do everything. An app for editing text becomes an IDE, then an OS. An app for displaying hypertext documents becomes a mail reader, then an OS.”

This feels more relevant today than it ever has, and I wanted to talk about it in the context of AI-assisted development. With tools like Cursor, Kiro, and Claude Code it’s making it trivially easy to ship features, and the temptation to say “yes” to every user request has never been stronger. Someone files an issue? Ship it. A user suggests a nice-to-have? Why not, it only takes ten minutes with an agent.

The problem is that feature bloat is no longer gated by engineering capacity. When the cost of adding a feature drops to near zero, the discipline of saying “no” becomes the most important skill a product team can have. Every feature you add is a feature you maintain, a feature that increases your test surface, a feature that makes your onboarding story more complex, and a feature that may dilute your core value proposition.

What are your own thoughts and experiences of this quote? Feel free to leave a comment below.

Leave a comment

🗳️ Poll of the Week

In the last poll, we asked the question, “With AWS services like App Runner and WorkMail now being deprecated, is this a worry for you when designing and architecting systems on AWS?”.

Interestingly, 75% said Yes, and 25% said no. I would say for me personally I sit firmly in the 75% camp, and it is always a consideration when building production grade apps. There are a few services I won’t use as I have a gut feel they may get relegated to the same list!

This issue, we ask the question, “In the AI-DLC, is there still room for delivery-focused roles?”

Please feel free to leave a comment below on your answer.

Leave a comment

📅 Serverless Events

The following serverless events are upcoming, so mark your calendars.

• ACD Türkiye - 9th May 2026.

• AWS Comsum Birmingham - 4th June 2026.

• ACD Midwest - 24th June 2026.

Would you happen to have any upcoming events that you would like to highlight? Message me below!

⭐ Spotlight

This issue’s spotlight falls on AWS Community Builder, Viola Lykova!

Viola is also a Software Engineer and Cypress Ambassador based in London, with a strong focus on software testing, open-source, quality engineering, and building clean, well-tested systems.

She’s also the Founding Engineer of the 5 A Day app and the creator and maintainer of snappycart, an open-source headless React cart system. She is also active as an AWS Community Builder, speaker, blogger, produces videos on YouTube, and more!

Thank you for everything you do for the community Viola!

Thank you for reading the latest Serverless Advocate Newsletter!

If you want to find out a little more about me, please have a look at:

https://www.serverlessadvocate.com/

See you next time,

Lee

Twitter | LinkedIn | YouTube | BlueSky | Website

Welcome

In the last issue, our open source expert was Ricardo Sueiras, and our spotlight fell on AWS Principal Developer Advocate Aaron Hunter!

This issue, our serverless expert is AWS DevTools Hero Kenta Goto, our spotlight is on AWS Principal Solutions Architect Heeki Park, and we look at the latest AWS service releases, blog posts, hints and tips, news, and more!

Subscribe now

This week’s newsletter is sponsored by Study From Experts, The Exclusive Home of AWS Deep-Dive Mastery. Gain access to a private library of advanced courses from the world’s most renowned AWS experts, with content so specialised, you won’t find it on any other platform.

📰 Articles that caught the eye

Here are some stand-out articles I read during the week in the World of Serverless, AI, engineering and architecture!

⭐ My favourite article over the past two weeks is by Danilo Poccia discussing ‘From Prompt Engineering to Agent Experience’,

Eric Johnson covers ‘Lambda Just Got a File System. I Put AI Agents on It’, a hands-on walkthrough of building a serverless code review system using S3 Files, where AI agents share a mounted S3 workspace orchestrated by a durable function.

Ran Isenberg has a great article titled ‘AI Made Everyone a Builder and That’s a Problem’, arguing that the low barrier to entry from AI is creating a wave of half-baked open source projects, spam PRs flooding maintainers, and security debt that nobody is counting.

Marc Brooker features ‘Spec Driven Development isn’t Waterfall’, making the case that specifications are living, iterative artefacts that the implementation flows from, not static up-front designs, and that the biggest velocity gains come from letting agents run autonomously with a clear map.

Vadym Kazulkin covers ‘Serverless Applications on AWS with Lambda Using Java 25, API Gateway and Aurora DSQL — Part 4 Using SnapStart with DSQL Request Priming’, diving deep into how Lambda SnapStart priming techniques with Aurora DSQL request priming can further reduce cold start times.

Danilo Poccia has a thought-provoking article titled ‘From Prompt Engineering to Agent Experience’, exploring how the shift from crafting single prompts to designing the environments in which agents operate is the new discipline.

Kenta Goto covers ‘AWS CDK Deployment Best Practices’, sharing four practical approaches for deploying with CDK, including static stack creation with Stage, synthesise once deploy many, and separating asset build from deploy phases.

Paul Stack features ‘AI: Why I’m All In’, a candid piece on why he’s an AI maximalist and why operations is about to get left behind again.

Davide De Sio covers ‘Building KiroGraph: a 100% Local Semantic Code Knowledge Graph for Kiro’, turning your codebase into a queryable semantic graph that dramatically reduces AI tool calls and token usage by up to 90%.

Paul Santus has a great article titled ‘AWS S3 Files Just Made Transfer Family SFTP Obsolete for Most Use Cases’, showing how S3 Files with ECS Fargate can replace Transfer Family at roughly 8x cheaper for low-to-medium traffic SFTP use cases.

Vinicius Senger covers ‘Why Quarkus + Vanilla JS + AWS Lambda + DynamoDB Is the Fastest Path from Idea to Production’, making the case that the gap between “works on my machine” and “running in production” can be almost zero with the right architectural choices.

🎓 Ask the Expert

Each issue, I ask a different AWS expert the same three questions to get their personal insights. This issue, we have AWS DevTools Hero Kenta Goto:

Opinions are the author’s and do not express the views of their employer.

1. What is one common mistake you see teams making when using open source in their solutions, and how can they avoid it?

One common mistake is assuming serverless is essentially free. In the early stages, costs seem negligible, but as workloads grow, teams are often surprised by how quickly the bill adds up.

The key is consistent, disciplined tuning. Take the time to measure and adjust memory allocation, timeout settings, and concurrency limits. Refactoring code to reduce cold start times and unnecessary execution duration can also cut costs significantly.

I'd also recommend embracing event-driven architectures and caching mechanisms to minimise redundant invocations and keep costs under control.

2. Which tool, package, or service are you most excited about right now, and why?

I'm most excited about Amazon Bedrock AgentCore. It's a fully managed platform for building, deploying, and operating AI agents at scale without managing infrastructure. Among its capabilities, AgentCore Runtime provides a serverless hosting environment for deploying and running AI agents.

What makes it even more interesting for me personally is that I contributed the hotswap support for AgentCore Runtime to AWS CDK. This lets you deploy agent changes in seconds instead of waiting for full CloudFormation deployments, dramatically improving the developer experience when building AI-powered serverless applications.

3. What is your favourite trick or tip that the readers may find interesting?

Always think stateless. Design every component as if it could disappear, break, or be recreated at any moment.

This mindset forces you to externalise state properly, make everything idempotent, and avoid relying on anything ephemeral. When something inevitably goes wrong, you'll be glad that you can simply tear down and rebuild without worrying about lost state or corrupted data.

This principle has saved me countless times in production and is the foundation of truly resilient serverless systems.

🧠 Tips & Tricks

This week’s tip is by Emanuele Bartolesi on LinkedIn:

An agent tone-of-voice configuration that tells your AI assistant to give direct, critical feedback, identify mistakes, weak assumptions, unnecessary complexity, and poor trade-offs without softening the message. (of course, this works with any agent).

No generic praise, no filler, no agreeing by default. When something is wrong, say exactly what is wrong, why it is a problem, and what should be done instead.

Here is the exact repo of instructions. I personally use very similar instructions with Kiro and Cursor.

Leave a comment

🚀 New Releases

Here are the latest and most interesting releases over the past two weeks in the AWS World:

⭐ My favourite release over the past two weeks is S3 Files - and here is a great video by Darko describing how they work!

• Amazon CloudWatch pipelines introduces new compliance and governance capabilities.

• Amazon Bedrock now supports cost allocation by IAM user and role.

• Amazon OpenSearch Serverless now supports Zstandard (zstd) codec for index compression.

• Amazon RDS Blue/Green Deployments now supports Amazon RDS Proxy.

• Amazon Bedrock Data Automation now supports custom vocabulary.

• AWS Agent Registry for centralized agent discovery and governance is now available in Preview.

• Amazon Location Service announces enhanced map styling with contour line density, traffic visualization, and 3D terrain.

• Amazon Bedrock AgentCore Browser adds OS-level interaction capabilities.

• Amazon Bedrock now offers Claude Mythos Preview (Gated Research Preview).

• AWS Lambda expands response streaming support to all commercial AWS Regions.

• Announcing Amazon S3 Files, making S3 buckets accessible as file systems.

• Amazon S3 starts rolling out new security best practice to new and existing buckets by default.

• Amazon Bedrock Guardrails announces general availability of cross-account safeguards.

• Amazon CloudWatch now supports OpenTelemetry metrics in public preview.

• Amazon CloudFront now supports SHA-256 for signed URLs and signed cookies.

• Amazon Bedrock AgentCore Evaluations is now generally available.

• AWS DevOps Agent is now generally available.

🔥 Tip: Check out https://aws-news.com/ for the very latest up-to-date serverless releases as they happen, created by the talented AWS Serverless Hero Luc van Donkersgoed.

✖️ Social of the Week

This week’s social is by Naval on Twitter/X:

I am down with this tweet 100%.. and personally finding building with AI tooling a similar feeling to building on Minecraft or Themepark (am I showing my age here? ha).

As it happens, I don’t vibe code; I am actually all in on spec-driven development, but the feeling of producing working software, scratching that itch on some open-source you have always wanted to build but didn’t have the time, or automating something that is a pebble in your shoe day to day, is wonderful.

What are your thoughts on this post? Leave a comment

Leave a comment

👷🏻 Tools & Frameworks

Check out the latest open-source frameworks, news, and tool updates from the past week.

⭐ My call-out tool this week is one of my own, my AWS educational platform Study From Experts!

For a long time, Mark Sailes and I noticed a frustrating gap in the market.

While there’s plenty of "introductory" AWS content out there, we struggled to find truly advanced, deep-dive mastery, the kind of knowledge you only get from years in the trenches. Content by true experts.

So, we decided to build it ourselves. We launched the exclusive home for AWS practitioners who want to move beyond the basics. This isn't just another learning platform; it’s a private library of specialised content you cannot find anywhere else.

Other recent releases:

• sample-automated-reasoning-formalization — An AWS sample project for automated reasoning formalisation, helping teams apply formal verification techniques to validate infrastructure and application logic.

• claude-coding-commands — A collection of useful coding commands for Claude, streamlining common development workflows and making agentic coding sessions more efficient.

• Kiro now supports MiniMax — Brian Beach highlights that Kiro has added MiniMax model support, expanding the range of models available in the IDE for developers who want more choice in their AI-assisted development workflows.

• iam-policy-autopilot — An AWS Labs tool that helps automate IAM policy generation and management, reducing the manual effort of crafting least-privilege policies and keeping your security posture tight.

• CDK Insights — Now GA! A tool for gaining visibility into your AWS CDK applications, helping you understand construct usage, dependency graphs, and deployment patterns across your infrastructure-as-code projects.

😂 Just for Fun

This week's "just for fun" post is by Stephen Telford on LinkedIn:

“A 34-year-old software analyst is out of a job this week after his employer's IT security team flagged an anomaly in his VPN login data that no internal policy had ever specifically anticipated: every single remote session for the past fourteen
months had been initiated from international waters.

Derek Callum, 34, had accepted a fully remote position in February of last year.

He had not mentioned that "fully remote" was a concept he intended to explore aggressively.

According to the termination report, the investigation began after a colleague noticed the Carnival Cruise Line logo on a travel mug visible during a video call.

HR cross-referenced the observation with Derek's IP address history.

The addresses corresponded to port stops across the Bahamas, Cozumel, and the western Caribbean basin.

The review found:

- 47 client calls completed from a balcony suite with documented ocean wind interference.

- 3 quarterly reports submitted from a vessel identified as the Carnival Sunrise.

14 instances of background applause attributed to a poolside steel drum band.

- One all-hands meeting during which Derek's camera angle successfully concealed a frozen drink with an umbrella in it.

His supervisor confirmed Derek had hit every deadline.

This did not factor into the termination decision.

Derek's travel mug declined to comment. Carnival Cruise Line said it looked forward to welcoming him back as a full-time guest.

He was employee of the month in October.“

If the guy hit all of his targets whilst getting a tan, why not? As somebody who has done a lot of international travel over the past 5 years, I have worked from swanky rooftop bars in Las Vegas to downtown coffee joints in Boston, and like this guy, I have always smashed all of my deadlines and targets! (Also, I take with a pinch of salt the credibility of the Fox News article!)

Leave a comment

🎙️ YouTube & Podcasts

Here are some of my favourite videos and podcasts this week covering serverless, AI, architecture, and software engineering.

⭐ My favourite video this week is by James Eastham discussing safe deployments with Lambda and the CDK - and I learnt a new feature of CodeDeploy called pre-traffic hooks!

• The Prime covers “Is Mythos too Dangerous?“ in this great video.

• Lenny’s Podcast covers “Head of Growth (Anthropic): “Claude is growing itself at this point”.

• The AWS Developers channel covers “Predicting Your Agent’s Cost“.

• Another great video from the AWS Developers channel with an ‘Intro to S3 Files‘.

• James Eastham has a great video titled ‘Stop Your Bad Lambda Deployments Before They Hit Production‘.

• Milan Jovanović discusses ‘Why Shopify Rejected Microservices (And What They Did Instead)‘.

• Derek Comartin has a great video titled ‘Why “Clean Architecture” is Killing Your Velocity‘.

• Matt Meckes covers ‘Building Agentic Applications with Spring AI‘.

Weekly Case Study 🔍

This week’s case study is by Just Eat on AWS:

Just Eat created its human resources application Jetty using Amazon Web Services and generative AI capabilities in Amazon Bedrock. The platform empowers employees with faster access to benefits and HR information, and the company has reduced ticket volumes by 70%. Elizabeth Eastaugh, Senior Director of Platform and Data Engineering, talks about how Just Eat has increased efficiency while maintaining a secure approach to data in the AWS Cloud, leveraging a stream-first approach to ensure data is both stored and used safely.

The key takeaway here is how Just Eat is using Amazon Bedrock to experiment quickly with text, voice, and images across the business, using the latest foundation models such as Amazon Nova. With Amazon Q, their engineers are building faster and more efficiently, and the platform scales across the 19 countries in which Just Eat operates.

🗣️ Inspirational Quotes and Thoughts

This week’s inspirational quote, law, or thought is by John J. Gumperz:

“If a program is useful, it will have to be changed.”

— Gumperz's Law (John J. Gumperz, 1922–2013)

Gumperz’s Law is a fascinating example of how the moment software becomes useful, the world around it shifts. Users discover new needs. Business requirements evolve. Edge cases emerge that nobody anticipated. The software that doesn’t adapt becomes the software that gets replaced.

This is why I’m such a strong advocate for evolutionary architecture and, specifically, hexagonal architecture. When you know that change is inevitable, the only sane response is to design for it from day one. Hexagonal architecture gives you exactly that: clean boundaries between your business logic and the outside world, so when the inevitable change arrives, whether it’s a new set of access patterns, a new API contract, a new event source, or a complete rethink of your delivery mechanism, you can swap out the adapters without gutting the core.

What are your own thoughts and experiences of this quote? Feel free to leave a comment below.

Leave a comment

🗳️ Poll of the Week

In the last poll, we asked the question, "Is AI development all just hype?"

Super interestingly, 100% of people voted no; they are having major successes! Im kind of suprised by this as there are still some sceptics out there on LinkedIn, but I guess the number says a lot about my core readers!

This issue, we ask the question, "With AWS services like App Runner and WorkMail now being deprecated, is this a worry for you when designing and architecting systems on AWS?"

Please feel free to leave a comment below on your answer.

Leave a comment

📅 Serverless Events

The following serverless events are upcoming, so mark your calendars.

• ACD Romania - 23rd April 2026.

• ACD Athens - 28th April 2026.

• ACD Türkiye - 9th May 2026.

• AWS Comsum Birmingham - 4th June 2026.

Would you happen to have any upcoming events that you would like to highlight? Message me below!

⭐ Spotlight

This issue’s spotlight falls on AWS Principal Solutions Architect, Heeki Park!

Heeki is a Principal Solutions Architect at Amazon Web Services and a prolific writer and builder in the agentic AI and serverless space.

His Medium articles are a fantastic resource for anyone working with agentic AI (Im a huge fan!), covering everything from agent platforms and spec-driven development to response streaming and credential management in agentic applications. Go check out his content at https://medium.com/@heeki, you won’t be disappointed.

Thank you for everything you do for our amazing AWS community Heeki!

Thank you for reading the latest Serverless Advocate Newsletter!

If you want to find out a little more about me, please have a look at:

https://www.serverlessadvocate.com/

See you next time,

Lee

Twitter | LinkedIn | YouTube | BlueSky | Website

Welcome

In the last issue, our serverless expert was AWS & AI Expert Paul Stack, and our spotlight fell on AWS Community Builder Jamie Walker!

This week, our open source expert is Ricardo Sueiras, our spotlight is on AWS Principal Developer Advocate Aaron Hunter, and we look at the latest AWS service releases, blog posts, hints and tips, news, and more!

This week’s newsletter is sponsored by Leighton.

Subscribe now

💡 A quick note from the author: I just wanted to let everyone know that going forward, the newsletters will be every two weeks, to allow me to curate an even better experience for you all - Lee.

📰 Articles that caught the eye

Here are some stand-out articles I read during the week in the World of Serverless, AI, engineering and architecture!

⭐ My favourite article over the past two weeks is by Luca Mezzalira on why agents don’t know what “good” looks like, and why that’s exactly the problem:

Ran Isenberg and Yaron Ben Ezra cover ‘Prevent Sensitive Data Leaks in Amazon CloudWatch Logs’, exploring how CloudWatch Logs Data Protection lets you detect and mask sensitive data before it’s written or forwarded, and how to enforce this centrally across every AWS account.

Allen Helton covered ‘Your AI agents are a security nightmare’ in this great article, discussing how maintenance is often forgotten about when we are building with AI.

Javi López features ‘The Real Bottleneck in Software’, arguing that the real constraint in software engineering is knowledge, not tooling, and why optimising for learning through small batches, TDD, and CI/CD is the path forward.

Jason Conway-Williams has a quick article titled ‘Handling Cross-Region SSM Parameters in AWS CDK’, walking through two practical approaches for sharing Parameter Store values across regions when CloudFront certificates need to live in us-east-1.

Joan Westenberg covers ‘”Collaboration” is Bullshit’, a provocative piece drawing on S.L.A. Marshall’s WWII research to argue that most collaboration tools simulate collective engagement without producing actual output, and that transparency got confused with progress.

Luca Mezzalira has a great article titled ‘Agents Don’t Know What Good Looks Like. And That’s Exactly the Problem’, reacting to the Neal Ford and Sam Newman fireside chat on agentic AI and software architecture, exploring the Dreyfus Model and why agents stuck at the novice level can’t distinguish a valid fix from deleting the test.

Agent Driven Development features ‘We Kissed Specs and PRDs Goodbye. Product Managers Pass POCs Now’, exploring how AI agents are collapsing the gap between product ideation and working software, with PMs building POCs directly instead of writing twelve-page documents.

Gunnar Grosch covers ‘Multi-Agent Systems on AWS Lambda with Durable Functions’, combining Strands Agents SDK with the durable execution SDK to build a purchasing coordinator where each specialist runs as its own Lambda function with checkpointing, parallel dispatch, and human-in-the-loop approval.

Ran Isenberg has a great article titled ‘Claude Code Best Practices: Lessons From Real Projects’, sharing practical lessons from shipping three real projects with Claude Code, covering BMAD vs plan mode, CLAUDE.md structure, and why domain expertise remains the bottleneck.

Jimmy Dahlqvist features ‘Authenticating to MongoDB Atlas with AWS Outbound Identity Federation’, a deep dive into using AWS IAM Outbound Identity Federation for machine-to-machine authentication, eliminating stored secrets by exchanging IAM roles for signed JWT tokens.

Darryl Ruggles covers ‘AWS Lambda Now Knows Where It Lives: Exploring the New AZ Metadata’, exploring the new metadata endpoint that exposes the Availability Zone ID of the execution environment with no VPC required and sub-millisecond response times.

Jeremy Daly has a thought-provoking article titled ‘The Convergence Problem: Rethinking the 2028 Global Intelligence Crisis’, arguing that AI isn’t just accelerating how we build software but compressing it toward sameness, and that the bottleneck has moved from building to deciding.

Kawsar K covers ‘AI Agent Steering File for AWS Serverless Applications’, sharing a practical steering file template for ensuring AI-generated applications follow secure, scalable, well-architected patterns on AWS.

🎓 Ask the Expert

Each week, I ask a different AWS expert the same three questions to get their personal insights, this week, we have Open Source Expert Ricardo Sueiras:

Opinions are the author’s and do not express the views of their employer.

1. What is one common mistake you see teams making when using open source in their solutions, and how can they avoid it?

This is such a great question, and one that, in my experience, changes depending on the background and experience of the developer.

A very frequent mistake is that for a lot of developers they think that open source is about being able to download software and use it for free. And whilst this might look like a characteristic of open source, it is only one part of it. The thing that makes something open source is when the creator of the “work” (in our case, source code) applies a license (specifically, an open source initiative, or OSI, license). These licences provide developers with “freedoms” on what they can do with the source code. You might hear of the four freedoms of open source: the freedom to study (view) source code, the copy the source code, to modify the source code, and to run the source code.

This leads to another common mistake that I see developers make, which is forgetting to add a license to their project, effectively making it proprietary. This effectively means that no one can really use your software in perhaps the way you or they want to. This is important because that license is what provides developers the ability to use software without the negotiation of the rights holder - in essence, it is this that allows everyone to build using and on top of open source. I would argue that this has been one of the key driving forces in software innovation over the past decade.

The final related mistake I see, although this is a lot less frequent these days as the licences being used by projects have changed over the years, is developers not understanding what obligations they are required to follow when they decide to use open source projects. For example, this might be that they need to generate an attribution document that outlines all the open source components they are using within their project and make this available via some mechanisms. You might have seen this if you go to your mobile phone and check the About or Legal sections of an app to see a list of open source projects that are being used.

I also want to add a generative AI footnote here. There are a LOT of mistakes being made about “open source” LLM’s. A lot of this started when META incorrectly stated that Llama was open source - it isn’t. Since then, I keep seeing both the term “open source” incorrectly used to refer to LLM’s. Remember, unless it has an OSI licence, it isn’t open source.

Also, please, please, it is NEVER “open-source” and ALWAYS “open source”!

2. Which tool, package, or service are you most excited about right now, and why?

This is a hard one as I regularly review really nice projects, from both large organisations and individual developers. That said, there are a few projects that for me are stand out and that I use on a regular (daily/weekly)

• Strands Agents - this project provides an SDK to developer agentic applications, and what I love about it is both its simplicity and its capabilities. Start simple to get the hang of things, but then keep diving in to add more features as you need them. I love the fact that it works across different model providers and has a growing community that is developing nice tools that you can use. Also, there is some really great content being created to help you really dive into how to create production-quality agentic applications. Top notch, go check it out.

• Finch - this is a drop in replacement for Docker Desktop, which I stopped using at the beginning on 2024 as an experiment. I wasn’t sure whether I would be able to stick with it as Docker Desktop is a super nice tool, and I love it. Fast forward to today, and I have been able to stick with Finch and have not had too many issues. The biggest one is that as I use ARM-based MacBooks, I have been caught out building containers that when deployed on AWS default to x86 CPU architectures, and Finch does not support the buildx command switch - but there are other ways to build multi-architecture images, so it’s not a biggie.

• Cedar - if you need to do authorisation within your application, then this is still one of the great “unsolved” problems that developers have to deal with. Amazon introduced Amazon Verified Permissions, allowing you to take a policy-based approach, like you might be familiar with when working with IAM policies within your AWS environments, and apply these to your application code. I spent a lot of time building content and demos using Cedar, and it is totally awesome. Last week I was blown away by Sondera, which has created extensions so you can use it within your Claude sessions to help define policies on what Claude can and cannot do.

• Zed- I switched to Zed last year, and it makes all other IDEs look pedestrian. I guess I have always been a terminal/CLI person, and whilst IDEs like Kiro and VSCode are awesome in their own way, Zed feels like a breath of fresh air and is my daily editor now.

3. What is your favourite trick or tip that the readers may find interesting?

So I guess the first tip for those of you who are Python developers is to make sure you check out “uv”. It is one of those tools that really does a lot of heavy lifting and makes working with Python SO much easier. I made the transition last year, and it’s now a staple in my agentic AI rules files.

The second tip is about finding cool tools to help you with your project or tasks. Some folk might know that for the past three or four years, I have been running an AWS open source newsletter (you can check out the full list here). I come across a lot of great tools that are created by the community, and I use many of these on a frequent basis. Thanks to the great AWS Community, they provided feedback about creating a repo that listed all the tools and projects that I have featured, and so I put together the newsletter-oss-projects as a result. If I am working on something that might need a new tool, I can normally find a tool to help me out (for example, last week I was building an app that used Amazon S3 to store assets, so I used stree, which works exactly like tree but for your S3 buckets). I have added hundreds of tools to this repo, so if you have not seen that, then check it out - I am sure you are going to find more than a few tools you will be wondering how you ever did without.

The final tip I want to share is about how you use AI coding tools to interact with open source projects. These tools are great, and they are getting better all the time. They have lowered the barrier to being able to contribute code back, and you might be tempted to do this - especially if there is a feature request or bug that you need adding/resolving. If you do, then please be mindful. Maintainers have never been under as much stress and pressure, and many have become quickly burdened with a large influx of new PRs to review. If you do decide to use AI coding agents, then make sure you spend the time to ensure that you read the contribution guidelines (many for example are holding off or pausing AI coding agent contributions), and take the time to check the code generated to make sure that not only does it work, but it follows the norms and expectations of the community (coding conventions, documentation, tests, etc). The bar should be higher from a review perspective on the contributor side, and you need to make sure that you sense-check the code contribution against what the project might be expecting. Avoid huge code dumps - tricky given that our agentic AI coding tools love nothing more than creating code! Finally, be kind to maintainers of the projects you use, as they have probably never been busier.

✅ Bonus tip: join the hashtag#believeinsls discord! There is a community there to answer any questions you may have without getting overzealous on serverless or without judgment! Check it out!…

🧠 Tips & Tricks

This week’s tip is by Stephen Siegert on LinkedIn:

“Kiro Tangent mode (/tangent): an experimental feature that creates conversation checkpoints, allowing you to explore side topics without disrupting your main conversation flow. Enter tangent mode, ask questions or explore ideas, then return to your original conversation exactly where you left off.”

A really handy feature when you want to go down a rabbit hole without losing your place!

Leave a comment

🚀 New Releases

Here are the latest and most interesting releases over the past two weeks in the AWS World:

⭐ My favourite releases over the past two weeks are the ones from the Lambda team, as they have been busy, plus the AWS Elemental MediaTailor updates, since I am building in the VoD area at the moment!

• Amazon CloudWatch now supports multi-account and region log centralization based on data source.

• AWS Elemental MediaTailor now available in Europe (London).

• Access Cluster Insights through the Amazon OpenSearch Service Console and Amazon EventBridge events.

• AWS Step Functions adds 28 new service integrations, including Amazon Bedrock AgentCore.

• AWS Lambda supports up to 32 GB of memory and 16 vCPUs for Lambda Managed Instances.

• AWS Lambda increases the file descriptor limit to 4,096 for functions running on Lambda Managed Instances.

• Aurora DSQL launches connector that simplifies building Ruby applications.

• AWS AppConfig adds enhanced targeting during feature flag rollout.

• Accelerate AI-assisted development with Agent Plugin for AWS Serverless.

• Amazon Bedrock AgentCore Runtime now supports managed session storage for persistent agent filesystem state (preview).

• Amazon Bedrock AgentCore adds support for Chrome policies and custom root CA.

• AWS MCP Server (Preview) now with enhanced monitoring and semantic search capability.

• Amazon Bedrock AgentCore Runtime adds WebRTC support for real-time bidirectional streaming.

• AWS Lambda now supports Availability Zone metadata.

• AWS Security Agent now supports downloading penetration testing reports.

🔥 Tip: Check out https://aws-news.com/ for the very latest up-to-date serverless releases as they happen, created by the talented AWS Serverless Hero Luc van Donkersgoed.

✖️ Social of the Week

This week’s social is by Prabhu G. on LinkedIn:

The anatomy of the .Kiro/ folder: a fantastic visual breakdown of how Kiro’s features come into existence, how the files are structured, and what purpose each one serves.

The key insight: global = who you are, project = what you’re building. If you’ve been experimenting with Kiro and things have gone over your head with steering, specs, hooks, powers, and agents, this is the visual guide you need!

What are your thoughts on this post? Leave a comment

Leave a comment

👷🏻 Tools & Frameworks

Check out the latest open-source frameworks, news, and tool updates from the past week.

⭐ My call-out tool this week is Figma’s MCP Server for GitHub Copilot. With Figma’s MCP server, you can now make changes directly to your Figma files using GitHub Copilot CLI or Visual Studio Code.

durable-viz - A tool by Gunnar Grosch that turns AWS Lambda durable function handlers into flowcharts using static analysis. Point it at a handler, get a diagram showing every step, parallel branch, invoke, callback, and conditional. Run it as a CLI or as a VS Code extension with a live diagram panel.

TypeORM v1.0 Beta - After more than 10 years, over 190 million downloads, and 4 million every week, TypeORM is finally approaching its v1.0 milestone. The beta is available now via npm install typeorm@beta.

CDK Contribution Skill - An Agent Skill from CDK Labs that combines multiple specialised subagents, AWS tooling, and Agent SOPs to streamline the entire AWS CDK contribution process from issue analysis to PR submission. Implements an orchestrator pattern with human approval gates.

Everything Claude Code - The agent harness performance optimisation system with 120K+ stars. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond. Now at v1.9.0 with selective install architecture and 12 language ecosystems.

ECR Scan Verifier - An AWS CDK Construct that blocks deployments to ECS, Lambda, and other services when ECR Image Scanning detects vulnerabilities, and optionally verifies container image signatures with Notation or Cosign.

Kiro MCP Manager v1.2.0 - New release of Ryan Cormack’s lightweight macOS menu bar app for managing Kiro CLI MCP server configuration and settings. Now with cleanup period settings, full settings sync with Kiro CLI docs, and knowledge pattern management.

delstack - A CLI tool for deleting AWS CloudFormation stacks, from routine deletions to force deleting stacks with undeletable resources. Unlike CloudFormation’s built-in FORCE_DELETE_STACK, which leaves failed resources behind, delstack actually cleans them up with no orphaned resources.

⚠️ Also, a note for NodeJS and TypeScript developers that axios@1.14.1 and axios@0.30.4 are compromised.

😂 Just for Fun

This week’s “just for fun” post is by Dr Milan Milanović on LinkedIn:

I think we are fine for now lol (check out the link to the video above)

Leave a comment

🎙️ YouTube & Podcasts

Here are some of my favourite videos and podcasts this week covering serverless, AI, architecture, and software engineering.

⭐ My favourite video this week is by the My 50 cents channel featuring Susanne Kaiser on ‘Architecture for Flow’!

• Debbie O’Brien covers ‘Build Better AI Agent Skills With Skill Creator v2 from Anthropic’.

• No Priors features Andrej Karpathy on ‘Code Agents, AutoResearch, and the Loopy Era of AI’.

• The PrimeTime covers ‘I Did Not Expect This Ending’.

• Derek Comartin at CodeOpinion features ‘Stop Using Event-Driven Architecture for Your CRUD Apps’.

• The My 50 cents channel covers ‘Architecture for Flow with Susanne Kaiser’.

• Logicata Podcast Season 5, Episode 12 covers ‘Buckets, Chips, and Legal Quips’.

• The Modern Software Engineering channel covers ‘Rethinking The Software Lifecycle: How To Look After Software For The Long Term’.

• Lenny’s Podcast features Jessica Fain on ‘The Art of Influence: The Single Most Important Skill Left That AI Can’t Replace’.

• Domain-Driven Design Europe features Younès Zeriahi on ‘Living Domain Model’ from DDD Europe 2025.

• Lex Clips features NVIDIA CEO Jensen Huang on ‘Will AI Replace Programmers?’.

Weekly Case Study 🔍

This week’s case study is by AssemblyAI on AWS:

AssemblyAI, a leading speech AI company, partnered with AWS to build a scalable platform for turning audio data into actionable insights.

Customers upload audio data to the AssemblyAI API or submit references via Amazon S3. The data is then transcoded and sent through an inference pipeline powered by an orchestrator that decides which models to call and in what order.

The pipeline leverages Amazon SQS for message queuing, Amazon ECS for running and scaling dozens of ML models, and Amazon S3 for storage. The platform supports features like speaker labelling and sentiment analysis, with models constantly being iterated on and deployed. Amazon ECS automatically scales the ML models up and down based on customer demand, enabling AssemblyAI to efficiently manage its growing model fleet.

🗣️ Inspirational Quotes and Thoughts

This week’s inspirational quote or thought is by Gordon Bell:

“The cheapest, fastest, and most reliable components are those that aren’t there.”

— Gordon Bell

It’s the ultimate architect’s/engineer’s mantra. Every line of code you don’t write is a line you don’t have to test, document, or fix at 3:00 AM. Every component you don’t deploy is one less thing to monitor, patch, and pay for. The best architects don’t add complexity; they remove it, and they make the case for why less is more.

This has never been more relevant than right now in the age of AI. When generating code costs almost nothing, the temptation to add features, abstractions, and “just in case” components has never been higher.

AI makes it trivially easy to create new features that nobody will ever use, new services that nobody asked for, and new layers of indirection that nobody needs. The discipline of restraint, knowing what not to build, is now the most valuable skill an architect can have. Just because you can generate it in thirty seconds doesn’t mean you should ship it.

What are your own thoughts and experiences of this quote? Feel free to leave a comment below.

Leave a comment

🗳️ Poll of the Week

In the last poll, we asked the question, “Have you now moved to spec-driven development in your AI-DLC?”

Interestingly, 75% of respondents have never tried the AI-DLC, or perhaps don’t know what it is, with the remaining 25% saying they love it! Im fairly suprised by the numbers, in all honesty.

This week, we ask the question “Is AI development all just hype?”

Please feel free to leave a comment below on your answer.

Leave a comment

📅 Serverless Events

The following serverless events are upcoming, so mark your calendars.

• ACD Romania - 23rd April 2026.

• ACD Athens - 28th April 2026.

• ACD Türkiye - 9th May 2026.

• AWS Comsum Birmingham - 4th June 2026.

Would you happen to have any upcoming events that you would like to highlight? Message me below!

⭐ Spotlight

This issue’s spotlight falls on AWS Principal Developer Advocate, Aaron Hunter!

Aaron is a Principal Developer Advocate at Amazon Web Services and a passionate advocate for the AWS developer community. He tweets/posts about AWS, learning, certification, and tech, and is known for his super-engaging video content and community presence! Go give him a follow on LinkedIn if you don’t already to stay on top of the latest AWS news and service releases!

He is a regular at AWS events, Conferences, User Groups, AWS live-streams, the AWS Developer Channel, and much more besides! Here is a great recent example on YouTube of his awesome content:

I also have a fond memory of spending time with Andres, Rob, Allen and Aaron out in Dallas, whilst out there for a work trip, where they took great care of us and showed us the local food joints out in McKinney! Aaron is one of the nicest guys in tech, for sure!

Thank you for everything you do for our amazing AWS community Aaron!

Thank you for reading the latest Serverless Advocate Newsletter!

If you want to find out a little more about me, please have a look at:

https://www.serverlessadvocate.com/

See you next time,

Lee

Twitter | LinkedIn | YouTube | BlueSky | Website

Welcome

In the last issue, our serverless expert was AWS Hero Kristine Armiyants, and our spotlight fell on AWS Senior Product Manager Michael Gasch!

This week, our serverless expert is AWS & AI Expert Paul Stack, our spotlight is on AWS Community Builder Jamie Walker, and we look at the latest AWS service releases, blog posts, hints and tips, news, and more!

This week’s newsletter is sponsored by Leighton.

Subscribe now

💡 A quick note from the author: I just wanted to let everyone know that going forward, the newsletters will be every two weeks, to allow me to curate an even better experience for you all - Lee.

📰 Articles that caught the eye

Here are some stand-out articles I read during the week in the World of Serverless, AI, engineering and architecture!

⭐ My favourite article this week is by Boris Tane on slop creep and the “enshittification” of software:

• Darryl Ruggles has a fantastic article titled “Powertools for AWS Lambda - Best Practices By Default”, a deep dive into how three decorators give you structured logging, distributed tracing, and CloudWatch metrics out of the box.

• Boris Tane covers ‘Slop Creep: The Great Enshittification of Software’, exploring how coding agents remove the natural circuit breaker that used to prevent bad architectural decisions from compounding unchecked.

• Ruth Kaseke features ‘AIDeas: Breaking the Silence on Crisis Support in Africa’, showcasing Safe Voice built on AWS.

• Ran Isenberg covers ‘AI Didn’t Wait for Security. Now What?‘ in this great article.

• Alex Moening has a great article titled ‘Skills vs MCP — Why Your AI Needs an Orchestration Layer’, explaining why tool access alone doesn’t equal tool intelligence and how skills sit above MCP to solve context rot.

• Heeki Park discusses “Collaborating with Agents: Teams in Claude Code”, looking at multi-agent orchestration patterns for development workflows.

• Faye Ellis has a great tutorial titled ‘Secure Serverless RAG in 5 Minutes with Amazon Bedrock + S3 Vector Store’, walking through Knowledge Bases, Guardrails, and PII filtering.

• Gunnar Grosch covers ‘Circuit Breakers on AWS Lambda: Why In-Memory State Silently Fails’, explaining why isolated execution environments mean your circuit breaker might never actually trip across the fleet.

• Vadym Kazulkin has another great Java article, this time ‘AWS Lambda Managed Instances with Java 25 and AWS SAM – Part 6 Lambda Function Performance Improvement Approaches’, taking cold start response latency from 1,579ms down to 60ms through static initialiser optimisations and DynamoDB priming.

• Gunnar Grosch features ‘Building Multi-Agent Systems with RISEN Prompts and Strands Agents’.

• Marco Frattallone covers ‘Standardising Construct Properties with AWS CDK Property Injection’, a feature that automatically applies organisational defaults to constructs without modifying existing code.

🎓 Ask the Expert

Each week, I ask a different AWS expert the same three questions to get their personal insights - this week we have AWS & AI Expert Paul Stack:

Opinions are the author’s and do not express the views of their employer.

1. What is one common mistake you see teams making when implementing serverless solutions, and how can they avoid it?

One mistake I see often is teams treating observability as something they will add later. That becomes a serious problem in serverless systems. A single request might trigger several Lambda functions, pass through queues, and start asynchronous workflows. Each step runs independently and writes its own logs. So when something breaks, you end up with scattered log entries and no easy way to reconstruct what actually happened.

The key is to build structured logging and correlation IDs into your systems from day one, treating observability as part of your SDLC. Use a shared logging utility so every function follows the same format and automatically propagates the correlation ID.

Tools like AWS Lambda Powertools make this straightforward. Setting it up early takes very little effort. Adding it later, once you already have dozens of functions in production, becomes a much bigger project.

2. Which serverless tool or service are you most excited about right now, and why?

Right now I’m most excited about AWS Lambda durable functions. They solve a problem that anyone building multi step workflows in serverless runs into. You have to either manage state yourself across functions, which gets complicated quickly, or you use AWS Step Functions and move a lot of your logic into a JSON state machine instead of normal application code.

Lambda durable functions allow you to take a different approach. You write your workflow as normal sequential code. The runtime handles checkpointing, state persistence, and replay behind the scenes and if the function pauses or fails, it resumes from the last checkpoint automatically.

What makes this particularly interesting right now is AI workflows. Many AI systems need to pause while waiting for something, a model response, a user approval, or a long running task. So this makes AWS Lambda durable functions a natural fit for agent style workflows that need orchestration but still want to live close to the application code. I think we are going to see a lot of experimentation there over the next 6 months or so.

3. What is your favourite trick or tip when working with serverless that the readers may find interesting?

One tip I’ve seen that works really well is using real production events as test fixtures instead of handcrafting synthetic ones. When people write tests for a Lambda function, they usually copy an example event from the AWS docs and fill it with realistic looking values. The problem is that real production traffic is almost always messier than what you expect. These real user requests will find all sorts of edge cases appear that you never considered.

The approach is to capture real events at the start of the handler and write them to an S3 bucket with a lifecycle rule that expires them after a set period. Over time you build a library of events your system has actually processed, including the strange ones. The one requirement is proper sanitisation as real events often contain PII or sensitive data, so you need a step that masks or removes those fields before storing them. Once that is in place, the fixture library becomes more valuable over time with almost no extra effort.

When a production bug happens because of an unexpected event shape, that exact event goes into the fixture library. Your test suite slowly becomes a record of the real inputs your system has seen.

✅ Bonus tip: join the hashtag#believeinsls discord! There is a community there to answer any questions you may have without getting overzealous on serverless or without judgment! Check it out!…

🧠 Tips & Tricks

I will write a full article on this soon, but I am having great success with using custom agents in Kiro alongside the GitHub MCP server in my development flow! My flow is like this below:

A great example was creating a custom agent in Kiro for Google Analytics Integration, which reviewed all of the code, and created a markdown file detailing the high, medium and low issues (bugs and features), as well as creating issues in GitHub for me automatically using the MCP integration.

I can then point Kiro at one or more of the issues in GitHub to create a spec for (design, requirements and tasks), to implement, and then subsequently have the same custom agent review the changes at the end to make sure everything was implemented correctly.

I then push the changes to GitHub in the form of a PR, and perform my own ‘human in the loop’ code review before merging.

I’m aware I could fully automate the loop, but in all honesty, I want to be able to prioritise the issues raised that I think are most important, sometimes group smaller changes, and sometimes they get closed as false positives, or where I have made an alternative design choice (where I wouldn’t want them implemented automatically).

What kind of flows are you using with Kiro currently that work well?

Leave a comment

🚀 New Releases

Here are the latest and most interesting releases over the past two weeks in the AWS World:

⭐ My favourite release over the past two weeks is the Amazon S3 regional namespaces!

• Accelerate serverless application development with new SAM Kiro power.

• AWS Lambda Managed Instances now supports Rust.

• New LZA MCP Server for AI-assisted configuration management.

• OpenSearch UI supports Cross Account Data Access to OpenSearch domains.

• AWS CDK Mixins is now generally available.

• Amazon S3 introduces account regional namespaces for general purpose buckets.

• Amazon Bedrock AgentCore Memory announces streaming notifications for long-term memory.

• AWS Glue zero-ETL integrations with Amazon DynamoDB as the source support new configurations.

• AWS Elastic Beanstalk launches Deployments tab with in-progress deployment logs.

• Amazon Bedrock AgentCore Runtime now supports stateful MCP server features.

• Amazon Bedrock now supports observability of First Token Latency and Quota Consumption.

• Amazon Route 53 Global Resolver is now generally available.

• Amazon CloudWatch Logs announces increased query concurrency and API limits.

• Amazon EventBridge Scheduler now provides a higher default quota for the CreateSchedule API.

• Amazon Redshift Serverless now maintains datashare permissions during restore.

• AWS Elastic Beanstalk now offers AI-powered environment analysis.

• Accelerate Lambda durable functions development with new Kiro power.

• Amazon Lightsail now offers OpenClaw, a private self-hosted AI assistant.

🔥 Tip: Check out https://aws-news.com/ for the very latest up-to-date serverless releases as they happen, created by the talented AWS Serverless Hero Luc van Donkersgoed.

✖️ Social of the Week

This week's social is by Eli Igra Serfaty on LinkedIn:

This is crazy…one AI engineer. No JavaScript experience. $200. 14 days. A fully functional Slack clone in production. Channels, threads, reactions, files, huddles, desktop, mobile. Everything.

Slack has around 2,500 employees, raised $1.4B in funding, and took 7 years to build what it is today. The takeaway? Code is no longer a moat. Your product is only defensible if it's hard to leave. Brand, distribution, community, network effects.

I would love to deep dive into the quality of this solution!

What are your thoughts on this post?

Leave a comment

👷🏻 Tools & Frameworks

Check out the latest open-source frameworks, news, and tool updates from the past week.

• CDK Insights (Beta) - AI-powered AWS CDK analysis and security tool, now in beta. Run a command, and it analyses your CDK stacks to find potential issues and suggest improvements. Check it out at cdkinsights.dev/beta.

• Safe Voice - An AIDeas project featured on the AWS Builder Center, breaking the silence on crisis support in Africa. Built on AWS services to provide accessible crisis support. Read more.

• AWS SAM Kiro Power - AWS SAM is now available as a Kiro power, bringing SAM workflows directly into the Kiro IDE. See the announcement on LinkedIn.

• Swamp - A new tool from System Initiative for managing AI-native automation pipelines. CLI-first interface for creating, validating, running, and managing workflows. github.com/systeminit/swamp.

• AWS Lambda Rust Runtime v1.1.1 - New release of the AWS Lambda Rust runtime with improvements and fixes. See the release.

• Kiro MCP Manager - A tool by Ryan Cormack for managing MCP server configurations in Kiro. github.com/ryancormack/kiro-mcp-manager.

• Terramate v0.17.0-rc1 - New release candidate for Terramate, the infrastructure orchestration tool. See the release.

• Agent Speaker - A tool for adding voice capabilities to AI agents. github.com/NetDevAutomate/Agent-Speaker.

• Peon Ping - Game character voice lines the instant your AI agent finishes or needs permission. Warcraft III Peon notifications for Claude Code, Codex, IDEs, and any AI agent. Stop babysitting your terminal. peonping.com.

• Build Applications with Lambda Durable Functions - Now available as a Kiro power at kiro.dev/powers.

• Wispr - A macOS menu bar app for local speech-to-text transcription built by Sébastien Stormacq. wispr.stormacq.com.

• Data API Client v2.1.4 - New release from Jeremy Daly. See the release.

😂 Just for Fun

This week's "just for fun" post is by Jaime Garcia on LinkedIn:

Why spend money on Claude Code when you can use Chipotle in your AI-DLC!

Sometimes the real productivity hack is burrito related. I love it.

Leave a comment

🎙️ YouTube & Podcasts

Here are some of my favourite videos and podcasts this week covering serverless, AI, architecture, and software engineering.

⭐ My favourite video this week is by the Confluent Developer channel comparing MCP to Skills!

• AWS Bites covers ‘LLM Inference with Bedrock’ in this great episode.

• Logicata Podcast Season 5, Episode 10 covers ‘CLI Updates, OpenAI Partnership, and Data Center Attacks’.

• The Prime asks the question ‘What is wrong with us?!‘.

• Barry Stahl covers ‘Solutions That Evolve: Building Self-Improving Systems with Genetic Algorithms‘ at NDC London.

• Mo Bitar covers a production incident and how ‘Amazon is regretting AI‘.

• The Confluent Developer channel asks the question, ‘Agent Skills or MCP in the era of Claude Code?‘.

• Serverless CrAIc Episode 82 covers ‘AI Is Changing Software Engineering — Why Your North Star Matters’.

• Lenny’s Podcast interviews Qasar Younis covering ‘The real AI revolution isn’t software. It’s farms, mines, and trucks‘.

• The My 50 cents channel interviews Vlad Khononov and covers ‘Why AI is the Third Coming of Domain-Driven Design‘.

• Maciej Jedrzejewski covers ‘Master Software Architecture: From Simplicity to Complexity‘ in this GOTO 2025 session.

• The Modern Software Engineering channel discusses ‘The Truth About Developer Productivity in the AI Age (IT’S A TRAP)‘.

• The AWS Developers channel covers ‘I Used an Agent to Order A Pizza’.

Weekly Case Study 🔍

This week’s case study is by Omnicom on AWS:

Omnicom, one of the world’s largest advertising and marketing services groups, partnered with AWS to revolutionise their global marketing operations at scale. By moving to AWS, Omnicom addressed three critical challenges.

First, managing data and analytics at scale using Amazon S3 Intelligent Tiering to manage over 90 petabytes of data, integrating more than 100 separate data sources and processing 400 billion daily events through Amazon Redshift.

Second, AI-powered campaign optimisation using Amazon SageMaker AI for real-time analytics and Amazon Bedrock for hyper-personalised messaging across omni-channel campaigns.

Third, redefining their creative process by fine-tuning campaign generation models on AWS, creating a comprehensive AI-powered platform serving over 75,000 employees across their global network.

🗣️ Inspirational Quotes and Thoughts

This week's inspirational quote or thought is by Neal Ford:

“Everything in software architecture is a trade-off. If an architect says they have a solution without a trade-off, they aren't an architect—they're a salesperson.“

- Neal Ford

It’s the first law of software architecture, and it holds true every single day. Every decision we make in our systems involves giving something up to gain something else. The moment someone presents a solution as having no downsides, that’s the moment to start asking harder questions. The best architects don’t hide trade-offs, they make them explicit and let the team decide which compromises are worth making.

What are your own thoughts and experiences of this quote? Feel free to leave a comment below.

Leave a comment

🗳️ Poll of the Week

In the last poll, we asked the question, “You’re starting a brand-new project on AWS today, what is your ‘Default’ data store?”

Interestingly, 43% said Amazon DynamoDB, 29% Amazon Aurora DSQL, and 29% stating Amazon RDS. In all honesty the results don’t surprise me, as most of my services now use DynamoDB or Aurora DSQL as standard!

This week, we ask the question: “Have you now moved to spec-driven development in your AI-DLC?”

Please feel free to leave a comment below on your answer.

Leave a comment

📅 Serverless Events

The following serverless events are upcoming, so mark your calendars.

• ACD Pune - 21st March 2026.

• ACD Romania - 23rd April 2026.

• ACD Athens - 28th April 2026.

• ACD Türkiye - 9th May 2026.

• AWS Comsum Birmingham - 4th June 2026.

Would you happen to have any upcoming events that you would like to highlight? Message me below!

⭐ Spotlight

This issue’s spotlight falls on AWS Community Builder Jamie Walker!

Jamie is a Principal AWS Consultant at Steamhaus and an active advocate in the AWS community as the organiser of the AWS User Group Manchester, AWS Community Buiklder, and a regular speaker at AWS events. He is also an organiser of ServerlessDays Manchester, helping to bring the serverless community together in the more Northern parts of England! (if you are from Newcastle like me, then Manchester is “down south”!)

Here is a great interview on YouTube that Jamie did with friend of the newsletter Mark Sailes whilst out at AWS re:Invent a few years ago! (never expected to see these two on a gondola ride together!)

Thank you for everything you do for our amazing AWS community Jamie!

Thank you for reading the latest Serverless Advocate Newsletter!

If you want to find out a little more about me, please have a look at:

https://www.serverlessadvocate.com/

See you next time,

Lee

Twitter | LinkedIn | YouTube | BlueSky | Website

Welcome

In last week’s issue, our serverless expert was AWS Community Builder Darya Petrashka, and our spotlight fell on AWS Hero Matt Lewis!

This week, our serverless expert is AWS Hero Kristine Armiyants, our spotlight falls on AWS Senior Product Manager Michael Gasch, and we look at the latest AWS service releases, blog posts, hints and tips, news and more!

This week’s newsletter is sponsored by Leighton.

Subscribe now

💡 A quick note from the author: I just wanted to let everyone know that going forward, the newsletters will be every two weeks, to allow me to curate an even better experience for you all - Lee.

📰 Articles that caught the eye

Here are some stand-out articles I read during the week in the World of Serverless, AI, Engineering and Architecture!

⭐ My favourite article this week is by Davide De Sio comparing Durable Functions to Step Functions:

• Matheus das Mercês has a fantastic article titled “AWS Lambda Durable Functions on Hexagonal Architecture: The Pattern You’ve Been Looking For”.

• Davide De Sio covers ‘AWS Lambda Durable Functions vs Step Functions: A Real-World Comparison’.

• Jeremy Daly has an epic article called ‘Context Engineering for Commercial Agent Systems’.

• The Register covers ‘From Agile to AI: Anniversary workshop says test-driven development ideal for AI coding’.

• Sai Donepudy has a great article titled ‘The Dark Factory Alternative: From Vibe Coding to Spec-Driven Development’.

• Pubudu Jayawardana discusses “Understanding Lambda Tenant Isolation“.

• Ran Isenberg asks the question ‘Claude Built My Wix Website in 3 Hours - Is SaaS Dead?‘.

• Aaron Sempf has a great article titled ‘From Orchestration to Authority‘.

• Kenta Goto has another great CDK article, this time ‘Why is cdk.out (Cloud Assembly) Necessary in AWS CDK?‘.

🎓 Ask the Expert

Each week, I ask a different AWS expert the same three questions to get their personal insights - this week we have AWS Hero Kristine Armiyants:

Opinions are the author’s and do not express the views of their employer.

1. What is one common mistake you see teams making when implementing serverless solutions, and how can they avoid it?

This may be an extreme one but I had a case when a customer spent several thousands of dollars in a month on CloudWatch. The thing was that the developer built a serverless flow that was using CloudWatch and used the request ID as a dimension!

In CloudWatch, every unique combination of dimensions creates a brand new custom metric, which cost $0.30 a month for the first 10.000, and plus the API requests and log ingestion, every single run of that function created a new billable metric.

How to avoid it: Never put the request IDs as a dimension. Neither put them in plaintext logs. Consider using CloudWatch Embedded Metric Format (EMF) which is best for generating cost-effective custom metrics from serverless apps and containers too.

2. Which serverless tool or service are you most excited about right now, and why?

From numerous interesting announcements of re:Invent 2025, I’m really excited about S3 Vectors becoming generally available.

Excited to use it as a vector storage engine for Bedrock knowledge base. Love its simplicity & cost effectiveness.

First there were tables and now vectors: S3 is definitely going in an interesting direction!

3. What is your favourite trick or tip when working with serverless that the readers may find interesting?

I would suggest smart usage of wonderful Step Functions. With standard usage of them you are paying for the number of state transitions and that can easily be quite expensive.

Reserve standard workflows only for business-critical workflows, and my favorite case is like payment processing. The real resilience and all the audit trail that Step functions are giving you for the payments case completely justifies the cost.

✅ Bonus tip: join the hashtag#believeinsls discord! There is a community there to answer any questions you may have without getting overzealous on serverless or without judgment! Check it out!…

🧠 Tips & Tricks

This week’s tip is by AWS in their article ‘6,000 AWS accounts, three people, one platform: Lessons learned‘:

Lots of great tips in this article on how to implement a large-scale account-per-tenant model on AWS, and how that model shifts complexity from service code to platform operations.

Leave a comment

🚀 New Releases

Here are the latest and most interesting releases this week in the AWS World:

⭐ My favourite release this week is the AWS Observability Power for Kiro, which I am going to try out this weekend!

• Amazon Lightsail expands blueprint selection with a new WordPress blueprint.

• Amazon Bedrock batch inference now supports the Converse API format.

• Amazon CloudWatch logs centralization rules now support customizable destination log group structure.

• Amazon Bedrock announces OpenAI-compatible Projects API.

• Amazon SNS now supports push notifications in the Europe (Spain) Region.

• Amazon Cognito enhances client secret management with secret rotation and custom secrets.

• AWS Lambda Durable Execution SDK for Java now available in Developer Preview.

• Amazon Location Service introduces LLM Context as a Kiro power and Claude Code plugin to improve AI performance.

• Aurora DSQL launches new integrations for Visual Studio Code SQLTools and DBeaver.

• AWS launches a playground for interactive Aurora DSQL database exploration.

• AWS Observability now available as a Kiro power.

🔥 Tip: Check out https://aws-news.com/ for the very latest up-to-date serverless releases as they happen, created by the talented AWS Serverless Hero Luc van Donkersgoed.

✖️ Social of the Week

This week's social is by Kelsey Hightower on LinkedIn:

Kelsey always has a way of cutting through the noise and getting to the heart of the matter.

What are your thoughts on this post? What are your own experiences?

Leave a comment

👷🏻 Tools & Frameworks

Check out the latest open-source frameworks, news, and tool updates from the past week.

• Bref - Bref 3 is released.

• taskmd - Markdown-based task management designed for both humans and AI coding assistants. Store your tasks as .md files with YAML frontmatter, and use the CLI or web interface to manage, visualise, and track your work. Includes a Claude Code plugin for slash command integration.

• PR-Agent - The original open-source AI-powered code review agent. It handles PR reviews, improvements, and descriptions using a single LLM call per tool. Platform agnostic across GitHub, GitLab, BitBucket, and Azure DevOps, and supports multiple AI models.

• AWS MCP Server Deployment Agent SOPs - AWS has launched deployment Standard Operating Procedures (SOPs) available in the AWS MCP Server. These automated procedures let you deploy web applications to your AWS account using natural language prompts from any MCP-compatible IDE, including Kiro, Cursor, and Claude Code. It generates CDK infrastructure, deploys CloudFormation stacks, and creates CI/CD pipelines with recommended security best practices.

• mmdflux - Render diagrams as Unicode text, ASCII text, SVG, or JSON. Supports Mermaid syntax and is built for diagram-as-code pipelines with fast rendering, terminal-friendly output, linting, and machine-readable graph data for tooling and agents.

• PR Lens - AI-powered GitHub PR code reviewer for teams. It is codebase-aware, injecting co-change history, directory siblings, and paired test files into every review so the AI understands context beyond the diff. Supports Anthropic Claude and OpenAI GPT-4o, and stores team review history in a shared GitHub Gist with zero infra.

• aws-lambda-env-modeler - v3 is out! AWS-Lambda-Env-Modeler is a Python library designed to simplify the process of managing and validating environment variables in your AWS Lambda functions.

😂 Just for Fun

This week’s “just for fun” post is by Abdirahman Jama on LinkedIn:

Geez, this is the kind of ‘Dad Joke’ I usually make!

Leave a comment

🎙️ YouTube & Podcasts

Here are some of my favourite videos and podcasts this week covering serverless, AI, architecture, and software engineering.

⭐ My favourite video this week was a video I missed 3 weeks ago from James Eastham on Immutable Patterns of System Design!

• Adrian Cockcroft asks the question ‘AI Agents Will Do What Cloud Did to Ops - Are You Ready?‘.

• The Modern Software Engineering channel asks the question ‘Testing in Production” Actually the Safest Way to Ship?‘.

• Aleksander Stensby covers ‘10 Tips To Level Up Your AI-Assisted Coding‘.

• James Eastham covers ‘Immutable Patterns of System Design; From Monolith to Agentic AI’ at NDC London 2026.

• Lennys Podcast chats with the ‘Cisco President on the AI revolution‘.

• Thomas Coopman at DDD Europe 2025 covers ‘Scaling Teams with Ownership‘ in this short video.

• Mitchell Hashimoto’s “new way of writing code” on the Pragmatic Engineer.

• The Prime and The Standup covers “AI Personal Assistants are ruining people lives”.

• The vBrownBag talks through “AI Agents Made Simple: Everything You Need to Know“.

Weekly Case Study 🔍

This week’s case study is by Mercado Libre on AWS:

Mercado Libre, Latin America’s largest e-commerce platform, partnered with AWS Partner Mutt Data to develop GenAds, an AI-powered tool that empowers small and long-tail sellers to quickly go to market with visually impactful product listings.

Leveraging Stability AI’s Stable Diffusion model in Amazon Bedrock, the tool creates contextualised images that place products in visually appealing backgrounds, refining details like reflections and shadows for professional-grade quality. The system uses Anthropic’s Claude 3 Sonnet to generate prompts, stores assets in Amazon S3, and indexes them in Amazon DynamoDB.

Weekly, category-based backgrounds are created and daily selection processes identify popular items for banner generation. Only the most visually engaging and technically precise images make the final cut, adhering to strict advertising parameters. This automated approach reduces the burden on sellers and accelerates the production of impactful ads in the competitive e-commerce space.

🗣️ Inspirational Quotes and Thoughts

This week’s inspirational quote or thought is by Derek J. de Solla Price (Price’s Law):

“The square root of the number of people in a domain do 50% of the work.“

- Derek J. de Solla Price

The law essentially means that in a company of 100, 10 people do half of the major/important work.

In software, these "vital few" are usually the architects or senior devs who understand the entire codebase. If one of these people leaves (or gets "burned out"), the project doesn't just slow down by 10%; it can effectively grind to a halt because the remaining 90% of the staff lacks the tribal knowledge to navigate the core logic.

What are your own thoughts and experiences of this quote? Feel free to leave a comment below.

Leave a comment

🗳️ Poll of the Week

In last week’s poll, we asked the question, “How much do you typically pay for online courses?”

Interestingly, 50% said $10-$50, and the other 50% said $100-$200. Personally, I am happy to pay anywhere up to $300 for a great online course.

This week, we ask the question: “You’re starting a brand-new project on AWS today, what is your ‘Default’ data store?”

Please feel free to leave a comment below on your answer.

Leave a comment

📅 Serverless Events

The following serverless events are upcoming, so mark your calendars.

• ACD Romania - 23rd April 2026.

• ACD Chennai - 7th March 2026.

• ACD Slovakia - 11th March 2026.

• ACD Pune - 21st March 2026.

• ACD Romania - 23rd April 2026.

• ACD Athens - 28th April 2026.

• ACD Türkiye - 9th May 2026.

• AWS Comsum Birmingham - 4th June 2026.

Would you happen to have any upcoming events that you would like to highlight? Message me below!

⭐ Spotlight

This week’s spotlight falls on AWS Senior Product Manager Michael Gasch!

Michael is a Senior Product Manager at AWS within the Serverless organisation, where he works on services like Amazon EventBridge and AWS Lambda.

He is obviously deeply passionate about all things distributed, streaming, and event-driven systems (and has answered many of my questions over the years, sending me articles and information to always back up suggestions - so a personal thank you from me, Michael!).

Many of you will recognise Michael from his fantastic deep dive session at re:Invent 2025 alongside Eric Johnson, where they unveiled AWS Lambda Durable Functions to the world. Check it out below:

His work is always helping shape the future of serverless orchestration and event-driven architectures on AWS, and he is also an all round great guy, and I have been lucky to meet him a few times.

Thank you for everything you do for our amazing AWS community, Michael!

Thank you for reading the latest Serverless Advocate Newsletter!

If you want to find out a little more about me, please have a look at:

https://www.serverlessadvocate.com/

See you next time,

Lee

Twitter | LinkedIn | YouTube | BlueSky | Website

Welcome

In last week’s issue, our serverless expert was AWS Community Builder Damien Jones, and our spotlight fell on AWS Principal Cloud Technologist Morgan Willis!

This week, our serverless expert is AWS Community Builder Darya Petrashka, our spotlight falls on AWS Hero Matt Lewis, and we look at the latest AWS service releases, blog posts, hints and tips, news and more!

This week’s newsletter is sponsored by Leighton.

Subscribe now

📰 Articles that caught the eye

Here are some stand-out articles I read during the week in the World of Serverless, AI, Engineering and Architecture!

⭐ My favourite article this week is by Boris Tane, and how the AI-DLC now differs massively from how we have done things for years!

• Jacob Moxham covers ‘How to scale your impact across multiple teams as an individual contributor‘.

• Boris Tane has a fantastic article titled “The Software Development Lifecycle Is Dead“.

• Alexander Abramovich covers ‘Product Ownership in Practice: Beyond Processes and Backlogs‘.

• Luca Bianchi has a cool article called ‘Vibe Coding Got Me a POC in Days. Production Broke Every Assumption I Had‘.

• Ricardo Sueiras discusses ‘Using aws-vault to manage access to your AWS resources from Kiro CLI‘.

• Morgan Willis covers ‘From POC to Production-Ready: What Changed in My AI Agent Architecture‘.

• Massimo Re Ferrè has a great article titled ‘Adding Kiro support to Backlog.md using Backlog.md‘.

• Rehan van der Merwe covers ‘Scaling ECS Fargate like Lambda‘.

• Darryl Ruggles discusses ‘A Serverless Recipe Assistant with Bedrock AgentCore, Knowledge Bases, and S3 Vectors‘.

🎓 Ask the Expert

Each week, I ask a different AWS expert the same three questions to get their personal insights - this week, we have AWS Community Builder Darya Petrashka:

Opinions are the author’s and do not express the views of their employer.

1. What is one common mistake you see teams making when implementing serverless solutions, and how can they avoid it?

I still see most issues stemming from poor planning and miscommunication. Often, teams (or individual members) fail to properly agree on event schemas, API contracts, or input/output formats. A common scenario is when a team member makes a change and updates their side of the integration, but forgets to notify others, leading to broken workflows.

To avoid this, plan as early as possible and keep the team consolidated with a clear division of responsibilities. Implement transparent documentation that serves as a single source of truth. Additionally, utilising Lambda versions and aliases is a best practice to ensure that updates don't inadvertently disrupt existing production environments. It may sound idealistic, but aiming for high standards in communication is essential for serverless success.

2. Which serverless tool or service are you most excited about right now, and why?

I am still very excited about AWS Lambda and how its roadmap is evolving. For instance, using Lambda to implement action groups for Amazon Bedrock agents was a brilliant move to give agents greater flexibility and a richer "toolbox".

I also love the recent introduction of AWS Lambda durable functions. They make agentic orchestration and multi-step AI workflows much simpler by allowing you to write sequential code that automatically checkpoints progress. The ability to suspend execution for up to one year (without paying for idle compute during wait periods) is amazing. I definitely see massive potential for using them in ML training pipelines (yes, a "classical" ML is still a thing!)

3. What is your favourite trick or tip when working with serverless that the readers may find interesting?

Since I work mostly with Python, a frequent challenge is bringing external packages into the Lambda execution environment. Because Lambda runtimes are paired with specific Amazon Linux distributions (like AL2 or AL2023), libraries packaged on Windows or macOS often face an architecture mismatch.

For example, the Pillow library. It contains a compiled C extension that must match the specific OS of the Lambda runtime. If you simply run pip install locally and zip it, the binary likely won't be compatible. Since there is no AWS-managed Pillow layer, you have to build it yourself using an environment that matches Amazon Linux, typically via Docker or an EC2 instance.

My trick is to use Lambda itself to produce the Lambda-compatible library. By running a build script inside a Lambda function that matches your target runtime, you ensure the resulting library is perfectly compatible with the environment. Once the build is complete, the resulting .zip archive can be stored in an Amazon S3 bucket and used to create a Lambda layer. It will be available across all your functions.

✅ Bonus tip: join the hashtag#believeinsls discord! There is a community there to answer any questions you may have without getting overzealous on serverless or without judgment! Check it out!…

🧠 Tips & Tricks

This week’s tip is by Brian Beach on LinkedIn:

It’s another week, and another release from the Kiro team! I have an article coming in the next few days about how to use Agent Skills and Kiro together (specifically the Playwright Skill), but in the meantime, check out this awesome list of skills you can use https://github.com/sickn33/antigravity-awesome-skills

Leave a comment

🚀 New Releases

Here are the latest and most interesting releases this week in the AWS World:

⭐ My favourite release this week is the Aurora DSQL simplified IAM authentication through a set of libraries!

• Amazon SNS now supports sending SMS in additional AWS Regions.

• Aurora DSQL launches new Go, Python, and Node.js connectors that simplify IAM authentication.

• Amazon Aurora DSQL now integrates with Kiro powers and AI agent skills.

• Kiro is now available in AWS GovCloud (US) Regions.

• Amazon OpenSearch Service now supports storage optimized i7i instances.

• Amazon MSK now supports dual-stack (IPv4 and IPv6) connectivity for existing clusters.

• Amazon Bedrock reinforcement fine-tuning adds support for open-weight models with OpenAI-compatible APIs.

• Amazon Aurora now supports Server-Side Encryption by default.

• Claude Sonnet 4.6 now available in Amazon Bedrock.

• AWS Glue 5.1 is now available in 18 additional regions.

🔥 Tip: Check out https://aws-news.com/ for the very latest up-to-date serverless releases as they happen, created by the talented AWS Serverless Hero Luc van Donkersgoed.

✖️ Social of the Week

This week’s social is by AWS Community Builder Benjamen Pyle on LinkedIn:

What are your thoughts on this post? What are your own experiences?

I personally think Ben hits the nail on the head - we are now becoming architects, product specialists, and orchestrators - and guess what… it is very fun!

Leave a comment

👷🏻 Tools & Frameworks

Check out the latest open-source frameworks, news, and tool updates from the past week.

• Teleprompter - A browser-based one, with data fully local to your browser.

• acp-ui - A modern, cross-platform desktop client for the Agent Client Protocol (ACP). Connect to AI coding agents like GitHub Copilot, Claude Code, Gemini CLI, and Qwen Code from a unified interface.

• parallel-dev - Run multiple AI coding agents on the same codebase without merge conflicts.

• zen - A worktree orchestrator for AI-assisted PR reviews and feature work with Claude Code.

• cdk-terrain - The Open Construct Foundation (OCF) today announces the first community release of CDK Terrain (cdktn), the open, community-governed continuation of the Cloud Development Kit for Terraform (CDKTF).

• all-agents-mcp - An MCP server that orchestrates multiple AI CLI agents — Claude Code, Codex, Gemini CLI, and Copilot CLI — through a unified interface. Delegate tasks, run cross-agent comparisons, and leverage each agent's strengths from any MCP-compatible host.

• antigravity-awesome-skills - The Ultimate Collection of 883+ Universal Agentic Skills for AI Coding Assistants — Claude Code, Gemini CLI, Codex CLI, Kiro CLI, Antigravity IDE, GitHub Copilot, Cursor, OpenCode, AdaL.

😂 Just for Fun

This week’s “just for fun” post is by Sanchit Narula on LinkedIn:

WOW, how many of the websites I am using day to day are doing this cheeky change without me knowing? lol…

Leave a comment

🎙️ YouTube & Podcasts

Here are some of my favourite videos and podcasts this week covering serverless, AI, architecture, and software engineering.

⭐ My favourite video this week was on Lenny’s Podcast interviewing Boris Cherny. So many insights in this video!

• Rehan Van Der Merwe covers ‘Scaling ECS Fargate like Lambda‘.

• The My 50 cents channel interviews Jacqui Read on ‘Architecture is BROKEN (It’s not the code but the communication)‘.

• The Modern Software Engineering channel discusses ‘How to Make the Best of AI Programming Assistants‘.

• Lex Fridman and Peter Steinberger discuss the ‘Best programming language for coding with AI agents‘.

• The Logicata channel season 5 episode 7: ‘Serverless Streams, Sovereign Clouds, and Surprising Costs’.

• The vBrownBag podcast discusses ‘Tech Employment in the AI Era‘.

• Lenny’s Podcast has guest Boris Cherny covering ‘What happens after coding is solved‘.

• Derek Comartin covers ‘Read Replicas Are NOT CQRS (Stop Confusing This)‘.

• AWSLondonON Meetup has Darryl Ruggles covering ‘Lambda Durable Functions vs Step Functions When Each Wins‘.

Weekly Case Study 🔍

This week’s case study is by WRITER on AWS:

WRITER is the enterprise platform where AI agents automate work, with security, compliance, and control built in. Designed for scale, WRITER provides an end-to-end environment for building, activating, and supervising AI agents that automate workflows and improve decisions. WRITER executed a comprehensive "AWS-first" strategy starting in 2023, ultimately training their breakthrough Palmyra X5 model on SageMaker HyperPod at 1/10th the typical cost. Now fully migrated to AWS, WRITER delivers 333% ROI for customers while maintaining strict security and compliance standards.

🗣️ Inspirational Quotes and Thoughts

This week’s inspirational quote or thought is by Adyen Principal Engineer Heitor Lessa:

“One of the heartbreaking moments I had when Serverless took off was watching individuals being let go from positions they didn’t want to change — restarting web servers, NOC positions, and many others that were necessary then but the environment and the pace made them less necessary or in that case redundant.

Seeing a similar pattern with AI now that is producing consistent and acceptable results in the enterprise, I can’t help but think the same will happen — except this time is not narrowed to Ops, DBAs, and DC roles anymore.

There’s so much we could achieve if we have an ounce of curiosity, keep our skepticism to help us validate assumptions and influencer’s BS, and remember all this exists to generate value for customers - be internal or external.

Hearing “it’s not technically possible, it’s too difficult, we need XYZ HC to to this” is increasingly become harder to defend without the inevitable - albeit silly at times - question:

If AI agent teams and typical Enterprise guardrails were present right now, what prevents us to deliver more for customers?

I’m starting to lose sympathy and this is no good”

- Heitor Lessa