What I heard about agents, organizational design, partner workflows, and the work that still requires humans

I attended Catalyst 2026 hoping that AI would be a major topic. What I didn't anticipate was that it would be the central focus.

Over two days, AI appeared in nearly every discussion of the partnership lifecycle. Sessions covered partner research, recruitment, onboarding, co-marketing, co-selling, marketplaces, enablement, service delivery, measurement, and organizational design. The discussions also touched on agents responding to partner questions, preparing QBRs, coordinating workflows, influencing vendor shortlists, and altering product discovery.

The amount of AI-generated content was considerable. The sophistication of the conversation was even more important.

The conference wasn't mainly about using AI to write emails faster or summarize meetings. The most significant sessions raised a more important question: what impact will AI have when it becomes part of the operating model for partnerships?

That's the main point I took away. What was formerly a set of tools used by the partnership teams is now becoming the operating system for how partnership work is carried out.

The conversation has moved past the copilot

In recent years the way that AI has entered into partnership situations has been through personal productivity—for example, by drafting an email, summarizing a call, producing an initial version of a joint value proposition, and researching a company before a meeting.

I still use those as practical examples. Yet they don't change the fundamental work. The same person carries out the process, transfers information between the same disconnected systems, and bears the same coordination effort. AI only helps complete the individual steps faster.

Catalyst demonstrated that the conversation could go beyond that model.

The approach has shifted from asking "How can AI help me with this task?" to "How should this process function now that people and agents work together?"

That is a more difficult question because it involves roles, workflows, permissions, data, measurement, and accountability. It offers greater opportunity than just saving a few minutes drafting.

AI as the operating system

Nelson Wang articulated this idea most clearly in his sessions on “10 Principles for AI in Partnerships” and the AI-native partner leader’s operating system.

He drew a distinction between partnership teams which use AI occasionally for low-value tasks and those which regard AI as the program's operating system, and his "90/10 flip" provides a practical means of understanding the aim.

Most partnership professionals spend too much time on operational tasks. They focus on securing deal registrations, reconstructing materials, updating trackers, finding relevant context, preparing reports, and coordinating work with teams they do not manage. The activities that generate real value—building trust, designing joint strategies, resolving ambiguity, and working with partners—get only the leftover time.

The opportunity is to reverse that ratio.

An AI-native operating model can prepare partner research, keep account context updated, turn meetings into coordinated follow-ups, draft onboarding plans, identify missing information, assemble executive updates, and monitor workflows for exceptions. Each task sounds modest alone. Together they represent a large share of the invisible work consuming partnership teams.

This is not about making one person equivalent to an entire organization. It is about redesigning the system so people spend less time moving information between processes and more time applying judgment where it matters.

AI is joining the team

Christopher Smith’s session was titled “Your AI Challenge Isn’t a Tech Problem. It’s an Org Design One.” Its central premise was straightforward: AI is not replacing the partner team. It is joining it.

That framing stuck with me because it demands a different set of decisions.

If agents are joining the team, what work do they own? What context can they access? Which actions require approval? When should they escalate? Who is accountable when an agent makes the wrong recommendation or fails to recognize an exception?

These are organizational questions before they are technology questions.

A partner-research agent can assemble information, but a person still has to decide whether a relationship is strategically meaningful. An onboarding agent can coordinate tasks, but a person still has to recognize when a partner is losing confidence. An AI system can prepare a QBR, but it cannot own the relationship or make an executive commitment.

The practical opportunity is to design the work around that distinction. AI can carry repeatable coordination, synthesis, and monitoring. Partnership professionals carry judgment, trust, negotiation, and accountability.

Without that design work, companies will just add AI tools to fragmented processes. Teams may produce more material, but underlying operating problems will remain.

From individual agents to an ecosystem operating layer

Dan O’Leary and Sugata Sanyal offered concrete examples of what the broader model can look like.

Dan’s sessions described an AI-native partner operating layer handling work such as onboarding, QBRs, agreement generation, and executive reporting. His chief-of-staff agent extended the idea by coordinating inbound and outbound work across people, other agents, applications, and data.

Sugata described an “Ecosystem OS” that turns workflows and team knowledge into agents capable of working across the partner lifecycle.

The interesting part is not any single agent. It is the compounding effect of connecting them to the work.

A successful onboarding motion can become a reusable workflow instead of being rebuilt for each partner. Decisions and context can become part of operating knowledge instead of disappearing into call notes. Reporting can become an output of the work rather than a separate monthly exercise. The system can identify where human attention is needed instead of requiring a person to inspect every step.

Every interaction can improve the system for the next interaction.

That is a meaningful change from the traditional partner technology stack, where information lives across CRM, PRM, email, spreadsheets, call recordings, and individual memories. Adding a chat interface to that fragmentation does not solve much. An operating layer that understands context and coordinates work across it might.

Old playbooks are under pressure

Cindy Zu’s session, “Building Partnerships at AI Speed (with No Playbook),” added another dimension to this shift.

Cindy ran through five hot takes. One of them stayed with me: the first zero-to-one partnerships hire at a technology company will fail.

I suspect many in the Catalyst community have experienced this. A company hires one person to define strategy, identify partners, create the operating model, and build internal alignment, but often the only clear objective is to deliver revenue. Maturity arrives long after results are due, yet the person is measured as if it already existed.

AI creates an opportunity to change that narrative by giving early partnership teams operating support from the start. Research, documentation, coordination, and reporting no longer have to begin from scratch every time.

But Cindy’s broader point was not just about giving the first hire more tools. AI-native companies operate in ecosystems where model capabilities, unit economics, buyer behavior, and partner roles change at the same time. The person who arrives with the most established playbook may not be best prepared for that environment.

The role increasingly rewards people who can test hypotheses, build lightweight systems, and make decisions with incomplete information. Experience still matters, but its purpose changes. The value is not arriving with a fixed map but knowing how to build the map while the ground moves.

That applies well beyond the first hire. Entire partnership organizations will need to become more experimental without becoming less disciplined.

AI is changing the ecosystem outside the partner team

The agenda also made clear that AI is changing more than internal operations.

Several sessions explored what happens when AI influences how buyers discover, evaluate, and shortlist companies. Partner marketers discussed visibility in AI-generated answers instead of traditional search rankings. Marketplace leaders discussed making listings agent-discoverable and products agent-callable. Other sessions questioned whether the best partner portal might be one partners never have to visit because information and workflows meet them inside the systems they already use.

This matters because partnership teams are not only adopting AI. They are partnering in markets being reorganized by AI.

Co-marketing changes when content can be personalized and localized at scale. Co-selling changes when agents help determine which vendors make the shortlist. Co-building changes when agents and MCP-based architectures become part of the product. Co-serving changes when AI participates in enablement, implementation, and customer support.

The traditional partnership playbook assumed that people at companies would discover one another, exchange information, build integrations, and coordinate execution. That model is not disappearing, but agents are orchestrating in more of those steps.

Partnership strategy now has to account for both human relationships and machine-mediated discovery and execution.

AI will not repair a weak partnership foundation

A tempting conclusion emerges: connect enough agents, automate the workflows, and the partnership program will work.

I am skeptical of that.

AI cannot create a compelling reason for another company to partner. It cannot manufacture executive sponsorship, resolve conflicting incentives, or make a product ready for an ecosystem. It cannot fix bad channel data without someone defining good data. It cannot improve partner activation if the company has never agreed on what activation is.

Several Catalyst sessions provided this counterweight. AI may be ready before the data is. Automating an undefined process only scales confusion. Buying tools cannot substitute for organizational design.

Companies also risk using AI to raise expectations without changing priorities. Giving partnership professionals agents and then adding more work to the queue is not transformation. It is the same operating model running faster.

The real question is whether AI helps partnership teams spend less time compensating for missing infrastructure and more time doing the work that creates mutual value.

What partnership leaders should take from Catalyst

The pattern across the conference was clear. The next stage of AI adoption will not be defined by how many tools a partnership team buys. It will be defined by which workflows the team redesigns.

The starting point is not a broad mandate to “use AI.” It is identifying where operational drag prevents people from doing higher-value work. Pick one workflow with a clear outcome. Define what the human owns, what an agent can own, what data the workflow requires, and where approval or escalation belongs. Then measure whether the redesign improves business results, not just whether it produces more activity.

This approach is less exciting than installing a new tool. It is also much closer to the work described in the strongest Catalyst sessions.

The technology is becoming capable enough to participate in the operation. The difficult part is deciding how the operation should work.

The human work becomes more valuable

My biggest takeaway from Catalyst was not that AI will automate partnerships.

It was that AI will automate more of the heavy lifting required to create successful partnerships.

As research, content, coordination, and routine execution become abundant, the scarce resources become trust, access, judgment, and the ability to align organizations around a shared outcome. Those have always been partnership capabilities.

AI can give partnership professionals more time and better context for that work. It can help organizations turn what they learn into repeatable systems. It can make partner experiences more responsive and let smaller teams operate more consistently.

But it does not remove the need to partner. If anything, it raises the value of people who know how.

Catalyst 2026 made the direction increasingly clear. AI is joining the partner team, entering the partner ecosystem, and becoming part of the operating model. The companies that benefit will not be the ones that automate the most tasks. They will be the ones that make the clearest decisions about what people and agents should accomplish together.

Where is operational drag keeping your partnership team from doing its most valuable work?

Every few days, a new headline declares that AI is either coming for your job or that the fear was overblown all along. Both camps sound absolutely certain. The data underlying the story is a lot messier. A new study tracking 22,000 companies found that the firms investing most heavily in AI are hiring faster than everyone else, not slower, and that includes entry-level roles. At the same time, other researchers are warning that the specific tasks junior employees used to cut their teeth on are quietly disappearing. Those two things can both be true, and figuring out how they fit together matters more than picking a side.

This week I pulled three pieces that refuse to give you a clean answer about AI and work. One challenges the layoff narrative with real numbers. One shows a company using AI to make its people more valuable instead of fewer. And one makes the uncomfortable case that we are breaking the bottom rung of the career ladder without a plan to rebuild it. Here are three things that caught my attention this week, along with three work agents worth comparing if you want to see what is actually reshaping the job.

1. Is AI causing layoffs? A new report says it’s complicated

A study from Ramp and Revelio Labs tracked 22,000 US firms and found the opposite of what the layoff headlines suggest. Companies investing most heavily in AI are hiring faster than their peers, with high-intensity adopters posting a 10.2% jump in headcount and 12% growth in entry-level hiring over two years. Tech leaders including Nvidia’s Jensen Huang and OpenAI’s Sam Altman have started pushing back on the idea that AI is behind recent job cuts, calling it a lazy narrative and a convenient scapegoat for decisions companies were making anyway.

The catch is in the fine print. Those hiring gains are concentrated in larger, VC-backed, faster-growing firms in the information sector, and the study notes AI adoption still has not reliably translated into higher productivity or profits. So the headline finding may say more about which companies can afford to bet big on AI than about AI creating jobs on its own. The most useful line for an individual is the study’s own advice: pick the firms that are actually using AI well, because that is where the growth is showing up.

2. A.I. Doesn’t Have to Mean Layoffs

This piece takes aim at the executive reflex that treats AI success as a headcount reduction. Economist Erik Brynjolfsson argues the bigger prize comes from using AI to raise what each worker can do, not from cutting the roster. Schneider Electric is the example, deploying AI to automate repetitive and tedious tasks across its nearly 160,000 employees so people can spend their time on work that actually needs a human.

The important word here is choice. Augmentation is not the default outcome of buying AI licenses; it is a decision that requires redesigning how the work happens. That is harder and slower than a layoff, and it doesn’t yet seem to show up on quarterly reports, which is exactly why so many leaders default to the cut. The companies that get real value are the ones rebuilding their processes around the tools, not just handing employees a chatbot and hoping productivity appears.

3. It’s time to address the looming crisis in entry-level work

Here is the counterweight to the first story. AI is increasingly doing the routine junior tasks that early-career workers have always used to build experience, and that is landing on top of an already soft market for recent graduates. The old advice to learn to code looks shaky now that AI handles much of the routine coding, and the piece argues we are quietly dismantling the training system the whole economy depends on.

Both this and the Ramp study can be right at once, which is the part worth sitting with. Firm-level headcount can climb while the specific rungs people used to climb disappear. If entry-level roles are how skills actually get built, automating them away is a bit like eating our seed corn. The author’s fixes point in sensible directions, with schools adapting, governments nudging early-career hiring, and students getting genuinely AI-fluent, but nobody has yet answered the hard question of who pays to train people once the training tasks are gone.

3 Things I Think You’ll Like: AI Work Harnesses

OpenAI launched ChatGPT Work this month, so the battle for the business-user workspace is heating up. Claude Cowork has led the business-user workspace for a while, so ChatGPT Work has some catching up to do with Anthropic framework. That makes this a real comparison rather than a one-horse race. Here are three different bets on where it goes: the fast-improving challenger, the incumbent, and an open-source option you run yourself.

1. ChatGPT Work

This is OpenAI’s push to move ChatGPT from a chat window into a work agent, and it is the tool that impressed me most this month. Powered by GPT-5.6, it pulls context from your tools and files, plans an approach, and takes action across your apps to produce finished spreadsheets, docs, and slides. It connects to more than 1,400 plugins, runs recurring scheduled tasks you can check from your phone, and includes a Plan mode that waits for your approval before it starts. What stands out is how closely this now mirrors the business-user framework Claude built first. I ran out of tokens on Claude this week and gave Work a try. I don’t know if I’ll go back. The app is good, and the new 5.6 models are really good.

2. Claude Cowork

I know I have used Cowork multiple times in this segment. Cowork is the tool that defined this category for business users, and it is still the one I reach for first in the morning. You hand Claude a task, and it works across your files, calendar, email, messaging, and the web until the job is done. It now runs on web and mobile with background sessions that keep going after you close the laptop, plus an approval gate so nothing sends until you review it. I’m still using Cowork daily, including to help pull this newsletter together, and that approval step is what makes me comfortable handing it real work.

3. Microsoft 365 Copilot

Microsoft continues to make efforts, but in my experience, it remains one of the most unintuitive frameworks for developing desktop agents. They tend to over-engineer solutions, but for companies tied to the Microsoft ecosystem—many of which are—this is still worth exploring. Their approach to models is inconsistent: initially focusing solely on OpenAI, then collaborating with Anthropic, and now engaging with Chinese open-source models like DeepSeek. They are even developing their own models. It seems they are trying to lower internal costs associated with providing these agent solutions. However, this often appears to compromise the customer and user experience, though they keep trying.

Bottom line: Claude Cowork has led the business-user workspace and remains my default, but ChatGPT Work has closed the gap enough that the choice now comes down to your ecosystem and how much control you want. If you’ve been a die-hard ChatGPT user, then it’s time to try Work. Keep working with Microsoft solutions if, you know, you’re tied to that ecosystem and don’t have options.

How is your organization using AI? Same work with fewer people, or more output with more people?

I’ve been working out some definitions lately, because the words we use for this technology have started to blur together. Generative AI, agentic AI, AI agents, agentic systems. People reach for them interchangeably, and they don’t mean the same thing. I spent a couple of hours this week arguing with Perplexity, and here’s how I’ve come to sort them out.

We started with generative AI. That’s the part most of us experienced first with ChatGPT. You give it a prompt, and it produces text - a paragraph, an image, a block of code, a summary. The output is information. It’s useful, but it sits there waiting for you. It generates, then it stops.

Then we moved on to agentic AI, where I struggled and had to change my thinking. I used to hear “agentic” and picture autonomous robots taking everyone’s jobs. That’s not what the term actually meant. The real shift was tools gaining agency, the ability to do things rather than just produce text. An agentic AI tool can take an action, connect to another system, and do work on your behalf. The shift from producing text to taking action is what agentic AI was really about.

Once you have that foundation, you can build the thing people actually mean when they say AI Agents. These really are autonomous software systems that pursue a goal through several steps with minimal supervision. They plan, act, check the result, and adjust, all while using tools and connecting systems. We’re building these now, whereas 12 months ago most of what was called an agent was vaporware. Today it’s real software doing real work on your behalf.

Finally, a distinction worth working out is the difference between desktop and cloud agents. A desktop agent runs on your machine. This started with Claude Code, quickly became Claude Desktop, and now includes Codex and other desktop tools. It can see your local files, your open apps, and even use your computer, which makes it personal and immediate. A cloud agent runs on remote infrastructure. It’s always on, it scales, and it keeps working when your laptop is closed, but it lives a step removed from your day-to-day. In my mind, this is where we move from agents working for individuals to agents serving teams and companies. This is the most interesting transition point to me at the moment.

Here are three things that caught my attention this week, along with three tools I think you’ll like.

3 Things for Tuesday, June 30, 2026:

1. Agentic AI vs. Generative AI

IBM offers a clean outside check on the vocabulary. Generative AI creates content from a prompt. Agentic AI describes systems that decide and act toward a goal with limited supervision, in a loop of perceive, reason, act, and learn. The key point is that agentic AI is the broader framework, and AI agents are the components that do the work within it. IBM is honest that the proven use cases still sit on the generative side while the agentic ones are mostly early. The framework is real; the proof is still being written.

2. AI Agents Plunged the Tech World Into Chaos

Steven Levy traces how we got from generative output to working agents in about a year, through Claude Code and the open-source OpenClaw, which lets you run a personal agent from WhatsApp or iMessage and hit 366,000 GitHub stars by May. It’s the build moment, and a clear look at why agentic systems are still unsettled. The same tool that automates your work can leak your data and burn through tokens at a six- or seven-figure clip. The capability is here. How to run it safely is not.

3. All Model Labs Are Now Agent Labs

This is the evidence that the whole field made the move. The major labs have shifted from shipping standalone models to shipping agents, wrapped in harnesses, memory, and workflows, and then selling the package. We didn’t just get better at generating text. The industry rebuilt itself around tools that take action. The caution is that “agent” is now a label stuck on anything with a chat box, so look for the harness underneath. That’s the difference between a real agent and a generative model in a costume.

3 Things I Think You’ll Like: Three Ways to Run an Agent

Three ways to actually run one, from managed and polished to open and self-owned. All three run autonomous agents. What differs is where the agent lives and how much you manage yourself.

1. Claude Cowork

The desktop agent I use most. It runs on your machine, sees your local files and open apps, and acts on the work in front of you. Hand it a multi-step task and let it run while you do something else. The catch is it’s tied to your desktop, so it isn’t built for jobs that run around the clock without you.

2. Claude Managed Agents

The cloud-hosted side of the same idea, aimed at building something to ship. You define the tasks, tools, and guardrails, and Anthropic runs it on their infrastructure with a harness that handles tool calls and error recovery. Sessions run for hours and survive dropped connections. It’s in public beta at standard token rates plus eight cents per session-hour. I haven’t built on it, so treat this as a pointer.

3. Hermes

The open-source option for owning the whole stack. From Nous Research, it’s a native MIT-licensed app for Mac, Windows, and Linux, with one agent and one memory that reaches you across Telegram, Discord, Slack, WhatsApp, email, and the command line. It keeps persistent memory, generates its own skills, and runs on 300-plus models so you aren’t locked to one provider. I’ve been experimenting, and like the pace of change; there’s a lot of work going into it. If Cowork is too limited for you, try this.

Bottom line: Cowork for a polished desktop agent with no infrastructure to manage. Managed Agents to ship something that runs unattended in the cloud. Hermes to own the stack and experiment.

When you picture putting an agent to work, do you reach for one on your desktop or one in the cloud, and what’s driving that choice?

The AI hype cycle seems to have hit a wall over the past few weeks.

For the last 18 months, foundation model labs have handed power users 10x or 20x more value than they were actually paying for. That subsidy is ending. Uber burned through its entire 2026 AI budget in four months. One unnamed enterprise customer ran up a $500 million Claude bill in a single month because nobody set spending limits. Anthropic, OpenAI, GitHub, and Google are all quietly shifting from subscription pricing to usage-based billing, and the era of letting power users consume tokens without consequences is closing fast.

What I’m watching right now is how companies respond to this new reality. The smart ones will treat AI like any other expensive resource, with budgets, controls, and clear accountability for ROI. The less-smart ones will overcorrect, cutting headcount and management layers on the assumption that AI will fill the gap. Both responses will play out in real time, and the gap between them is the most interesting space to watch over the next 18 months.

Here are three things that caught my attention this week, along with three AI assistants worth knowing if you spend time on CarPlay.

3 Things for Monday, June 1, 2026:

1. The AI Token Shortage Begins

Nathaniel Whittemore at The AI Daily Brief argues that May 2026 marked the end of the AI subsidy era and the beginning of what he calls the token shortage era. Foundation model revenue has moved from per-seat subscriptions to per-token consumption, and with OpenAI reaching $30 billion in ARR and Anthropic hitting $47 billion in annualized run rate (up from just $3 billion a year ago), they still have huge revenue shortfalls to make up to justify the infrastructure buildout. Once tokens became the economic unit, power users started generating costs that flat subscription pricing was never built to absorb. GitHub Copilot, Anthropic, and Google have all rolled out usage-based billing in response. Cursor’s new Composer 2.5 model and DeepSeek’s permanent 75% price cut on V4 reflect the market response to compute scarcity, and Elon Musk has pivoted from a Grok cheerleader to a compute provider, with SpaceX renting capacity on Colossus 1 to Anthropic.

The podcast is worth listening to because Whittemore clearly frames the second-order effects. We’re moving from an era in which companies could let employees experiment freely on subsidized plans to one in which every prompt has a measurable cost. Some of that is healthy. The capability overhang between what models can do and what most companies actually get out of them is widening, and a real cost signal might finally force discipline on AI adoption. But the people who burned $500 million accidentally are not going to be replaced overnight by people who can build durable AI ROI frameworks. That gap is where the next 18 months of consulting, services, and infrastructure play will get built. The question worth asking right now is which side of that gap your organization sits on, and whether the tooling and discipline are in place to land on the right one.

2. Claude Spending Tops $500M When One Client Forgot Usage Limits

An unnamed enterprise customer racked up a $500 million Claude bill in a single month after rolling Anthropic’s API out to thousands of employees without spending caps, usage limits, or monitoring. The story broke late last month and is the most extreme example of a pattern that has become common in the last six months. Microsoft, Uber, and Amazon have all publicly disclosed AI cost issues. Most companies that adopted Claude or ChatGPT at scale did so under a flat seat pricing model and never built the cost governance muscle that legacy SaaS budgets require. When the underlying pricing shifted to token consumption, those gaps showed up immediately in the form of runaway monthly invoices.

Treating AI as a checkbox procurement decision is going to keep producing stories like this one. The technology bills more like AWS than like Salesforce, but most procurement teams are still buying it like Salesforce. The fix is not complicated, though it is uncomfortable. Hard spending caps at the workspace and user level. Role-based access controls so that not every employee has the same API authority. Usage dashboards that finance teams actually monitor. And clear accountability for which business outcomes the spend is supposed to generate. Companies that put those guardrails in place now will avoid the worst headlines, and they’ll also be in a stronger position to negotiate when their enterprise contracts come up for renewal. The cost story is not someone else’s problem, and the organizations that put the right people at the table this quarter will spend the second half of 2026 in a much better position than the ones that wait.

3. Everyone wants to kill the middle manager role. The data says don’t do it

Fast Company published a useful counterweight to the Great Flattening trend, the pattern of companies cutting middle managers on the assumption that AI can absorb their work. Korn Ferry data shows 41% of employees say their company trimmed management layers in the last year. Gartner predicts that through 2026, 20% of organizations will use AI to eliminate more than half of their middle management positions. The author, an HR veteran and employment lawyer, clearly lays out the appeal of the move. Management layers are expensive, bad managers exist, and AI is genuinely capable of handling the busywork of taking notes, drafting goals, and flagging underperformers in a dashboard.

The argument against the trend is that the data does not support it. Gallup research finds that managers account for at least 70% of the variance in employee engagement across business units. Engagement in turn correlates with profitability, productivity, turnover, theft, and safety. When a CEO asks whether managers are still needed in an AI-augmented world, they are effectively asking whether they can cut 70% of the variance in their results and trust that the remainder lands on the right side. The honest answer is that AI can absorb much of the administrative burden managers have been carrying, which is a real opportunity. But replacing managers with dashboards confuses the visible work of management with the actual job. Coaching, judgment, and accountability drive engagement, and none of those scale with an LLM.

3 Things I Think You’ll Like: AI Assistants on CarPlay

iOS 26.4 opened CarPlay to third-party AI chatbots in early 2026, and I’ve been testing the major options on my drives around Denver and the Front Range. ChatGPT, Perplexity, Claude, Gemini, and Grok all shipped support in the first wave. One shared limitation is worth flagging up front. None can be invoked by voice command. Siri still owns the wake word, so you have to tap the app open before talking, which is the biggest reason in-car AI experiences vary so much from one app to the next.

1. Perplexity

Perplexity has been my go-to AI on CarPlay since iOS 26.4 shipped. It pairs LLM responses with cited web sources, which is useful when I want to confirm something quickly between meetings without taking my eyes off the road. The voice mode handles long-form questions and follow-ups without losing context, and the answers feel built for the in-car listening experience rather than for staring at a screen. Same Siri limitation as the others, you still have to tap to open before talking, but once you’re in the app, the flow is the smoothest I’ve used, and the voice response doesn’t lag. If you’re going to install one of these on CarPlay, start here.

2. ChatGPT

OpenAI’s flagship arrived on CarPlay alongside the other big chatbots when iOS 26.4 shipped. The general capability is strong, with image, document, and web tools available outside the car, and voice mode handles natural conversation well in most contexts. My experience on CarPlay specifically has been the weakest of the three I’ve tested. The tap-to-open flow breaks the rhythm of in-car use, and the voice mode struggled more than I expected with the ambient noise from highway driving. If you already live inside the OpenAI ecosystem for the rest of your work, the continuity is still valuable, but it would not be my first pick if I were choosing for CarPlay alone.

3. Grok (xAI)

Grok shipped CarPlay support alongside ChatGPT and Perplexity, and 9to5Mac confirmed it as one of the three working CarPlay AI apps in their May coverage. The in-car UI lists recent conversations for quick resume, which is a nice touch when you’re picking up a thread from earlier in the day. The conversational tone is less filtered than the others, and the real-time X data can be useful if you’re tracking breaking news while driving. Grok is not my bag, so I will not make strong claims about CarPlay-specific quality, but if you already use Grok for the X integration, the in-car version is a natural extension.

Bottom line: Perplexity is my pick for CarPlay use. ChatGPT is fine if you live in the OpenAI ecosystem. Grok makes sense if you’re plugged into the X data feed for work.

Are you using an AI assistant in the car yet, and if so, what’s the use case that’s actually worked for you?

The articles I cover this week don’t have a clean through-line, and I think that’s actually appropriate. We’re in a moment where different parts of the AI conversation are moving in very different directions at the same time.

Block published a detailed essay arguing that AI can replace the coordination functions of corporate hierarchy entirely. Accenture and Wharton published a report arguing something closer to the opposite: as AI agents become more capable, human accountability becomes more consequential, not less. And Anthropic quietly disclosed that its new Claude Mythos Preview model can autonomously identify and exploit zero-day vulnerabilities in major operating systems and browsers.

That last one deserves more attention than it’s probably getting. A model that can find and exploit previously unknown security flaws is a meaningful capability threshold, and Anthropic’s response, launching a defensive program called Project Glasswing and urging the industry to shorten patch cycles, suggests they know it.

The Block and Accenture pieces can coexist, by the way. AI can take over coordination and information routing in organizations while also demanding that humans take greater ownership of outcomes. The question of who is responsible when an AI system makes a consequential mistake doesn’t get easier just because the AI is doing more of the work.

Here are three things that caught my attention this week, along with three tools I’ve been experimenting with.

1. From Hierarchy to Intelligence

Block (layer-off of 1000s, owner of Square, Cash App, et al.) published a detailed essay arguing that the traditional corporate hierarchy, a structure that has evolved over 2,000 years from the Roman Army to the modern org chart, is fundamentally limited by human “span of control.” The piece proposes replacing those coordination layers with AI: a system built around a “Company World Model” that continuously tracks internal operations and a “Customer World Model” derived from transaction data across Cash App and Square. In this structure, managers whose primary function was routing information between teams are replaced by AI-driven orchestration, while human roles are redefined as deep specialists, problem owners, and people developers who combine building with coaching.

The core question here is whether this is a genuine organizational model or a well-written piece of AI positioning. Block’s framing is more developed than most AI transformation essays, but it’s still written by a company with an obvious interest in positioning itself as an AI-first organization. The actual test will be whether Block publishes measurable results over time rather than letting this become another aspirational document. That said, the specific mechanisms they describe, world models, DRI-based accountability, and player-coaches replacing traditional managers, are concrete enough to take seriously. Worth watching whether other companies start borrowing this language and whether Block ever reports back on what actually worked.

2. ‘Intelligence may be scalable, but accountability is not.’

A joint report from Accenture and the Wharton School, titled “The Age of Co-Intelligence,” lands as a useful counterweight to the more utopian visions of AI-run organizations. Its central argument: as AI agents become smarter and more capable, they don’t reduce the demands on human leadership. They increase them. The report estimates that more than 50% of American working hours will be reshaped by AI, and that the productivity gains AI generates typically show up as “capacity freed” rather than as automatic output growth. That freed capacity has to be deliberately redirected to higher-value work, or it simply disappears. The report also flags a growing “governance gap” as AI agents spread faster than formal oversight structures can form around them.

The accountability argument is the most important thread here, and it cuts against a common assumption in AI adoption conversations. Deploying AI agents doesn’t transfer accountability away from leaders. It makes leadership more consequential because errors can now multiply at the speed and scale of whatever agent is doing the work. The report’s suggestion of a “chief agentic resources officer” title is worth considering, but the underlying point is sound: before the governance gap widens further, most organizations would benefit from clarifying who is actually responsible when an AI agent makes a consequential mistake.

3. Assessing Claude Mythos Preview’s cybersecurity capabilities

Anthropic published a detailed capability assessment of Claude Mythos Preview, a new model it has not released publicly, that autonomously identifies and exploits zero-day vulnerabilities in major operating systems, web browsers, and open-source codebases. The model significantly surpasses previous Claude versions in exploit development, can reverse-engineer closed-source software, and can chain multiple known bugs into sophisticated attacks. Anthropic describes the moment as a “watershed” in AI cybersecurity capability and has launched Project Glasswing, a program using Mythos Preview for defensive purposes, to help the industry prepare. Notably, these capabilities emerged from general model improvements rather than from security-specific training.

The timing and framing of Anthropic’s disclosure are worth noting. Publishing a detailed capability assessment for a model they’re not releasing is an unusual move, and the recommendations they offer, shorten patch cycles, automate incident response, and re-evaluate defense-in-depth strategies, read more like an industry-wide warning than a product announcement. The honest implication is that a model capable of autonomously finding and exploiting major vulnerabilities will eventually be available to actors with very different motivations from Anthropic’s. The defensive window between that capability existing inside a controlled lab and existing in the open is probably shorter than most organizations’ current security roadmaps assume. In fact, Anthropic’s 90-countdown will start soon.

3 Things I Think You’ll Like: Random Bits

No single theme tied these tools together this week. They’re just three things I’ve been using to build, experiment, and launch things, and each one solved a specific problem I ran into. Sometimes that’s enough.

1. Paperclip

Paperclip is an open-source AI agent orchestration platform built around the concept of a “zero-human company,” an organization where AI agents run every function, from development to marketing to quality assurance. You manage the agents like a team: org chart, goals, tasks, budgets, and role templates. The mental model shifts from “prompting an AI” to “running a company with AI employees.” I’ve been deep into this one all week, using it to organize and push forward a stalled micro SaaS project. I set up an organization, which in turn built a launch plan, created and launched a website, and began implementing a GTM plan. It’s a meaningfully different experience from any automation tool I’ve used before, and if you’ve been looking for a way to augment a solo pursuit or small team with agents, it’s worth serious attention in my book.

2. AgentMail

AgentMail is an API-first email platform designed for AI agents to participate in two-way email conversations. Most email APIs are designed for one-way notifications: your app sends, humans reply. AgentMail flips that model by giving your AI agents the ability to read, understand, and respond within an email thread, as a human participant would. If you’re building workflows where an agent needs to handle inbound inquiries, coordinate scheduling, or follow up on multi-step processes, this addresses a real infrastructure gap. I’ve just given my Paperclip agents access and they’ve started drafting outreach messages.

3. Hermes Agent

Hermes Agent is an open-source, MIT-licensed AI agent framework from Nous Research that runs as an autonomous server process and, notably, learns and improves over time rather than starting fresh with each session. This is another agent framework that I’m leveraging with Paperclip, with the intention of offloading some AI workload onto cheaper and possibly local LLMs.

Which of these capabilities, autonomous agent orchestration, AI-handled email, or self-hosted agent frameworks, seems closest to something you’d actually put to work in the next month?

I’ve been spending a lot of time lately in Claude’s world, testing Anthropic’s tools and writing about their approach to enterprise AI. So this week I decided to shift the lens and look at what’s happening across the street at OpenAI. The picture is... interesting.

What I’m seeing is a company that went all-in on the consumer side of AI and is now scrambling to figure out what comes next. They launched Sora, built a web browser, expanded ChatGPT into a commerce platform, and are now planning to bundle everything into a single desktop app. Meanwhile, Anthropic has been quietly scaling revenue by going after business customers with tools like Claude Code and Cowork, and OpenAI’s investors are starting to notice.

The three stories below paint a picture of a company at a crossroads. OpenAI is cutting projects, consolidating products, and making big bets on consumer engagement, all while the enterprise market they largely overlooked keeps growing around them. Whether this is a strategic pivot or a sign of deeper problems depends on who you ask.

Here are three things that caught my attention this week, along with three AI voice assistants worth exploring if you’re ready to talk to your tools instead of typing.

1. Panicked OpenAI Execs Cutting Projects as Walls Close In

OpenAI is pulling back on side projects and refocusing resources on coding and enterprise users. Despite recent launches (and cancellations) like Sora and the Atlas browser, the company is burning through billions and has scaled back its AI infrastructure spending plans from $1.4 trillion to $600 billion through 2030. Fidji Simo, OpenAI’s CEO of Applications, declared a “code red” internally (another one?), telling staff the company “cannot miss this moment because we are distracted by side quests.”

The timing of this shift is hard to ignore. Anthropic’s Claude has been gaining real traction in the enterprise AI market, and OpenAI’s scattered product strategy has left investors questioning where returns will come from. Cutting projects is never a sign of strength, but it can be a sign of clarity. The question is whether OpenAI can execute a focused enterprise strategy fast enough to compete with companies that have been building for that market from the start.

2. OpenAI is planning a desktop ‘superapp’

OpenAI plans to merge ChatGPT, Codex, and Atlas into a single desktop application. The move is part of the broader consolidation effort led by Fidji Simo and OpenAI President Greg Brockman, aimed at reducing product fragmentation and building “agentic” capabilities that enable the AI to work autonomously on tasks like writing software and analyzing data.

This feels like the right move structurally, even if it raises questions about execution. Bundling disconnected products into one experience makes sense when your user base is confused about which tool does what. But the real test will be whether the superapp can compete with purpose-built tools that already do each of these things well. A browser that’s also a coding tool that’s also a chatbot is a hard product to get right, and OpenAI’s track record on shipping polished apps has been mixed at best.

3. ChatGPT Expands Into Commerce with 14 App Partners

OpenAI launched an Apps SDK with 14 integration partners, including Spotify, Uber, DoorDash, Expedia, Figma, and Canva. The setup transforms ChatGPT into a discovery and planning layer where users can build playlists, browse restaurants, or search flights, then get handed off to the partner app to complete transactions. The integrations are built on the Model Context Protocol (MCP) and are currently limited to the US and Canada due to GDPR restrictions.

Can you say whiplash? This is a counter signal that indicates OpenAI still sees itself as a consumer platform rather than an enterprise solution. The App Store comparison is obvious and intentional. Partners gain distribution to ChatGPT’s 800 million users, and OpenAI gains deeper engagement and longer session times. What’s less clear is how OpenAI plans to monetize this layer. They haven’t disclosed whether they’ll take revenue shares or charge for placement, and history suggests that controlling the discovery layer without a clear business model is a risky bet. It’s worth watching, but I wouldn’t call it a strategy yet.

3 Things I Think You’ll Like: AI Voice Assistants

Voice interfaces for AI are moving beyond simple question-and-answer into something more useful. Dictation is great, but the latest generation of AI voice tools can take actions, connect to your work systems, and handle multi-step tasks through conversation. If you’ve been waiting for voice AI to become practical rather than just novelty, these three options represent different approaches to the problem.

1. 11.ai

ElevenLabs, widely recognized as a leader in AI voice technology, launched 11.ai as a personal AI voice assistant that goes beyond answering questions to take actions within your daily workflows. It uses the Model Context Protocol (MCP) to integrate with tools like Linear, Perplexity, Slack, and Notion, letting you plan your day, research prospects, manage projects, and summarize communications through voice commands. The system offers ultra-low latency and automatic language detection. It’s currently available as a free alpha, which means it’s early but worth testing if you want to see where voice-first productivity is heading.

2. Lindy Phone Calling

Lindy, which I’ve featured before as an AI employee platform, now offers phone-calling capabilities that let your AI agents make and receive calls. The feature extends Lindy’s workflow automation to voice interactions, handling tasks such as appointment scheduling, lead qualification, and customer follow-ups through real conversations. Based on what I’ve seen, it’s a good effort but not quite there yet. Lindy has indicated more improvements are coming, so it’s worth keeping on your radar even if the current version feels like an early step.

3. VAPI

VAPI takes a developer-first approach to AI voice agents, offering a platform for building custom voice-powered applications. Rather than providing a ready-made assistant, VAPI gives you the infrastructure to create voice agents tailored to specific use cases, from customer support lines to internal tools. I haven’t tested this one personally, so I’m including it based on what I’ve seen in the documentation and community discussion. If you have development resources and want full control over your voice AI experience, this is the option to evaluate.

Bottom line: Start with 11.ai if you want a ready-to-use voice assistant with real integrations. Watch Lindy if you’re already in their ecosystem and want voice added to existing workflows. Look at VAPI if you have dev resources and need a custom solution.

What’s your take on voice as an AI interface? Are you still a keyboard-first person, or have you started talking to your tools?

The gap between the AI narrative and AI reality keeps widening. This week’s articles share a common thread: what companies claim about AI and what actually happens when people use it are often two very different stories. Block claims eventual AI efficiency justified cutting 4,000 jobs; the reality is likely something else. An eight-month HBR study found that AI doesn’t reduce work at all; it intensifies it. And a CNBC survey reveals that men and women experience AI adoption in fundamentally different ways.

Here are three things that caught my attention this week, along with three tools for connecting your AI agents to the systems they need to be useful.

1. I Worked for Block. Its A.I. Job Cuts Aren’t What They Seem.

Jack Dorsey’s Block laid off 4,000 employees, framing the cuts as part of an AI-driven efficiency push. Dorsey has championed the idea that AI will enable “single-digit teams” to accomplish what previously required large organizations. The problem, as the author (a former Block employee) points out, is that Block had grown to over 12,000 employees despite Dorsey’s long-held philosophy of staying small and decentralized. Sam Altman weighed in this week as well, noting that many companies claiming AI-driven layoffs are actually engaging in “AI washing,” using AI as cover for correcting pandemic-era overhiring that had nothing to do with AI capability.

AI washing is becoming a real pattern worth watching. When a company announces layoffs and credits AI efficiency, it’s worth asking whether AI drove that change or whether it’s a convenient narrative layered on top of a headcount correction. The distinction matters because it distorts the broader conversation about what AI can and can’t do. If we attribute every restructuring to AI, we lose the ability to evaluate where AI is genuinely changing the nature of work versus where it’s just a talking point.

2. AI Doesn’t Reduce Work, It Intensifies It

What is the optimum outcome of the AI revolution? A reduced workload for knowledge workers, or pushing every employee to their productivity limit? An eight-month study at a U.S.-based technology company with about 200 employees found that generative AI tools consistently intensified work rather than reducing it. Researchers identified three forms of intensification: scope expansion, where workers took on tasks previously handled by others (product managers writing code, researchers doing engineering work); temporal blurring, where AI’s low-friction prompting style caused work to seep into breaks, lunches, and evenings without feeling like “real work”; and parallel overload, where employees managed multiple AI-assisted threads simultaneously, creating constant context-switching. The company didn’t mandate the use of AI. Employees adopted it voluntarily because AI made “doing more” feel possible, and in many cases, rewarding.

This research suggests an interesting dynamic at the individual level: workers start out accomplishing more with AI, but the initial productivity surge can give way to cognitive fatigue and burnout. The researchers recommend developing an “AI practice,” a set of intentional norms around when to use AI, when to stop, and how to prevent scope from quietly growing beyond what’s sustainable. If you’re leaning heavily into AI tools, it’s worth asking whether your workload is shrinking or just changing forms.

3. AI’s Got a Gender Gap: Women Are More Skeptical

A CNBC/SurveyMonkey survey found a significant gender gap in AI attitudes and adoption. 69% of men view AI as a “valuable assistant” compared to 61% of women. 50% of women say using AI at work “feels like cheating” versus 43% of men. The usage gap is even wider: 64% of women report never using AI at work compared to 55% of men. Sheryl Sandberg warned that if women fall behind in AI adoption and training, the consequences could be disproportionate, both for individual careers and for economic participation more broadly.

The survey also found that both men and women want more AI training, which suggests the gap isn’t about capability but about comfort, framing, and organizational support. That’s an important distinction. It means the gap is addressable, but only if organizations take the implementation approach seriously rather than just making tools available and assuming people will figure it out.

3 Things I Think You’ll Like: Agentic / Tool-Calling Ecosystems

One of the practical challenges with AI agents is connecting them to the systems where your work gets done. An agent that can write well but can’t access your CRM, calendar, or project management tools is limited in what it can accomplish. These three platforms take different approaches to solving that problem, from managed tool libraries to unified APIs to full MCP runtimes.

1. Composio

Composio provides a managed library of tool integrations designed specifically for AI agents, supporting connections to over 250 applications, including CRMs, calendars, project management tools, and communication platforms. I’ve been making significant use of this tool in my agent experiments. It provides a strong control mechanism for managing which tools agents can access, thereby enabling the principle of least privilege. The command line interface is particularly useful if you’re working with Claude Code, as it allows direct agent and tool management from the terminal. The pricing is also reasonable, making it practical for smaller teams to start experimenting with agentic workflows.

2. Merge

Merge takes a different approach by offering a unified API that normalizes data across hundreds of third-party platforms. Instead of building individual integrations for each tool, you build once into Merge’s API and get access to integrations across CRM, HRIS, accounting, ticketing, file storage, and ATS categories. The value is in the normalization layer: if you’re working with systems that use different data structures, Merge standardizes them so your tools and agents work consistently regardless of what sits on the other end. Merge also handles ongoing integration maintenance, which is worth considering if you don’t have dedicated engineering resources.

3. Arcade

Arcade is an MCP runtime platform that handles the security and authorization layer for AI agents performing real actions across business systems. Where Composio focuses on the tool library and Merge on data normalization, Arcade’s differentiator is governance: agents operate with user-specific permissions rather than service accounts, integrating with existing OAuth and identity providers. They offer what they describe as the largest catalog of MCP tools built specifically for agent reliability rather than simple API wrappers. I haven’t explored this one yet, but the focus on enterprise-grade authorization and compliance makes it worth investigating if security governance and user-level permissions are a priority. Arcade supports cloud, VPC, on-premises, and air-gapped deployment options.

Bottom line: I have hands-on experience with Composio, but not the other two options. Composio is the most accessible starting point for connecting agents to tools quickly. Merge may be the better fit if your core challenge is normalizing data across different systems. Arcade is worth evaluating if security governance and user-level permissions are your primary concern.

Have you found a better way to connect your systems to your agents?

A heavy dose from Salesforce this week, with Marc Benioff laying out his vision for the “agentic enterprise” and the company overhauling its partner program to match. Meanwhile, Anthropic finds itself caught in a whirlwind of opposing forces, where blowback from a dispute with the U.S. government triggers a wave of new users large enough to bring down its systems.

Here are three things that caught my attention this week, along with three OpenClaw alternatives worth exploring.

This Friday: Get More Out of Claude

Claude just hit #1 in the App Store. If you are one of the millions of new users wondering what this tool can actually do beyond chat, I am running a Partnering with AI workshop this Friday. We will cover Skills, Cowork, and Claude Code, and other capabilities that most people don’t even know exist, and build real automation together. Friday, March 6, 12-2pm MT. Register here.

1. The Truth About AI

Marc Benioff published an essay in TIME outlining what he calls the “agentic enterprise,” a model where AI agents move beyond answering questions to autonomously completing tasks, making decisions, and collaborating within organizations. The vision describes businesses that operate more like adaptive organisms, with AI handling routine operations and humans focusing on creativity, strategy, and complex problem-solving.

The interesting part is the framing. Benioff positions this as an inevitable evolution rather than an aspirational one, and the language maps directly to Salesforce’s Agentforce product line. For ecosystem members, this signals where the largest CRM ecosystem is directing its investments. Whether or not you buy the full vision, the practical implication is clear: Salesforce is betting its future on agentic AI, and any partnerships built around Salesforce’s ecosystem will need to follow suit.

2. Salesforce Pushes Partners to Prioritize Agentic AI Outcomes

Salesforce is revamping its consulting partner program to focus on “agentic AI outcomes” for customers, consolidating from multiple tracks into two tiers (Summit and Select) and replacing 170 legacy badges with 28 core certifications. The restructuring emphasizes partner specializations, competencies, and measurable customer satisfaction rather than traditional implementation metrics.

This is significant because it treats AI adoption as an operational problem rather than a technical one. Salesforce is essentially telling its partner ecosystem that deploying Agentforce 360 is not enough; partners need to prove that customers are getting real financial returns from their AI investments. The shift to outcome-based evaluation mirrors what partnership professionals across the industry are dealing with: the focus is moving from “have you implemented AI?” to “did AI produce measurable results?” The concept of “agents-as-a-service” replacing traditional SaaS is worth watching closely.

3. Anthropic’s Claude Reports Widespread Outage

Anthropic’s Claude services experienced widespread disruptions this week, with thousands of users reporting login issues affecting Claude.ai and Claude Code. The API remained functional, and Anthropic identified the issue and implemented a fix, but outages lingered through the week. The disruption followed a surge in new users that pushed Claude to the number one spot in Apple’s App Store, driven in part by attention around a dispute with the Pentagon and President Trump’s directive for federal agencies to stop using Anthropic products.

The pattern here is fascinating. A political dispute generates media coverage, which generates curiosity, which generates a user spike large enough to take down production systems. Free user signups reportedly jumped by more than 60%. For anyone building workflows on top of AI platforms, this is a reminder that reliability planning matters. The tools we increasingly rely on for daily work can still be disrupted by factors unrelated to the technology itself. It also highlights how rapidly public perception can simultaneously influence a company’s momentum, for better and worse.

3 Things I Think You’ll Like: Top OpenClaw Alternatives

Last month, I published a special-edition deep dive on OpenClaw. The conclusion I came to was that it is a fascinating experimental platform, but one built for roughly 0.1% of the general population. I believe that true AI agents are coming from the major LLM vendors in 2026, but we are not there yet. In the meantime, these three tools provide a more stable commercial offering than the bleeding-edge open-source OpenClaw approach.

1. Manus

Manus is an autonomous AI agent platform that uses a multi-agent architecture where one agent plans, another executes, and a third reviews the work. Now part of Meta, the platform recently shipped version 1.6 Max with stronger planning and problem-solving capabilities, plus a “Wide Research” mode that spins up multiple full Manus instances as subagents to tackle broad research tasks in parallel. A live dashboard streams the agent’s actions in real time so you can see what it is doing and step in mid-task to redirect. I experimented with Manus a few months ago and plan to revisit it soon to see how it has evolved.

2. Copilot Tasks

Microsoft’s Copilot Tasks represents a shift from chat-based AI to action-based AI, essentially a to-do list that executes itself. You describe what you need in natural language, and Copilot works in the background with its own browser and compute environment to complete the task, reporting back when finished. Tasks can be one-time, recurring, or scheduled. I was not impressed with my first experience building AI workflows in Copilot Studio, so I am hoping Tasks provides a better entry point.

3. Perplexity Computer

Perplexity Computer is a multi-agent system that coordinates 19 different AI models to execute complex workflows, distributing subtasks based on each model’s strengths. It uses Claude Opus 4.6 for core reasoning, Gemini for research, and other specialized models for image generation, video, and rapid processing. The catch is availability and cost: it requires the $ 200-per-month Perplexity Max subscription, and early feedback has raised concerns about credit usage. I am looking forward to testing this when it becomes available on lower-tier plans.

Bottom line: If you are already in the Microsoft ecosystem, Copilot Tasks is the lowest-friction option to start with. If you want a dedicated agent platform, Manus offers the most structured multi-agent approach. Perplexity Computer is the most ambitious of the three, but also the most expensive and least accessible at this point.

What other OpenClaw alternatives have caught your eye?

When Anthropic released its Cowork plugins a few weeks ago, the stock market lost its mind. Software stocks dropped $830 billion in a single day. Analysts at Jefferies called it the “SaaSpocalypse.” And the trigger wasn’t a new AI model or some breakthrough in reasoning. It was plugins: bundles of markdown files that package up entire job functions (sales, marketing, legal, finance) and hand them off to AI. The message from Wall Street was clear. When an AI company can ship a set of config files that enable the AI to do the same tasks as purpose-built software, software has a problem.

That reaction looked dramatic at the time, but the sell-off continues. In the weeks since, Microsoft’s AI CEO predicted that most white-collar tasks will be automated within 18 months, a New York Times essay argued that AI coding has already made professional software dramatically cheaper, and OpenAI signed multi-year deals with McKinsey, BCG, Accenture, and Capgemini to deploy its Frontier agent platform across the enterprise. The common thread is that AI companies are no longer building tools to help you use software. They’re building tools to replace it.

Here are three things that caught my attention this week, along with three Claude Cowork plugins worth installing if you want to see what all the fuss is about.

1. The A.I. Disruption Has Arrived, and It Sure Is Fun

Paul Ford, programmer and co-founder of the AI-driven software platform Aboard, wrote a guest essay for the New York Times arguing that AI coding tools have crossed a meaningful threshold. His core claim: using Anthropic’s Claude Code at $200 a month, he’s completing projects worth tens of thousands of dollars on weekends and evenings. He acknowledges the code isn’t production-grade, but argues that speed and cost matter more than polish for a huge range of unbuilt software.

That honesty is refreshing in a landscape full of breathless predictions. For knowledge workers, the practical takeaway is that the cost of building custom tools, dashboards, and integrations is collapsing. The skill that matters now isn’t writing code. It’s defining what needs to be created, prompting for it, and evaluating whether the result is valuable enough to redefine your work.

2. Microsoft AI CEO predicts ‘most, if not all’ white-collar tasks will be automated by AI within 18 months

Microsoft AI CEO Mustafa Suleyman predicted that within 12 to 18 months, AI will achieve “human-level performance” in most white-collar tasks, including law, accounting, and project management. He noted that software engineering is already seeing this shift through AI-assisted coding. Senator Bernie Sanders responded by calling the prediction an “economic earthquake” and reiterated his call for a moratorium on new AI data centers.

Suleyman has made bold predictions before, and 18 months is an aggressive timeline. But the pattern is worth paying attention to: when the CEO of a major platform company makes this kind of public statement, it shifts corporate planning. For partnership professionals, this means the organizations you work with are likely accelerating their AI strategies, whether or not the timeline holds. The practical response isn’t to panic. It’s to make sure you’re already experimenting with AI tools so you can speak credibly when partners and leadership start asking questions.

3. OpenAI partners with McKinsey, BCG, Accenture, and Capgemini to push its Frontier AI agent platform

OpenAI announced multi-year “Frontier Alliances” with four of the world’s largest consulting firms to sell and implement its Frontier enterprise agent platform. BCG and McKinsey will focus on strategy and operating model design, while Accenture and Capgemini will handle end-to-end systems integration. Each firm is building dedicated practice groups and certifying teams on OpenAI technology.

This is a significant partnership story. These consulting firms are deeply embedded with the very SaaS companies that Frontier could displace, including Salesforce, Microsoft, and ServiceNow. Having McKinsey and BCG actively evangelize an alternative AI platform to the C-suite creates real tension in those relationships. For partnership professionals, this is worth watching closely. When consulting firms start building certified practices around AI agent platforms, it signals that the enterprise sales conversation is shifting from “add AI to your existing tools” to “replace your existing tools with AI.”

3 Things I Think You’ll Like: Claude Cowork Plugins

Anthropic open-sourced 11 plugins when it launched the Cowork plugin system, covering everything from sales to legal to biology research. Each plugin bundles skills (domain knowledge Claude automatically draws on), slash commands (actions you trigger explicitly), and connectors (integrations with external tools like CRMs and project trackers). These are GitHub repositories, which not everyone may know well, but everyone should start familiarizing themselves with them. The plugins are built as simple Markdown (formatted text) and JSON (data structure) files, so you can customize them for your own workflows. Here are three that are particularly relevant for partnership professionals.

1. Sales Plugin

The sales plugin turns Claude into a prospecting and pipeline management assistant. It includes slash commands for call preparation, pipeline review, forecast generation, and drafting outreach, all of which connect to your CRM, email, and meeting tools through MCP connectors. The practical use case for partnership work is straightforward: run /sales:call-prep before a partner meeting and Claude pulls recent interactions, company intel, and competitive context into a briefing. I’ve been testing this one and the call prep workflow alone saves significant time when you connect it to your actual data sources.

2. Marketing Plugin

The marketing plugin handles content creation, campaign planning, SEO audits, and brand voice management. You can configure your brand voice, style guide, and target personas in a local settings file, and commands like /draft-content and /brand-review will automatically apply your standards. For partnership professionals managing co-marketing programs or partner content, this means you can maintain consistency across partner-facing materials without having to manually enforce guidelines every time. The SEO audit command is particularly useful if you’re evaluating partner content or landing pages.

3. Productivity Plugin

The productivity plugin gives Claude persistent context about your work through task management, workplace memory, and a visual dashboard. The standout feature is the two-tier memory system: Claude learns your shorthand, your people, your projects, and your terminology. Say “ask Todd to handle the QBR for Acme,” and Claude knows who Todd is, what a QBR involves, and which account you mean. For partnership professionals juggling multiple partner relationships, the ability to teach Claude your internal language and processes is where plugins start to feel less like tools and more like an actual working relationship.

Bottom line: Install the productivity plugin first to teach Claude your working context before layering on the specialized plugins. Add the sales plugin if you spend significant time on partner recruitment or co-sell motions. Try the marketing plugin if co-marketing and content consistency are priorities.

Which partnering function would benefit most from having an AI plugin built for your workflow?

Unless you’ve been living on a beach in Tahiti, you’ve been inundated by references to OpenClaw for the past month. The open-source AI agent hit 150,000 GitHub stars in under a month (the geek equivalent of “going viral” on social media), and demos of it autonomously managing emails, browsing the web, and running tasks across messaging platforms had people calling it the closest thing to JARVIS we have seen (Iron Man reference). My reaction was cautious curiosity, as I wrote a couple of weeks ago. The security concerns were real, but after a couple of weeks, I decided, what the hell, let me see for myself. What I found was more nuanced than the hype or the fear suggested.

Here is what I learned while setting up and running OpenClaw over the past week, along with five resources to make your own setup easier.

1. Building a Zero-Trust Company with AI-Augmented Operations

Peter Hallen runs a one-person compliance consultancy that operates like a team of ten. He built a virtual agency with four AI agents orchestrated through an OpenClaw gateway on a zero-trust infrastructure for under $200 a month. The article covers the org chart and team norms for his agents, the security architecture organized layer by layer using the OSI model, and the specific tooling: Tailscale mesh networking, WireGuard encryption, EC2 with zero open ports.

What made this my starting point is how Hallen treats AI agents, as you would treat human employees, from a security perspective. Least privilege access. Scoped tool permissions. Isolated sessions. Human-in-the-loop approval for all external actions. I used this framework to guide every decision during my own setup. For anyone considering OpenClaw, read this before you touch a configuration file.

2. I Used Superpowers in Claude Code to Deploy OpenClaw on AWS

Setting up OpenClaw on AWS is not really a coding project. It is a configuration and infrastructure project that requires running hundreds of commands through SSH and the AWS CLI, editing YAML and Markdown files, and managing dependencies. I launched Claude Code from a local project folder and used the Superpowers plugin to brainstorm, plan, and execute the entire deployment. Claude Code ran virtually every command, from provisioning the EC2 instance to configuring the gateway.

I sat with Claude Code, suggested the architecture, reviewed security requirements, and then executed step by step. Hundreds of CLI commands that would have taken me days to research and run got handled in a few focused sessions. Using one AI agent to deploy another is where operational workflows are going. Claude is also managing my weekly security audits. There is still a lot of fiddling, but at least I’m not doing the typing.

3. Virtual Server vs Hardware: $1.93 vs. $1,000

The conventional wisdom in the OpenClaw community is to run it on a Mac Mini on your home network. I set up my instance on an Amazon EC2 virtual server, and after a full week, the total infrastructure cost has been $1.93. That is not a typo. Your Claude/LLM costs increase with usage, regardless of where you host, but the infrastructure itself can be practically free.

I chose AWS for the security infrastructure. A virtual server gives you security groups, VPC isolation, CloudWatch monitoring, and user controls that are hard to replicate on a box connected to your home router. The tradeoff is some familiarity with AWS or a willingness to let Claude Code handle it for you.

3 (to 5) Things: OpenClaw Resources

These are the five resources I found most useful during my own deployment.

1. OpenClaw Discord Community

This is where the active community lives, and it was the most useful resource I found during setup. Dedicated channels for troubleshooting, security, skill development, and platform-specific guides. When you hit a wall at 11pm, and the docs don’t cover your configuration, this is where you find someone who has already solved it.

2. OpenClaw Getting Started Guide

The official docs walk through installation, connecting your first messaging platform, setting up API keys, and configuring the gateway. My advice: read the entire guide before you start. Understanding the full architecture upfront saves you from making early decisions you'll later want to undo.

3. OpenClaw GitHub Repository

The source of truth. Beyond the code, the README and issues section are useful for understanding current limitations, known bugs, and where development is headed. The project is MIT-licensed, and the commit history gives you a sense of how active the maintainers are.

4. OpenClaw Manager Plugin for Claude Code

This made my deployment dramatically smoother. The plugin turns Claude Code into a dedicated OpenClaw assistant that handles installation, channel configuration, security hardening, and troubleshooting. Instead of copying commands from docs and hoping you have the syntax right, you describe what you want, and Claude Code runs it. Community-built and MIT-licensed.

5. Tailscale Gateway Configuration

If you are running OpenClaw on a remote server, you need a secure way to access it without exposing it to the public internet. Tailscale keeps the server secure while handling HTTPS, routing, and identity-based authentication through your tailnet. Cryptographic identity verification means authenticated requests bypass token auth. This was a key piece of my security setup.

A note on ClawHub: the skills marketplace is promising in concept, but I have held off. Security researchers have found hundreds of malicious skills that combine prompt injection with traditional malware. Because skills are defined in Markdown, it is easy to embed hidden instructions. The project has partnered with a security firm to audit skills and added virus scanning, but I would hold off until the vetting process catches up.

Bottom line: OpenClaw is a fun project if you are technically inclined and want to be on the bleeding edge of the AI conversation. That is probably 10% of the people investing in building their AI skills right now. Of that group, maybe 1% will turn it into a truly autonomous agentic tool that moves the needle in their professional life. For everyone else, your time is probably better spent exploring commercial workflow AI tools like Claude Cowork. I believe the leading LLM companies will provide these types of agentic toolkits in the very near future. Meta acquired Manus and its agent frameworks. OpenAI hired the founder of the OpenClaw project last week. If that is not a signal, I do not know what is.

Where do you think agentic AI tools are headed? Are you waiting for the commercial platforms or building your own?

The past two weeks have revealed a rift between AI and traditional software. Between Anthropic releasing plugins that turn Claude into a genuine workflow tool, Microsoft’s AI CEO suggesting that vibe coding could replace entire categories of software, and OpenAI launching an enterprise agent platform, the message is clear: AI companies are no longer building tools that complement existing software. They’re building tools that replace it.

I’ve been watching this pattern closely because it could directly affect partnership professionals. The software ecosystem that most partner programs are built around is about to look very different. When an AI platform can handle core tasks, generate presentations, manage pipelines, and connect natively to your data sources, the value proposition of standalone SaaS point solutions becomes harder to justify. The question isn’t whether this disruption is coming. It’s how quickly it arrives.

Here are four things that caught my attention this week, along with three interactive Claude connectors worth exploring if you’re ready to integrate AI directly into your workflow tools.

3 Things for Thursday, February 12, 2026:

1. Something Big Is Happening

An AI startup founder went viral with what reads like an urgent warning about the pace of AI advancement. His core argument is that current models, including GPT-5.3 Codex and Opus 4.6, now perform his technical coding work better than he can, including self-testing and iteration. The piece traces AI’s trajectory from struggling with basic math in 2022 to passing the bar exam in 2023 to actively contributing to its own development today.

The practical takeaway is blunt: the window for adapting is narrower than most people realize. The author argues that nothing done on a computer is safe in the medium term, which tracks with what we’re seeing in the software space. The skills that will continue to matter most are the ones AI can’t replicate: building relationships, being physically present, and exercising judgment in ambiguous situations. Partnership professionals who are already integrating AI into their workflows while maintaining their humanity will have a significant head start over those who haven’t figured it out yet.

2. Anthropic Rolls Out Plugins for Claude Cowork Workflows

Anthropic introduced plugins for Claude Cowork, providing a library of verticalized capabilities while integrating with third-party applications and services. Initial partners include Google Search, calendar applications, and meeting transcription services like Zoom. The feature moves Claude beyond a traditional chatbot interface toward an operating system for knowledge work, where it can perform actions, access real-time information, and handle complex, multi-step tasks.

This is significant for knowledge workers because it represents the verticalization trend in action. When Claude can summarize your meetings, check your calendar, search the web, and manage your CRM/PRM from a unified interface, the need for separate point solutions diminishes. I’ve been using Claude’s connector capabilities for weeks now, and the difference between having Claude do something on your behalf versus manual data entry is nothing short of life-altering. The connectors are what make it a coworker rather than a water cooler.

3. Microsoft AI CEO says vibe coding is making apps easier to build

Microsoft AI CEO Mustafa Suleyman stated that vibe coding is dramatically lowering the barrier to building applications, making it possible for anyone to create web apps in seconds without deep technical skills. He suggested this shift could put traditional software categories at risk, as AI agents become the dominant interface for getting work done. The trend has already triggered investor anxiety and sell-offs across the software sector.

The pattern here connects directly to what we covered in January about vibecoding. The difference now is that a major platform CEO is saying it publicly, which signals that this isn’t just startup hype. For partner managers, the implication is that the software products your partner programs are built around may face serious disruption. If you can build your own customized AI-powered solutions instead of buying packaged software, the partner management function changes significantly.

4. Introducing OpenAI Frontier

OpenAI launched Frontier, a platform designed to help enterprises build, deploy, and manage AI agents at scale. The platform focuses on giving agents shared business context, onboarding and feedback loops, and explicit permissions and guardrails. OpenAI positions it as moving beyond isolated AI pilots to deploying agents as dependable coworkers that operate across multiple systems.

The interesting part is the framing. OpenAI is explicitly using workplace language: onboarding, coworkers, identities, and permissions. This isn't about building better chatbots. It's about creating an AI workforce management platform. This represents another entry point into the enterprise AI agent market, alongside Anthropic's approach with Claude and plugins. Will 2026 really see the adoption of truly collaborative cross-functional AI agents?

3 Things I Think You’ll Like: Interactive Claude Connectors

Connecting your AI tools to your data is a key step in moving beyond chatbots and into coworkers/agents. Claude has significantly expanded its list of connectors, and the interactive connectors are where things get interesting. These three options connect Claude directly to presentation, sales, and creative tools, letting you create and enrich compelling content simply by describing what you want and having Claude go to work.

1. Gamma MCP Server

Gamma’s connector brings AI-powered presentation creation into Claude through an MCP integration. Rather than switching to a separate tool to build slides, you can create and iterate on presentations within the same conversation where you’re analyzing data or drafting content. I’m a fan of Gamma’s approach to presentations, and integrating it with Claude and your other connectors takes it to another level. The practical use case for partnership professionals is to build partner-facing presentations that draw on your actual data and analysis, all within one workflow.

2. Clay MCP Server

Clay’s Claude connector brings go-to-market data enrichment directly into your AI workflow. Clay is quickly becoming a core GTM tool for sales and partnership teams, and having direct access in Claude opens up the ability to enrich contacts, build prospect lists, and research accounts without leaving your AI workspace. For partnership professionals managing partner recruitment or co-sell motions, the combination of Claude’s analytical capabilities with Clay’s data enrichment means you can research, qualify, and prepare for partner conversations in a single workflow.

3. Canva AI Connector

Canva’s MCP connector lets you control Canva directly from Claude, enabling you to create and edit designs without switching applications. You describe what you need, and Claude generates or modifies Canva designs using your existing brand assets and templates. For partnership professionals creating co-marketing materials, partner decks, or social media content, this eliminates the back-and-forth between writing copy in one tool and designing in another. The interactive experience happens entirely in Claude, no need to swivel between applications. Going from concept to finished design inside a single conversation is a significant workflow improvement.

Bottom line: Start with Gamma if you build frequent presentations and want to streamline the process from research to finished deck. Look at Clay if your focus is on partner recruitment, co-sell, or account-based motions. Try Canva if your work is heavily visual and you create partner marketing materials.

What type of interactive connection would you like to see between Claude and your critical work applications?

The past week in AI felt like watching a train wreck in slow motion. Clawdbot launched, renamed itself to Moltbot to avoid trademark issues, rebranded as OpenClaw, and disclosed security vulnerabilities serious enough that I’m glad I paused my own experiments this weekend. The whole saga demonstrates why 99% of us should be spending our time with commercial AI agent offerings rather than the DIY chaos. These companies have already solved the hard problems in security, reliability, and integration that open-source experiments are still grappling with. If you want an AI assistant that actually works without turning your accounts into a security nightmare, check out the options below.

Here are three things that caught my attention this week, along with three commercial AI agent platforms worth exploring if you’re ready to delegate actual work.

1. This whole AI thing is simpler than you think

Fast Company reports that only 25% of AI initiatives have achieved expected returns over the past three years, according to Microsoft research. The article argues this isn’t a technology problem or a people problem, but an organizational one. Most companies are trying to implement AI using 20th-century industrial age architectures that aren’t aligned with 21st-century goals.

The solution proposed is a “human-first” approach that restructures the organization around AI opportunities rather than trying to integrate AI into existing workflows. This means developing workplaces that account for human systems that can utilize and grow with AI, rather than reprogramming the workforce around AI’s needs. The key insight is that operating systems designed for efficiency and productivity at the expense of human well-being won’t unlock AI’s potential.

2. The AI adoption story is haunted by fear as today’s efficiency programs look like tomorrow’s job cuts

McKinsey research identifies three key areas where leaders can strengthen what people bring to the table in the age of AI: don’t allow fear to shrink ambition, use AI as an input rather than a default, and keep humans at the center of value judgments. This requires creating protected space for AI experimentation, designing decision-making to ensure AI informs judgment rather than replaces it, and articulating clear boundaries for AI decision-making.

The article emphasizes that judgment, ethics, and values cannot be outsourced to AI. Leaders who get this moment right will deploy AI tools in ways that tap into psychological safety, human judgment, and ethical clarity. Companies like Siemens and Toyota explicitly protected jobs while reinventing their production systems, gaining long-term innovation as a result. The difference between empowerment and replacement will define how history judges this transition.

3. ServiceNow chooses Claude to power customer apps and increase internal productivity

ServiceNow selected Claude as the default AI model for its Build Agent and a preferred model across its AI Platform. The partnership focuses on integrating AI into enterprise operations at scale with security as a priority. Claude now powers ServiceNow Build Agent, allowing developers to create apps and automations using natural language, with usage expected to quadruple.

Internally, ServiceNow reports that a Claude-powered coaching tool reduces sales preparation time by up to 95%, and Claude Code assists engineers with writing, reviewing, and debugging code. The implementation aims to reduce customer implementation time by 50% and supports agentic applications in sectors like healthcare and life sciences. This makes Claude available to tens of thousands of ServiceNow enterprise customers and its global workforce, demonstrating how enterprise AI adoption looks when properly implemented.

3 Things I Think You’ll Like: All-in-One AI Agent Platforms

The Clawdbot/Moltbot/OpenClaw chaos highlighted real demand for AI agents that can operate autonomously across multiple systems. Rather than wrestling with DIY solutions that require technical expertise and pose security risks, commercial platforms now deliver this capability with appropriate guardrails. These three options offer different approaches to the all-in-one agent problem, and I’d focus on these commercial offerings rather than taking risks with OpenClaw at this point.

1. Marblism

Marblism positions its offering as “AI Employees” that handle tasks automatically without requiring prompt engineering skills. The system requests input or approval only when necessary, allowing you to define the work rather than manage execution. The platform integrates with popular software, manages multiple businesses under a single account, and learns from user input to improve over time. It includes a 7-day money-back guarantee, live chat support, and guided onboarding with an account manager, which addresses the setup friction that makes DIY solutions challenging.

2. folk

folk markets itself as “the CRM that works for you” and takes a different approach to the AI agent problem by embedding intelligence into relationship-management workflows. Rather than building a general-purpose assistant, folk focuses on automating work related to contacts, companies, and deals. The platform handles data entry, follow-up reminders, and context gathering automatically, letting the AI agent operate within a defined domain where errors are containable and value is immediate.

3. Lindy

Lindy bills itself as your “first AI employee” and focuses on delegating entire workflows rather than individual tasks. The platform connects to your tools and learns your processes, then handles recurring work autonomously. What distinguishes Lindy is its emphasis on judgment-based tasks rather than simple automation. The system can handle ambiguity, recover from failures, and identify alternative approaches when initial attempts fail. This matches what successful Clawdbot users were trying to achieve, but with a security model designed for business use.

Bottom line: If you want to test the AI agent concept, start with Marblism for guided setup and learning. Check out folk if your primary use case centers on relationship management and CRM workflows. Take a look at Lindy if you’re ready to delegate judgment-intensive tasks and need cross-system autonomy with robust security.

Here’s my litmus test for truly autonomous AI: “What time does my mother-in-law’s flight arrive?” The first agent that can handle that question without additional context will represent AGI in my mind.

What would your ultimate AI agent use case look like if security and privacy weren’t concerns?

I’ve been neck-deep in Claude’s skills system this week. Started out curious about the feature, and ended up rebuilding most of how I handle partner data reporting. The difference between one-off prompting and having reusable skills turns out to be bigger than I expected. Instead of explaining my data structure every time I need help, Claude now just knows it. Instead of getting slightly different approaches each time, I get consistent results. Feels less like using a tool and more like having a new coworker who actually remembers what you told them last week. FYI, ChatGPT is moving in the Skills direction as well, but seems to be for developers only at this point.

Ready to take your Claude usage to the next level? Check out our workshop this Friday: https://luma.com/6j8tpzez

Here are three things that caught my attention this week, along with three resources for building and sharing Claude skills.

1. I broke up with ChatGPT

A long-time ChatGPT user explains why they switched to a multi-AI workflow, citing declining performance, ads, and the recognition that the era of a single all-purpose AI is over. The piece breaks down specific strengths across current platforms: Claude 4.5 for long-form writing and document analysis, Perplexity for cited research, Gemini 3 for multimodal processing, and ChatGPT for conversational voice interactions. The core argument is that specialized tools outperform general-purpose ones for specific tasks.

For partnership professionals, the takeaway is that it’s worth diversifying and understanding the strengths and weaknesses of different tools. Not everyone has the time or resources to experiment with every option, but if you’ve only been working with a single LLM, it’s a good idea to explore at least one other. I’ve been doing something similar for months: Claude for writing and analysis, Gemini for images when I need them, Perplexity for citations when they matter. The switching cost is lower than you’d think, and the quality difference can be significant. The annoying part is managing context across platforms, but you adjust.

2. Thanks But No Thanks on the Claudeswarms, Kevin Roose

This article pushes back on Kevin Roose’s claim that Silicon Valley insiders are dramatically ahead of everyone else in AI adoption through intensive “vibecoding” and building apps in hours or minutes. The author argues that much of this activity amounts to busywork with limited transformative impact, noting AI’s current limitations, particularly for tasks requiring physical interaction beyond pure software development.

The interesting part is that there’s a wide gap between using AI in every aspect of your life and where most people are today. But it’s worth learning concepts like vibecoding and having the ability to create simple apps and workflows that make your life better. Not everything has to be done to the nth degree. Building a quick app to track partner engagement might break when your data schema changes, but the ability to build it in the first place is valuable.

3. Clawdbot Showed Me What the Future of Personal AI Assistants Looks Like

The article reviews Clawdbot (now renamed to Moltbot), an open-source personal AI assistant that runs locally on your computer and operates as an agent with direct access to the shell and filesystem. Powered by Claude Opus 4.5 and accessed via Telegram, it can execute commands, install new skills, integrate with services like Notion and Todoist, and replace cloud-based automations with local cron jobs. The author’s customized version demonstrates highly personalized capabilities, including image generation, audio transcription, and adaptive learning through local Markdown files.

This matters because it shows where personal AI assistants go once you get past browser-based chatbots. Running an AI agent locally with full system access opens up workflows that aren’t possible otherwise. The catch is you need technical chops to set this up and maintain it, which puts it out of reach for most people right now. But I can feel another dive down the rabbit hole coming. I’m going to have to figure out Moltbot this weekend.

3 Things I Think You’ll Like: Claude Skills Resources

The skills ecosystem is expanding fast. Multiple directories and repositories have popped up to help you find and share what people are building. Here are a few samples:

1. Anthropic’s Skills Repository

Anthropic’s official collection showing what you can do with Claude’s skills system. Each skill lives in its own folder with a SKILL.md file that contains the instructions Claude reads. The repository includes document-focused skills for Word, Excel, PowerPoint, and PDFs, plus examples of web testing and brand guidelines. Worth checking out if you’re building custom skills, since the examples show different ways to structure instructions and bundle scripts. Most of these are included by default with Claude, but this repository is also starting to include third-party skills.

2. Agent Skills Directory

A community project trying to standardize how AI agent skills get defined and shared. Uses a structured YAML format with descriptions, examples, and implementation links. Skills are organized by category, like web browsing and data analysis. I haven’t spent much time with this one, but it looks like they’re trying to be the central hub for skill discovery across multiple AI platforms, not just Claude.

3. MCP Market Skills Leaderboard

MCP Market tracks Claude skills by popularity and shows which ones are getting the most use. The leaderboard format surfaces quality through community validation. Useful when you’re trying to figure out which skills are worth installing rather than building everything yourself, though the platform is still young and the metrics might not fully reflect quality for specialized use cases.

Bottom line: Start with Anthropic’s official repository to understand how skills work and see reference implementations. Use the Agent Skills Directory for cross-platform discovery if you’re working with multiple AI tools. Check MCP Market when you want community-validated options rather than comprehensive catalogs.

Interested in getting started with Claude skills? Check out this YouTube video:

Ready to take your Claude usage to the next level? Check out our workshop this Friday: https://luma.com/6j8tpzez

Check out our Workshop Lineup

Each session stands on its own. Take one, take all four, pick and choose based on what’s relevant to your work right now:

Beyond Prompting: Claude’s Toolkit for Getting Work Done

Most people treat Claude and other LLMs like a search engine. This session covers the features that turn it into a real work tool: Projects, Skills, Artifacts, Code, and now Cowork. This workshop will explain all of the options and the workflows that connect them.

Vibe Coding: It’s Not as Hard as It Sounds

You don’t need a technical background to build personalized software. This session walks through creating simple applications using natural language. If you can describe what you want, you can build it. Vibe coding is not just for developers!

Stop Repeating Yourself: Designing AI Workflows That Scale

If you’re copying and pasting the same prompts every week, there’s a better way. This session introduces the PowerFlow Automation Canvas, a framework for designing workflows that help you scale and get more done. Having a mental model for designing workflow will take your AI skills to a new level.

AI Agents: Create Your First Digital Assistant

Everyone's talking about AI agents, but what does that actually look like in practice? This session uses Lindy.ai to build a real assistant, step by step, that you can put to work immediately.

Each workshop is $159. If you refer someone who registers, you get 100% back or credit for a future session. Not a discount. The full amount.

I also host free Office Hours every Monday at 2pm Denver time. No agenda, just drop in with whatever you’re working on.

Full calendar and registration: luma.com/partneringwithai

Anything else to adjust?

In the beginning was the command line, but most business users have had little use for it after 1995. Until now. AI is charging into the future while bringing us back to the past. For years, the command line belonged to developers. That's changing. I've been experimenting with Claude Code for the past few weeks to create personal software, tools, and workflows. Tasks that would've meant learning Python, bash scripting, and API documentation six months ago now happen through simple conversation.

The shift isn’t just that more people can access these tools. It’s what becomes possible when building software becomes more about communication than technical expertise.

Here are three things that caught my attention this week, along with three command-line AI coding tools worth exploring if you’re ready to move from using AI to building with it.

1. Claude Code hits different

Claude Code with Opus 4.5 represents a shift in how software gets created, moving from artisanal craft toward casual process. The tool makes software creation more accessible by emphasizing human creativity and clear problem definition over technical skills. Beyond coding, Claude’s CLI design allows it to use use and access your computer, along with managing tasks like email and calendars. This week, Anthropic extended these capabilities to non-developers with Claude Cowork, a simpler desktop interface built on the same foundations.

For partnership professionals, the gap between identifying a workflow problem and deploying a solution is collapsing. Teams should be thinking about custom tools and integrations rather than waiting for IT or vendors to build what they need. The skill requirement changes from technical implementation to clear problem definition. If you can articulate what you need and evaluate whether it works, you can build it.

2. One reason CEOs tie layoffs to AI? It motivates remaining employees to adopt the technology

Some business leaders are publicly attributing layoffs to artificial intelligence adoption, not just for efficiency gains but also to motivate remaining employees. This strategy aims to accelerate AI integration within companies by signaling that embracing new technology is crucial for job security. The goal is to drive faster upskilling and a cultural shift toward AI utilization among the workforce.

The question is whether fear-based adoption actually works. Partnership professionals should recognize this pattern when AI mandates come from leadership. The pressure to adopt isn’t always about immediate productivity gains. It’s sometimes about establishing AI fluency as a cultural expectation. The teams that build actual workflows and solve real problems will be better positioned than those who simply attend AI training sessions.

3. 4 Googlers share how they reinvented their careers and pivoted to AI

Four Google employees successfully transitioned to AI-focused roles after about a year of dedicated preparation. Their paths varied but commonly involved extensive upskilling. One leveraged internal hackathons for hands-on experience, another committed two hours daily to learning while using content creation to reinforce knowledge, a third spent two and a half years reading books and building solo projects, and the fourth aligned her background in statistics and machine learning with an AI consulting role that emphasized non-technical communication.

The pattern is clear across these stories. Moving into AI roles requires hands-on building rather than passive learning. For partnership professionals, this means setting aside dedicated time for experimentation and project work. The two-hour daily commitment or weekend hackathon approach is more effective than occasional exposure to AI concepts and one-off prompting. The credibility that comes from building actual projects matters more than theoretical knowledge.

3 Things I Think You’ll Like: AI Command Line Coding Tools

Command-line AI coding tools represent a shift from GUI-based development to conversational programming. These tools live in your terminal and understand both your codebase and your intentions, letting you describe what you want in natural language rather than writing every line yourself.

1. Claude Code

Claude Code is Anthropic’s agentic coding tool that lives in your terminal, capable of reading your codebase, running commands, and handling repetitive work through natural language. You describe what needs to happen, and Claude Code figures out the implementation, runs commands, and confirms results. I’ve been using this for partnership workflow automation, building scripts that pull data from multiple sources. The advantage is speed on straightforward tasks, though you need to review changes carefully since it’s making actual modifications to your data.

2. Gemini CLI

Gemini CLI brings Google’s AI assistance into your terminal, letting you ask for what you want in plain English and get help generating commands and troubleshooting steps. It excels at generating shell commands, explaining error messages, and creating small scripts. For partnership professionals working with Google Cloud Platform or Google Workspace integrations, this is particularly valuable since it understands GCP-specific commands. Treat generated commands like a junior engineer’s draft and review before running.

3. OpenAI Codex

OpenAI Codex translates natural language into code across a dozen programming languages. Trained on billions of lines of public code, it powers GitHub Copilot and aims to make programming more accessible. I haven’t spent as much time with this one compared to Claude Code, but it represents OpenAI’s approach to the same problem. The limitation is that it generates suggestions rather than executing autonomously, operating more as intelligent autocomplete than an agent. Worth exploring if your organization standardizes on OpenAI’s ecosystem.

Bottom line: Start with Claude Code if you want an autonomous agent that can handle multi-step tasks. Use Gemini CLI if you’re working heavily with Google Cloud Platform. Consider Codex if you’re locked into the OpenAI ecosystem.

What workflow would you automate first if you could describe it in plain English and have an AI build it?

The holiday break provided a chance to step back for some perspective on where the AI landscape is headed in 2026. I spent some time neck-deep in vibe coding, testing different platforms to see which actually deliver on productivity promises and which are dead ends. What became clear is that Google’s stack offers the richest, most accessible set of tools for getting started. The barrier to entry has dropped significantly, which changes what partnership professionals need to focus on.

Here are three things that caught my attention over the break, along with three entry-level vibe-coding tools from Google worth exploring if you’re ready to build rather than just prompt.

1. Claude Code built in an hour what a Google team built in a year

Google Principal Engineer Jaana Dogan reported that Claude Code developed a distributed agent orchestrator in one hour, a task that took her team a year to build. This anecdote supports a broader trend: Anthropic’s CEO states that 90% of the company’s code is AI-generated, and Google and Microsoft report similar figures. If there’s any truth to the agentic AI hype from 2025, it’s that AI is taking over coding.

Dogan later clarified her comments, saying the code was “elementary” rather than production quality, but the comments are telling nonetheless. Keep in mind that while Google develops Gemini, it is also a major investor in Anthropic, the developer of Claude Code.

For partnership professionals, this means software development will increasingly be the focus of AI optimization in 2026. Partnership teams should consider custom tools and integrations rather than waiting for vendors to build what they need. The gap between identifying a workflow problem and creating a solution is narrowing, shifting the skill requirements from technical implementation to problem and requirement definition.

2. Meta acquires Manus for $2B, accelerating agent development

Meta is acquiring Manus, a Singapore-based AI startup that launched in spring 2025 and quickly scaled to millions of users and over $100M in annual recurring revenue. Manus focused on AI agent capabilities rather than on developing frontier models, making it an attractive acquisition target. Meta plans to keep Manus running independently while integrating its AI agents into Facebook, Instagram, and WhatsApp.

Meta will be embedding agents into their social platforms; expect to see similar integration in CRM and partnership management tools soon. The question isn’t whether agents will be part of your workflow, but whether you’ll be ready to use them effectively when they arrive. Keep an eye on Genspark, a Manus competitor, and now an attractive acquisition target.

3. A practical 10-week plan to build AI fluency

Looking for an AI New Year’s resolution? Nathaniel Whittemore at the AI Daily Brief created a self-guided 10-week challenge that cuts through AI theory and focuses on actually doing the work. Each weekend challenge tackles a concrete project: model mapping, deep research, data analysis, visual reasoning, automations, context engineering, and building an actual AI-powered app. The modules are designed to be completable in a few hours and immediately useful.

This resonates with what I’ve been seeing in the partnership community. Too many people are stuck in the learning phase, reading about AI rather than building with it. The gap between understanding AI concepts and implementing AI workflows is significant, and the only way to close it is through hands-on projects. If you’re setting professional development goals for 2026, this provides a structured path that will still matter six months from now.

3 Things I Think You’ll Like: The Latest Vibe Coding Tools from Google

I spent the holiday break testing various vibe coding tools, and Google’s stack emerged as the most comprehensive and an excellent place for partnership professionals looking to get started. These are three entry-level options that provide different paths into building with AI rather than just using it. Next week, I’ll cover the more advanced tools in the Google ecosystem.

1. Stitch - Design with AI

Stitch is Google’s AI-powered design tool for creating visual content quickly, where you can design your complete user interface before worrying about code. This tool is helpful if you’re building tools that need strong user interfaces rather than just backend automation. You can design the screens, buttons, and workflows visually, then use them as a blueprint for implementation in AI Studio, or hand them off to a developer.

2. Google Opal

This is Google’s no-code platform for building AI mini-apps using natural language. You describe what you want to create, and Opal builds multi-step workflows that chain prompts, model calls, and tools together, handling hosting and deployment automatically. I haven’t spent as much time on this one, but it appears to have the lowest barrier to entry. The no-code approach lets you start building without a technical background, making it useful for simple workflow automations, such as generating partner onboarding checklists or building meeting agendas based on partner history.

3. Google AI Studio

This is the most sophisticated tool on this list, but still very approachable and a great way to get started with vibe coding. Google AI Studio is a web-based development environment for building generative AI applications, primarily leveraging Google’s Gemini family of models. It allows you to quickly prototype, test, and deploy AI models with a free tier for getting started and integrates seamlessly with Vertex AI for production-grade scaling. The advantage over other platforms is that you’re working directly with Gemini models, which means you get the latest capabilities as they’re released, though you’re locked into Google’s ecosystem.

Bottom line: Start with Opal if you want to test whether vibe coding is worth your time. Move to AI Studio if you’re ready to build production workflows. Use Stitch if you want to design interfaces before worrying about functionality.

Which of these tools appeals the most to you?

The conversation around AI has shifted from whether to adopt it to how to build with it. Google’s release of its Deep Research agent via API this week is significant. Previously, advanced research capabilities lived behind chat interfaces. Now, users can embed that intelligence directly into applications and workflows. Meanwhile, OpenAI quietly rolled out skills support across ChatGPT and Codex CLI, following Anthropic’s lead. The pattern is clear: AI capabilities are becoming building blocks you compose, not finished products you adopt.

Here are three things that caught my attention this week, along with three training courses worth exploring if you’re ready to understand how AI can work for you.

1. Google releases Deep Research agent via API

Google made its Gemini Deep Research agent available via its API, enabling users to embed autonomous research capabilities into their workflows. This is a significant development. Previously, these deep research capabilities were only accessible through the chat interface.

The agent, powered by Gemini 3 Pro, is optimized for long-running context gathering and synthesis. It’s designed to reduce hallucinations and maximize report quality by iteratively planning investigations and navigating complex information landscapes.

The tool is already being used in financial services for due diligence and biotech for drug discovery. Google plans to integrate it into Search, NotebookLM, Google Finance, and the Gemini App. The practical application for partnership work is straightforward. Instead of copy-and-paste AI, you could build workflows that synthesize intelligence from multiple sources with deeper, more sophisticated research. The question isn’t whether this capability will be helpful, but how long it takes partnership teams to learn to use it.

2. OpenAI quietly adopts skills in ChatGPT and Codex CLI

OpenAI has quietly adopted the “skills” mechanism, previously seen with Anthropic, integrating it into both ChatGPT’s Code Interpreter and their Codex CLI tool. I’m firmly in the “team skills” camp as a candidate technology to be broadly adopted, enabling subject-matter expertise to be refined and exposed through chat interfaces. Think of it as a simple way to train your chatbot on your unique workflows and area of expertise.

The partnership angle here is about standardization. If skills become the way we extend AI capabilities, partnership teams need to start thinking about what partnership-specific skills would look like. For example, you could create a skill that instantly creates a partner performance report or a QBR presentation.

What knowledge or processes could you codify that would make AI more useful for partnership work?

3. OpenAI’s head of Codex says typing speed is the AGI bottleneck

Alexander Embiricos, OpenAI’s head of product for Codex, states that human typing speed is a major bottleneck to achieving Artificial General Intelligence. This is because humans are currently required to write prompts and validate AI’s work. Embiricos believes that progress will accelerate significantly when AI agents can review and validate their own work, leading to “hockey stick growth” in productivity and ultimately, AGI.

This point reinforces something I’ve been calling out for a while. I posted earlier about the benefits and some voice input tools. I would say probably 70% of my computer input these days is through voice. The constraint isn’t just typing speed; it’s the ability to get ideas out of your head.

We’re still operating AI through text prompts because that’s how we’ve always interacted with computers. But if agents can validate their own work, the interface shifts from precise instructions to high-level guidance. This changes what partnership professionals need to learn. Instead of mastering prompt engineering, you need to understand how to set objectives and evaluate outcomes.

4. BlackRock changes hiring priorities around AI skills

BlackRock’s head of talent acquisition, Nigel Williams, states that AI is significantly changing hiring priorities. Applicants are now expected to embrace AI, demonstrating fluency with various tools, a questioning mindset, and basic prompt engineering skills. Curiosity and the ability to critically evaluate AI outputs are crucial.

I have been expecting to see more of this in the press and anecdotally through people working through the job application process. The interesting part is the emphasis on critical evaluation rather than blind acceptance. This isn’t about using ChatGPT; it’s about knowing when AI is useful and when it’s not.

For partnership professionals, this signals that AI fluency is becoming table stakes. Whether you’re working to achieve success where you are or looking for your next role, your AI knowledge and skills will become increasingly important.

3 Things I Think You’ll Like: AI Training Courses

If you’re ready to expand your AI skills, these courses provide structured learning paths. The first is free and comprehensive, the second is paid with hands-on project work, and the third is specifically designed for partnership professionals. All three focus on practical implementation rather than theoretical concepts. I’m working my way through the first two, and attending the 3rd this week.

1. Microsoft AI Agents for Beginners

This GitHub repository offers a multi-language course with 12 lessons on building AI Agents. Each lesson includes Python code samples using Azure AI Foundry and GitHub Model Catalog, along with a written lesson and a video. It encourages users to star and fork the repo for code access and provides a Discord channel for support (if that sounds daunting, no worries, they walk you through it).

The advantage is that it’s free and covers the fundamentals without assuming you’re already a developer. The Python code samples give you something concrete to work with rather than just conceptual frameworks.

2. AI Engineer Core Track: LLM Engineering, RAG, QLoRA, Agents

This 8-week course, led by Ed Donner, focuses on mastering Generative AI and Large Language Models through practical, real-world projects. Students will learn to build advanced Generative AI products, experiment with over 20 groundbreaking models, and master state-of-the-art techniques like RAG, QLoRA, and Agents.

The course covers developing proficiency with platforms such as Hugging Face, LangChain, and Gradio, and deploying AI products to production with polished user interfaces and advanced capabilities. This is more comprehensive than the Microsoft course but requires a paid subscription and a significant time commitment.

3. Partner Success Gen AI Prompting Certification

AchieveUnite offers a certification program specifically designed for partner success professionals looking to build Gen AI prompting skills. This is the most directly applicable option if you’re working in partnerships and want training that speaks your language rather than forcing you to translate developer-focused content.

The certification approach provides credential validation alongside the learning, which is important if you need to demonstrate AI competency to your organization. The focus on partner success contexts means the examples and use cases should map more directly to the work you’re actually doing.

Bottom line: Start with the Microsoft course if you want free, hands-on experience with agents. The Udemy course makes sense if you’re ready to go deep on production deployment. The AchieveUnite certification is the right choice if you want partner-specific training with a credential at the end.

What’s your biggest barrier to understanding how AI can work for you - time, complexity, or figuring out where to start?

The conversation about AI agents has moved from “what” to “how.” AWS’s re:Invent announcements this week showed that we’re entering a phase where agents aren’t just demos or ChatGPT wrappers. They’re production infrastructure with actual marketplaces and deployment frameworks. Meanwhile, Google released Workspace Studio, which lets you build agents without code. The pattern across all major platforms is clear: everyone wants you building agents on their infrastructure.

Here are three things that caught my attention this week, along with three AI workflow tools worth exploring if you’re ready to build agents instead of just prompting them.

3 Things for Tuesday, December 09, 2025:

1. AWS shifts from chatbots to frontier AI agents

AWS re:Invent 2025 marked the official end of the chatbot hype cycle, replaced by what they’re calling “frontier AI agents” that can work autonomously for days rather than single exchanges. The headline product is Amazon Bedrock AgentCore, a managed service that handles state management and context retrieval so you don’t have to build that infrastructure yourself.

I’ve been working with Bedrock AgentCore this week, figuring out how to take vibe-coded AI agents into production. The difference between prototype agents and production agents turns out to be significant. Production agents need to handle errors, maintain state across sessions, manage security, and integrate with existing systems. AgentCore addresses most of these concerns, though the learning curve is steeper than I expected.

2. Google launches Workspace Studio as no-code agent builder

Google released Workspace Studio on December 3rd, enabling anyone to create AI agents that automate workflows across Gmail, Drive, Sheets, and third-party apps like Asana, Jira, and Salesforce. The pitch is simple: describe what you want in plain language, and Gemini builds the agent. No coding required.

The practical applications for partnership work are immediate. You can build agents that track partner engagement across email and calendar, generate weekly partner activity summaries, or flag deals that need attention based on velocity changes. The integration with Workspace apps means these agents have context about your actual work, not just access to APIs. The limitation is that you’re locked into Google’s ecosystem, which is fine if you’re already there but problematic if you need cross-platform workflows. I’m testing this against similar agent builders from Microsoft and OpenAI to see which handles partnership-specific workflows best.

3. The bento box metaphor for professional services transformation

This piece from Platforms, AI, and the Economics of BigTech uses the bento box as a metaphor for how AI is reshaping professional services. The core idea is that product constraints determine industry architecture. Professional services have been structured around human speed, document-based evidence, sampling, and fragmented data. AI removes many of these constraints, enabling the entire industry architecture to be reshuffled.

The film editing comparison is apt. When digital editing replaced physical tape cutting, the entire production workflow changed. New roles emerged, old constraints disappeared, and different kinds of stories became practical to tell. We’re seeing similar shifts in partnership management, though we’re still early in understanding what that means.

The constraints around processing partner data at scale are disappearing faster than our ability to figure out what to do with that capability. Most partnership teams still operate on quarterly review cycles and manual data collection; not because automation technology doesn’t exist, but because workflows haven’t kept pace with the infrastructure.

The work changes when you rebuild the processes, not just when tools become available. This isn’t about replacing partnership managers with AI. It’s about partnership managers doing fundamentally different work because the constraints that defined their workflows have disappeared.

I’ve added Reshuffle to my holiday reading list.

3 Things I Think You’ll Like: AI Workflow Tools

Every platform now offers a way to build AI agents rather than just interact with chatbots. These tools let you create automated workflows that reason and handle multi-step processes. The question isn’t whether you should explore agent builders; it’s which platform makes the most sense for your existing infrastructure and partnership workflows.

1. OpenAI Agent Builder

OpenAI launched Agent Builder with considerable fanfare a couple of months ago, then went relatively quiet. The tool lets you build agents on top of ChatGPT’s API with predefined actions and reasoning capabilities. The advantage is integration with OpenAI’s ecosystem if you’re already using ChatGPT for your team. The limitation is that agent capabilities remain somewhat constrained compared to what you can build with lower-level APIs. I’m curious whether OpenAI will invest more here or if this was a response to competitive pressure that they’ve since deprioritized.

2. Microsoft Copilot Studio

Microsoft’s no-code agent builder for Teams environments. If your organization runs on Microsoft 365, Copilot Studio provides the most natural integration path for building partnership workflow agents. You can create agents that monitor Teams channels, pull data from SharePoint, and trigger actions in Dynamics. The learning curve is moderate, capabilities are solid, and enterprise security is already addressed.

3. Google Workspace Studio

I covered this in the news section, but it’s worth including here as a direct competitor to Copilot Studio. Workspace Studio takes the same no-code approach but for the Google ecosystem. The advantage over Microsoft is tighter integration with Gemini’s reasoning capabilities and a more straightforward interface for specifying what you want; the drawback is the limited number of integrations.

Bottom line: The choice between these platforms depends more on your existing infrastructure than the tools themselves. If you’re in Microsoft 365, use Copilot Studio. If you’re in Google Workspace, use Workspace Studio. If you’re platform-agnostic or need custom integrations, OpenAI’s Agent Builder may offer more flexibility at the cost of complexity.

What’s stopping you from building your first AI agent for partnership workflows?

I try to avoid covering every LLM release since most advances are incremental and don’t always translate into immediate productive gains for partner leaders. However, the press coverage around Google’s Gemini 3 model this week, sandwiched between Anthropic’s Claude Opus 4.5 and OpenAI’s ChatGPT 5.1 releases, reminded me to revisit a few features I’ve either stopped using or haven’t tried yet. I’ve been using Claude as my primary daily driver, but Gemini has some unique features worth exploring further. Here are three things that caught my attention, along with three AI partnering assistants worth checking out.

📰 What’s Happening with Gemini

1. Generative UI: Custom interfaces on the fly

Gemini’s new generative UI capability, available through dynamic view in the Gemini app and AI Mode in Search, creates complete interactive interfaces in response to prompts. Rather than just generating text or code, it builds functional web-based tools, dashboards, and simulations tailored to your specific query.

I used this capability to create a partner QBR dashboard from a single prompt. The interface included interactive charts, data filtering, and navigation, all generated automatically. The output was impressive and can produce ready-to-use results easily. The challenge now is connecting this UI capability with real-time data access. I’ve been experimenting with Claude using skills and MCP servers, but this UI capability may leapfrog that work. While working on this article today, I realized I now have access to the HubSpot connector for Gemini, so I’m deep in the weeds connecting real-time HubSpot data with Gemini. Stay tuned for a deeper dive.

2. Nano Banana Pro: Image generation with actual reasoning

Google also released Nano Banana Pro (Gemini 3 Pro Image), their new state-of-the-art image generation model built on Gemini 3 Pro. What sets this apart from other image generators is its use of Gemini’s reasoning capabilities to create images with accurate, legible text in multiple languages and maintain consistency across up to 14 input images.

For partnership professionals, this is genuinely useful. You can create infographics with readable text, generate multilingual partner materials without separate translation steps, and maintain brand consistency across marketing assets by blending multiple reference images. I tested it by creating a word cloud image from the Google Blog. The ability to combine elements while maintaining visual consistency means you can create collateral quickly without waiting for design resources. It’s still not perfect, as some text misses the mark, but it’s the most capable tool I’ve seen for creating professional visuals with substantial text. If you’re still waiting for your marketing team to generate content for your partner communications, it’s time to take matters into your own hands.

3. Scheduled Actions: Set-it-and-forget-it AI tasks

Gemini’s scheduled actions feature lets you set up recurring or one-time automated tasks. You can ask Gemini to summarize your calendar and emails every morning, generate content ideas weekly, or track partner news on specific dates.

This isn’t specific to the Gemini 3 release, but it’s worth highlighting for partnership workflows. I’ve set up weekly news summaries and daily briefings. The setup is straightforward; you tell Gemini what you want and when, and it handles the rest through calendar integration. One reason I overlooked this functionality is that I didn’t see where you create these recurring tasks. It turns out you just have to ask for it. For partnership professionals, consider recurring partner engagement reports, weekly industry news summaries specific to your partners, or automated reminders to review partner performance data. I’ll incorporate this into the HubSpot demo I mentioned above.

🛠️ Three Things I Think You’ll Like: AI Partnering Assistants

The partnership space is starting to see purpose-built AI tools designed specifically for partnership professionals. These aren’t general AI assistants adapted for partnership work—they’re built from the ground up to address partner management workflows. Here are three emerging in this space:

1. PartnerBOT.ai - From the Partner Architects Group, this tool focuses on partnership-specific workflows and knowledge. It’s designed to understand partnership terminology and structures rather than requiring you to explain partnership concepts to a general-purpose AI.

2. Bluethread Partnership Revenue Lab 3.0 - Built as a custom GPT from BlueThread, this focuses on partnership revenue optimization and strategy. It’s accessible through ChatGPT and leverages BlueThread’s partnership frameworks.

3. PartnerAssistant.ai - Positioned as “AI-powered partner revenue, on autopilot,” this tool emphasizes automation of partnership workflows and revenue tracking. It aims to handle repetitive partnership tasks systematically.

Bottom line: These purpose-built tools are worth exploring if you’re serious about AI-enhanced partnership workflows. They won’t replace your partnership expertise, but they can accelerate common tasks by understanding partnership context without extensive prompting. Each takes a different approach—from strategic frameworks to workflow automation—so the best fit depends on your specific needs.

What AI features have you been ignoring that might actually be useful for your partnership work?

Start with the Problem Statement

The framework begins by identifying tasks that consume time and could be automated. Meeting notes offer a concrete example. Partnership professionals often run back-to-back calls—discovery conversations with prospective partners, regular cadence meetings with existing partners, and deal facilitation calls. The context switching between these meetings makes it difficult to capture and synthesize information. Note-taking tools like Fellow, Read.ai, or Fathom can handle note-taking, but the real value comes from designing what happens after the notes are captured. The framework pushes beyond just recording meetings to defining the desired outcome: synthesized action items, CRM updates, or follow-up communications.

Mapping Inputs, Outputs, and Process Flow

The canvas breaks down workflows into three components. Inputs are the data input feeding the workflow—meeting transcripts, CRM data, and partner communications. Output defines what the workflow produces—summarized notes, updated records, automated emails. The process flow connects these through specific tools and decision points. This structured approach prevents the common trap of selecting tools before understanding the actual workflow requirements.

Security and Compliance Considerations

The framework includes deliberate attention to data sensitivity. When pulling pipeline data with dollar values, close dates, and contact information into third-party workflow tools, partnership professionals need to document what data they’re exposing and verify tool certifications like SOC2 or HIPAA compliance. This preparation becomes critical when seeking internal approval from IT or steering committees, particularly since partnership teams often operate with less oversight than other functions.

Practical Application

The canvas works as a design exercise before implementing any automation. Taking one high-volume task—partner onboarding sequences, support requests, or post-meeting follow-ups—and working through each section of the framework reveals whether automation will genuinely reduce workload or shift it elsewhere. The goal isn’t automation for its own sake, but deliberately removing noise from your inbox and calendar.

Tools Mentioned:

• Meeting transcription: Fellow.app, Read.ai, MeetGeek, Google Meet, Teams, Zoom

• Workflow design: PartnerFlow Automation Canvas framework

The framework’s value is in forcing systematic thinking before tool selection. Partnership professionals have always been expected to accomplish significant goals without dedicated ops support or substantial technology budgets. The canvas provides a method for strategically deploying limited resources to create measurable impact rather than experimenting with tools in the hope that something sticks.