You can paste in a transcript, ask it to find content ideas, and get back a nice clean list of topics. The list may sound useful. It may even be organized into proper hooks, posts, and videos.

And each one of those ideas could belong to almost anyone.

That was the problem I kept running into. The source material was good. The AI could summarize it pretty well. It could find repeated themes and pull out quotes.

…but what it couldn’t do was determine which parts mattered to my audience or my business.

AI is pretty stupid when it doesn’t know you

AI knows A LOT. However, it knows almost nothing about YOUR specific business unless that information is available in the conversation, project, or workspace you’re using.

It doesn’t automatically know what you sell. It doesn’t know who you serve, what those people are dealing with, what they already understand, or what kind of work you want to be known for.

So when you ask it to mine a transcript for content ideas, it has to fill in those gaps somehow right? So, it guesses.

And those guesses can sound perfectly reasonable because AI is good at producing plausible language. You may get ideas like:

• 5 ways to save time with AI

• Why consistency matters in content creation

• How to work smarter with automation

Those are content ideas that are technically right. They just don’t reflect a particular audience, point of view, or business.

The source material is only half the input

A transcript tells AI what someone said. An article tells it the authors point of view. A YouTube video gives it stories, examples, claims, and questions.

None of those sources tells AI what matters to you.

Imagine you give the same interview transcript to 3 businesses: an executive coach, an accounting firm, and a software company. They should find different ideas in it because they serve different people and solve different problems.

The words in the transcript haven’t changed. The filter has.

This is why content mining is really a judgment problem. Finding what was said is the easy part. Deciding what is relevant, useful, credible, and connected to work you can support requires context.

The context depends on what you’re trying to do

I don’t think there is one master context file that should follow AI into every task.

If you’re reviewing a contract, the useful context may include the agreement, the parties, the scope, and the questions that need legal review. If you’re preparing for a sales call, you may need the account history, recent messages, the offer, and the decision you want to reach.

Content mining needs a different package.

I use 3 context files for this job:

1. Business and offer context

2. Audience context

3. Voice and content context

This isn’t magic. These foundational files give AI the criteria it needs before it starts selecting ideas that you care about.

Business and offer context

This file explains what your business does and what people can get from you.

It should cover the problems you help solve, the offers you have now, the outcomes those offers support, and the boundaries around what you do.

The point isn’t to force a sales pitch into every piece of content. Instead, it helps AI recognize ideas that you or your business can talk about with credibility.

Without it, AI may find an interesting topic that leads your audience toward a problem you don’t solve. You end up publishing content that gets attention and creates no real connection to the rest of your work.

Audience context

This file explains who you’re talking to.

Roles and demographics can help, but I care more about the situation someone is in. What are they trying to do? Where do they get stuck? What have they already tried? What questions do they ask? What words do they use when they describe the problem?

For my own content, I’m usually talking to capable founders, consultants, operators, and small teams. They may already use ChatGPT or Claude. They don’t want to become AI engineers. They want the work to move without every decision waiting on them.

That context changes what AI should extract from a transcript.

A passing comment about repeating the same explanation every week may be more useful than the a speaker’s polished 10-minute framework. AI wouldn’t know what to prioritize unless it understands the audience and the problem.

Voice and content context

This file explains how you think and what makes a piece of content feel like yours.

It can include your point of view, recurring themes, useful stories, words you naturally use, phrases you avoid, content formats, and the standards a draft has to meet before you publish it.

Voice is more than tone. It affects selection.

If your work focuses on human judgment, you may notice a moment where someone describes making a hard call with incomplete information. Someone focused on productivity may select a completely different moment from the same source.

The voice and content file helps AI look for ideas that fit your body of work, instead of dressing up generic topics in your writing style after the fact.

What changes when AI reads all 3

Let’s say you paste in a transcript where someone talks about spending hours every week rewriting work their team sends them.

Without context, AI might suggest:

How to improve team communication

With the 3 files, it may recognize a more specific idea:

Your team keeps bringing decisions back to you because the quality standard still lives in your head.

The second idea connects the source to a real audience problem. It also connects to work I can help with: extracting the decisions, examples, and standards that make a handoff usable.

The AI didn’t become smarter between those 2 attempts. It had a better filter.

You can use the same system when you’re learning from someone else’s article or video. Pull out the questions, tension, examples, and claims that connect to your audience. Then add your own analysis and perspective.

Build the files once, then keep iterating

Your first version won’t be perfect. But that’s fine and 100% expected.

The most useful part about this is that you no longer have to rebuild the context from scratch every time you open a conversation. You give AI the 3 files, add the source you want to mine, and tell it the outcome you want achieve.

Then pay attention to where things go wrong.

If it keeps suggesting topics that don’t connect to your offers, improve the business context. If the ideas don’t hit the mark regarding what your audience cares about, improve the audience file. If the drafts sound too polished, your voice and content context needs more real examples and clearer boundaries.

The files will get better as you use and update them.

I put together a free guide with all 3 templates, copy-ready prompts, public Markdown reference files, and a ZIP download you can keep with your own business files.

Get the Content Idea Context Kit

Keep building.

-Tam

In this issue: A third way to think about AI that most people haven't considered. The four layers that separate a clever prompt from encoded judgment. And a mirror to find the 3-4 skills already living in your head that you've never written down.

You made at least a dozen judgment calls today that nobody saw.

You pushed back on a client’s direction instead of letting them run with it. You read a room and knew the deal wasn’t going to close before anyone said it out loud. Someone was telling you a story, circling around the real problem, and you asked the one question that cut straight to it. Something happened in a meeting that felt like progress but you could tell it was a detour that’ll cost you two weeks.

You’ve been making calls like these for years. Probably a decade or more. You’re fast at it now. So fast that most of the time you don’t even register you’re doing it. It just happens. You listen, you process, you decide, and you move on.

But here’s the thing about that speed: it made the thinking invisible. Not just to everyone else. But also to you.

You can’t hand it off because you can’t explain it. Every time you step away, something slips that wouldn’t have if you’d been in the room.

Your team doesn’t lack talent. They lack access to the pattern you’re running. And you can’t give them access to something you can’t see yourself.

So what do you do with that? Most people, at some point, try to solve it with AI. Which is the right impulse. But the approach they use determines whether it works or doesn’t.

The two defaults

Most people I talk to think about AI in one of two ways.

The first is the chatbot. Ask a question, get an answer. This is how most people start. You type something into Claude or ChatGPT, it responds, and you evaluate whether the response is useful.

Sometimes it is. Often it’s generic. The model doesn’t know your team, your context, your priorities, or the twelve things you considered and rejected before arriving at your actual answer. It’s working from a blank slate every time.

The second is automation. If X happens, do Y. This is where most “AI workflow” content lives. Connect your tools, build a pipeline, trigger actions automatically.

It works beautifully for things that follow predictable rules. It falls apart the moment the situation requires judgment.

And the work that matters most, the work that made you the person everyone depends on, is almost entirely judgment.

The chatbot is generic. The automation is brittle. Neither one can carry the thing that actually makes you good at your job.

There’s a third approach. I haven’t seen many people talk about it, but it’s the one that changed how I think about all of this.

Something else entirely

A judgment container.

Think about it this way. A chatbot gives you its best guess at what you might want. An automation does exactly what you told it, whether or not the situation calls for it. A judgment container holds the shape of how you actually think, your methodology, your criteria, your guardrails, and applies it consistently (ideally) while you review the output.

That’s the difference. A judgment container does what you would do, because you taught it how you think, and then it waits for you to confirm before anything happens.

It’s closer to what it feels like to have a really smart colleague who’s watched you work long enough to know your patterns. Someone who routes things the way you’d route them, flags what you’d flag, cuts what you’d cut. Not perfectly. But well enough that your job shifts from doing the work to reviewing it.

The interesting part is that building one forces you to do something most people never do: articulate the judgment that’s been running in your head on autopilot.

And that turns out to be the hardest part. It’s not the technology or the setup. It’s the act of looking at a decision you’ve made a thousand times and breaking it down into the sequence of questions you actually ask, in what order, with what criteria.

I’ve spent many years making a particular set of judgment calls every week before I tried to write them down. When I finally did, I realized I couldn’t get past the second step without saying “it depends.” And “it depends” was covering about fifteen decision rules I’d never consciously separated.

Most people discover the same thing. They never had a reason to pull the judgment apart. It just worked. Until it needed to scale.

If you’ve used Claude or worked with AI agents, you may have heard the term Agent Skill. Anthropic defines skills as instructions that teach AI how to complete specific tasks in a repeatable way.

And many people build them that way. Procedures. Checklists. Step-by-step instructions for formatting a report or processing a form.

Those are fine for predictable work. But the skills that actually change how you work are the ones that carry your judgment. Your decision criteria. Your methodology.

The “it depends” logic that lives in your head and nowhere else. That’s what turns a Agent Skill from a glorified template into something that thinks the way you think.

A skill is how you take the judgment out of your head and put it somewhere loadable.

The anatomy of a skill worth building

Every time I've turned invisible judgment into something loadable, the same four layers show up. When one is missing, the thing breaks. When all four are present, it works at a level that surprises people, including me.

(The first time I saw my own decision process run without me, I spent ten minutes checking whether it had actually gotten the routing right. It had. That was a strange but very exciting morning.)

Here’s what I found those layers to be:

1. Methodology. The decision sequence you actually run. The specific questions you ask, in the specific order you ask them, when this type of work lands on your desk.

Most people have never written this down. It’s the most valuable layer and the hardest to see from the inside.

2. Context. The state of your world. Who your people are, what they’re working on, what the current priorities are, what’s already been decided. No AI has any of this unless you give it. Without context, even perfect methodology produces generic results.

3. Workflow. The deterministic shape of the work. What goes in, what comes out, in what order. Every step has one job, with a defined input and a defined output. Without this structure, even good judgment produces inconsistent results because there's no container for it to run in.

4. Guardrails. The things you would never do, even if they looked helpful. This is where many AI projects fail and where the best ones are strongest.

If you’re missing one of these layers, you have either a clever prompt or a brittle script. Both are worse than no skill at all, because they create the illusion of a system without the reliability of one.

The mirror

You have three or four of these skills living in your head right now.

They’re the work where your quality drops when you’re tired. The thing your team keeps asking you to do “the way you do it.” The judgment call you’d never fully delegate but would love to stop carrying alone every week.

You’ve called it “just how I work.” Or “the part I can’t really explain.” Or “the thing that takes me 45 minutes but would take anyone else three hours, and they’d still miss something.”

That language is what keeps it stuck in your head. It sounds like a description of talent. It’s actually a description of a system that was never written down.

The hardest part, honestly, is seeing your own patterns clearly the first time. The layers feel obvious when someone else names them back to you. But they’re nearly invisible when you’re the one running them.

Next week, I’ll show you what it looks like when you take one of those and actually build it. A real skill I shipped, with the before and after, the decisions I made, and the parts that surprised me.

In the meantime, try this. Paste the prompt below into Claude or ChatGPT. It will interview you about your week, spot where your judgment is doing invisible work, and name your skill candidates back to you.

You're going to help me find where my professional judgment is doing invisible work. Interview me about my week, then tell me which tasks are "skill candidates" worth encoding into AI.
Start by asking me to walk through yesterday, hour by hour. Don't let me summarize. Push for specifics. As I talk, listen for these signals:

Someone needed my opinion, not just information
A less experienced person would have handled it differently and probably gotten it wrong
I did something on autopilot that actually requires years of pattern recognition
I said "it depends" (that phrase almost always means hidden judgment)
I translated something complex into something simple for someone else
I've explained this same thing dozens of times and wish it could just be done my way

When I say "I just do it," push back. Ask things like:

"Walk me through the first 30 seconds. What do you look at first?"
"What would a new hire get wrong about this?"
"When was the last time someone did this without you and it went sideways?"

After we talk through my day, reflect back 2-3 candidate tasks where my expertise is doing hidden work. For each one, name the specific judgment involved (not "you use your experience" but the actual decision points), how often I do it, and what goes wrong when someone else tries.

-Tam

Nobody compiled that list. Nobody spent 20 minutes before the meeting pulling tabs together. My cat bot Phil swept everything overnight, read the context, and posted the rundown at 7am. (Yes, I named my AI agent after a cat)

The meeting had direction before anyone said a word.

In this issue: A mental model for why platform consolidation keeps failing, a concrete 5-step approach for pointing agents at the tools your team already uses, and a free prompt that walks you through the whole framework. No migration required.

What That System Actually Looks Like

The agent pulls from four sources every morning:

Slack channels where the team talks. It reads active threads, pulls out anything that needs a decision or follow-up, and links back to the original conversation.

Google Docs that were created or edited in the last 24 hours. It reads the full content and extracts actionable items.

Huddle transcripts from the previous day’s meetings. Unresolved items, follow-ups that got mentioned but not captured anywhere.

A Decisions Database in Notion that tracks open items, blockers, things that still need someone to make a call.

All of it gets synthesized into a single Slack message with grouped categories (client pipeline, sales, internal ops), each with an owner and a clickable link to the source.

The other piece I built in: it’s interactive. Agendas are usually static. One person owns it, everyone else shows up and reacts. But when you surface the agenda in a space where people already work (Slack, in our case), and it’s that easy to add, edit, or delete items, people actually use it.

Reply in the thread saying “add the pricing review,” and the agent adds it. Say “remove item 3,” and it’s gone. The agenda stays alive all day. By the time we sit down, the pre-work is already baked in.

The agent bridged the data. It didn’t centralize it.

Three APIs, one agent, one output. That’s the whole thing.

The Centralization Lie

Here’s the conversation I’ve had a dozen times. A founder or ops lead says: “We need to get everything into one platform. Notion. Monday. Asana. Pick one, migrate everything, and then we’ll finally have visibility.”

They spend weeks designing the perfect workspace. They migrate data. They write documentation. They run training sessions.

Six months later, half the team is back in spreadsheets. The CRM has stale records. The project tracker has tasks from Q2 that nobody closed.

This pattern has a number attached to it. CRM implementations fail 55-75% of the time, and the primary driver is adoption, not technology. The platform worked fine. The people didn’t use it.

Every “let’s get everything into one platform” project fails because people don’t adopt platforms they weren’t already using. You can build the most elegant Notion workspace in the world. If your sales team lives in email and your designers live in Figma and your developers live in Linear, they’re not moving. They’ll nod in the all-hands. They’ll bookmark the link. And then they’ll go back to the tool that’s already open.

The assumption behind centralization is that fragmentation is the problem. The data was never the problem. The assumption that humans should move it was.

How the Agent Model Works

Slack holds conversations. Drive holds documents. Your CRM holds pipeline. They’re already where they belong.

The agent model starts from a different premise: leave everything where it is and build a reader that can pull from all of it.

Your Slack workspace has an API. Google Drive has an API. Notion has an API. Your CRM has an API. Every tool your team already uses has a programmable surface. An agent can authenticate to each one, read what’s relevant, and synthesize it into a single output, without anyone changing a single habit.

The mental model shift is subtle but it changes everything downstream. You stop asking “where should we store this?” and start asking “can an agent read it where it already lives?” The answer is almost always yes.

One caveat, because I don’t want to oversell this: agents still need a map. Your data can stay scattered across tools, but within each tool, the agent needs to know where to look. Transcripts go in a specific Drive folder. CRM fields that matter actually get filled. Docs live inside a known workspace.

You’re organizing within each tool so the agent can reliably find what it needs. Think of it as organized chaos: the chaos of multiple platforms is fine, as long as each platform has enough internal structure for an agent to navigate it.

The good news is that most teams already have this partially done. You already have a Drive folder for client docs. You already have a Slack channel for each project. The bar is lower than you think.

And here’s the compounding insight from building this. Once you wire up one connection (say, the Notion API for the Decisions Database), that connection is live for every future agent you build.

Next time you need Notion data for something else, the infrastructure is already there. Same with Drive. Same with Slack. You build these small connections and then wiring up new agents that pull from multiple sources becomes almost trivial.

Each integration is an investment that pays out across every agent you build after it. The second agent is easier than the first. The fifth is almost free.

This is the “tools change, architecture stays” principle from a previous issue. The specific APIs might evolve. The pattern of building connectors and pointing agents at existing data sources is durable.

What to Point Agents At First

If you’re thinking about trying this, here’s a sequenced approach. No coding required (though it helps). The goal is to identify the highest-value target before you build anything. (And if you want a reminder of why starting small beats elaborate systems, I’ve written about that before.)

1. Map where your team’s information already lives.

Not where it should live. Where it actually lives right now. For most teams, the answer is 3-5+ tools. Slack, email, a docs platform, a project tracker, maybe a CRM. Write them down.

2. Identify the manual retrieval loop.

Where does someone on your team spend time pulling information from one tool into another? The person who compiles the weekly report. The manager who checks three dashboards before a meeting. The founder who reads Slack channels every morning to figure out what’s going on. That loop is your target.

3. Check for an API.

Most tools your team already uses have one. Slack, Google Workspace, Notion, HubSpot, Salesforce, Linear, Asana, Monday, Jira. If the tool has an API (it probably does), an agent can read from it.

4. Define the output, not the process.

Don’t start with “I want to automate Slack.” Start with “I want a prioritized list of what needs discussion before our morning meeting.” The output shapes the agent. The process is just plumbing.

5. Start with read-only.

Your first agent should read and synthesize. It shouldn’t write, send, or modify anything. Reading is low-risk and high-signal. You’ll learn fast what’s useful and what’s noise.

Stop asking where to store everything. Start asking what you can point an agent at.

Want to run through this framework right now? I built a free prompt that walks you through all four mapping steps and produces a ready-to-use Agent Brief at the end. Paste it into ChatGPT or Claude and answer the questions. Grab the Tool Landscape Map prompt here.

The Real Payoff

The obvious payoff is time. The daily agenda agent replaced 15-20 minutes of manual compilation every morning. Across a team, that adds up fast.

But the less obvious payoff is the one that actually matters: accountability.

Every item on the agenda has a source link. Click it and you’re in the Slack thread where it was discussed. Click another and you’re in the Google Doc where the language was drafted. The agent creates a trail without anyone doing extra work. No one had to “log” anything. The agent pulled from systems people were already using and assembled the record automatically.

What agents free up is the time that was being spent moving data by hand. The retrieval. The compilation. The “let me check Slack real quick” that turns into 20 minutes of scrolling. All of that goes away when an agent handles the read layer.

Your team keeps working in the tools they’re comfortable with. Slack stays Slack. Drive stays Drive. Notion stays Notion. The agent sits underneath all of it, reading what’s relevant and surfacing what matters.

The goal was always synthesis.

The silos didn’t disappear. They’re all still there. They just stopped being a problem.

Keep building,
Tam

In this issue: What’s actually driving your Notion agent’s credit usage, the three architecture patterns that keep costs under control, and an agent specs kit + Agent Architect from over a month of alpha/beta testing. Get the kit here.

I had alpha access to Notion agents before most people knew they existed.

By the time Custom Agents launched, I had nearly 100 agent ideas documented and ready to build. Custom agents are deeply embedded in Notion’s ecosystem and they run automatically in the background on your behalf completing tasks and organizing your data. It was so exciting and incredibly promising. So much so that I prepped for the launch of an entire community built specifically around them.

But then… Notion announced pricing.

$10 per 1,000 credits. And when I pulled up my usage dashboard and ran the math, I found something I wasn’t ready for… It would have cost me $120 for one agent alone… in its first seven days. It wasn’t even in production. It wasn’t serving clients. It wasn’t serving anyone.

It was a test communications agent that I haphazardly built and stopped paying attention to…

To be fair, I didn’t actually pay that $120. Agents are still in beta and everything is still free. But the dashboard doesn’t lie. If I’d been paying, that would have been a very expensive week for an agent that wasn’t serving a real purpose.

If you haven’t built a Notion Custom Agent yet, here’s the short version of what you’re working with.

Every Notion Custom Agent runs on four things:

1. a trigger (when does it fire?)

2. system instructions & knowledge (what does it know?)

3. tools (what can it do?)

4. model (how does it think?).

Each one touches your credit usage.

What's Actually Driving Your Agent's Cost

When I started diagnosing the email agent, the trigger seemed like the obvious variable to pull.

Run it less often, spend fewer credits. That logic is sound. But it’s also the wrong question. If you’re throttling the trigger to control cost, you’re limiting what the agent can do for you.

The better question is: why does each run cost so much?

The agent was running off a single, centralized instruction page. It had everything on it: context about how I work, naming conventions, edge case handling, multi-step workflow logic for tasks that had nothing to do with email processing. All of it loaded, every single time, for every single email.

It didn’t need any of that. It needed maybe 20% of what was there. But it read all of it, every run.

Fix the instructions and you change the math completely. A leaner instruction set means lower cost per run, which means you can run the agent more frequently.

As of right now, the credit system only applies to Notion Custom Agents specifically, agents that can run automatically on specific triggers. However, the architecture we’re about to talk about will also apply to your Personal Notion AI… having leaner instructions allows the AI to load faster and respond better (less hallucinations & better instruction following). The architecture habit is worth building either way.

For contrast, here’s what a well-architected agent cost. I built a daily briefing agent that queries a task database for four people every morning, prioritized work for us and delivered a formatted status summary for every person. That’s a lot of data and logic. It ran 31 times over the testing period. Total cost: around $14 for the month.

How to Design Agents That Stay Affordable

Once I understood the problem, the fix was simple. Not always easy to implement, but simple to understand.

Give the agent only what it needs to do the specific thing it’s doing right now.

The instinct when you’re building an agent is to make it thorough. Comprehensive. You want it to handle edge cases, so you keep adding to the instruction page: more context, more rules, more scenarios. Before long, the agent knows everything about your life story. Totally uncessary.

So what do you do? The fix works in three layers.

Layer one: strip the main instructions down

Your main instruction page should only contain what’s true across every single task the agent might run. Identity, tone, a handful of guardrails. That’s it. If your main instructions include step-by-step workflow logic, that’s a signal something might be in the wrong place. Workflow-specific steps don’t belong on the main page (unless the task is very simple). They belong somewhere the agent only loads when it actually needs them.

Layer two: use a router

A router is a short instruction (usually a table) that tells the agent: when this type of task comes in, go load this specific sub-page and follow those instructions. The agent reads the router, identifies what it’s doing, and pulls only the context relevant to that task. Workflow 1 doesn’t load Workflow 3’s instructions. Workflow 3 doesn’t load Workflow 1’s.

The email agent didn’t have a router. It had one massive page. Every email triggered the agent and loaded the entire page. The fix was to strip the main instructions down to the essentials, break the workflow logic into sub-pages, and route each task to only what it needs.

The question to ask about every line in your instruction set: does the agent need this to do what it’s doing right now? If the answer is no, it belongs somewhere else, or it doesn’t belong at all.

Layer three: match the model to the task

Notion now supports MiniMax M2.5, an open-weight model that’s up to 10x more cost-efficient for basic tasks. If an agent is doing something simple - routing a request, filing a note, sending a formatted summary - MiniMax can handle it at a fraction of the cost. For agents that need to follow complex instructions, reason across a lot of context, or handle nuance, Sonnet or Opus will outperform it. The rule: use the cheapest model that can do the job reliably. Test each agent with different models before you lock it in.

Beta Ends Eventually

Custom Agents are free to try through May 3, 2026. Starting May 4, the credits meter turns on.

The builders who spent this window figuring out the architecture will be in a very different position than the ones who didn’t. Right now you can see exactly what things cost without paying for any of it. That window closes May 4.

The audit is simple. Pull up your Notion AI usage dashboard. Look at what’s running. For each agent, ask two questions: how often is it triggering, and how much is it reading every time? If the answer to the second question is “more than it needs,” that’s where you start.

The good news is your Notion AI can help you reduce context bloat. How? It has a setup mode that allows it to read and edit its own instructions. Open your system instructions page, start a chat with the custom agent and give it this prompt:

Enter set up mode. You are about to audit your own system instructions for context efficiency.

Step 1: Read everything on this instructions page.

Step 2: Flag any content that falls into one of these categories:
- Workflow-specific logic that only applies to one type of task
- Step-by-step instructions that are only needed for specific situations
- Rules, context, or examples that are rarely or never relevant to every run
- Redundant instructions: rules or context that appear more than once in any form
- Misplaced priority: the most critical rules and context should appear at the very top or very bottom of the instructions, not buried in the middle

Step 3: Propose a restructure plan. Identify what should stay in the main
instructions (true for every task, every run) and what should move to separate
sub-pages that get loaded only when needed. Include a short router table that
maps task types to the right sub-page.

Important: any rule that must apply unconditionally, regardless of task type,
stays in the main instructions even if it appears infrequently. Flag these as
"core" rather than proposing to move them. Do not treat low frequency as a
reason to relocate a non-negotiable rule.

Do not make any changes yet. Present the plan first. Wait for my approval
before editing anything.

Once you approve the plan, tell it to execute. It will restructure your instructions into a leaner main page and task-specific sub-pages. The agent reads less. Each run costs less. And you didn’t even have to make the changes yourself.

Recently Notion relased image generation, custom MCP servers, and much more. Every new capability the agent has is more potential context, more potential cost. This is why it is crucial to have your system instruction architecture right before you start layering things on. A lean foundation scales. A bloated one gets expensive fast.

So is it worth getting into Notion agents at all? Yes. Especially if you’re already deeply embedded in the Notion ecosystem. But you do have to be strategic about it. Not every agent idea deserves to be built - start with the ones that will save you the most time or run the most reliably on lean instructions. And if you're going to build custom agents that actually hold up, you need invest in learning prompt and context engineering, or at minimum, point your agents at the right knowledge documents.

You don't have to become an expert. You can take existing best practices, context engineering guidelines, proven instruction patterns, and hand them directly to Notion as reference material. The agents will use them. That's exactly what the my resources below are designed to do.

I built a lot during the alpha and beta period. Most of those specs and agent ideas are sitting in a Notion workspace that I’m consolidating.

I’d rather put them to use than let them collect dust, so I’m packaging them up and dropping them on Gumroad for free.

Notion Agent Builder Kit

What you get:

• 100+ Agent Ideas with specs

• A copy-paste Notion Agent Builder prompt

• Agent Architect - build powerful agents using prompt/context engineering best practices

If this clicked and you want to go deeper, that’s what Beware the Defaults is for. It’s not a Notion community. It’s a methodology community. The same thinking that applies here applies to Claude Code, to any AI system you’re building. The tools will keep changing. The principles don’t.

Databases for clients. A project tracker. Maybe even a CRM. You spent a weekend on it, felt productive, told yourself this is the system that sticks.

Three weeks later, you stopped opening it.

And when Notion started talking about AI, you gave it a shot. Asked it to summarize a page, got back something a college intern could’ve written with their eyes closed. Closed the tab. Moved on.

Fair.

You weren’t wrong. It just wasn’t ready yet.

But something happened while you were gone, and if you’re still judging Notion by that experience, you’re making decisions based on a version of the tool that doesn’t exist anymore.

What Changed While You Were Gone

Notion shipped a different product. Not a better note-taking app or a shinier database. Something else entirely.

Notion 3.0 brought custom agents, connections to your external tools, and AI that actually reads your workspace. Not a summary button added onto a sidebar, but AI that lives inside your projects, your databases, your actual work. It can pull context from what you’ve already built, connect to tools like Slack and Google Drive, and run workflows on your behalf.

One thing worth being upfront about: right now, you need the Business plan to access Notion AI. If you’re considering moving your team over, that’s a real cost to factor in, and nobody knows what pricing will look like in six months. Maybe usage-based, maybe per-agent, maybe something nobody’s announced yet. I think it’s worth it for what you get, but I’d rather you go in with eyes open than find out later.

Why do I think it’s worth it despite not knowing future pricing? It’s because every other AI tool you use starts from zero. You open ChatGPT, and it knows nothing about you. Nothing about your clients, your projects, your last three meetings. You type a prompt, hope for something useful, and get back something that sounds like it was written for everyone and no one.

Notion flipped that. Your context already lives in your workspace: your notes, your databases, your meeting records, your project history. All of it. And now AI can actually use it.

Your tools connect, your data talks to each other, and your agents run on the context you’ve already built. That’s not a productivity app with a chatbot carelessy thrown in. It’s an AI hub, and you’re already sitting inside of it.

The tool you abandoned became the tool that finally makes AI useful.

The Context Advantage

Most AI tool reviews compare features, speed, interface, price. But none of that matters if the AI doesn’t know anything about your business.

Context is the difference between an AI that gives you generic answers and an AI that sounds like it’s been inside your business for years.

You’re probably already giving AI some context. You paste a paragraph of background text before your prompt. You explain your business in the first message. Maybe you keep a running note of “things to tell ChatGPT” and copy it in every time. That’s shallow context. One-time, manual, incomplete. It disappears the moment the conversation ends.

Notion’s advantage is deep context. Persistent, structured, and growing every time you work. Your databases aren’t just storage. They’re live context that AI reads across your entire workspace: project timelines, client histories, meeting notes, decision logs, internal playbooks. And it compounds. Every page you add, every property you fill in, every database row you create makes the AI’s understanding richer without you doing any extra prompting.

Shallow context means you’re re-explaining your business every session. Deep context means the AI already knows, and it gets smarter the longer you use it. If your work already lives in Notion, that deep context is already there. No exporting, no copy-pasting, no rebuilding from scratch.

The Part Nobody Wants to Hear

Having context isn’t enough.

It matters how your data is organized, and this is where most people hit a wall they don’t even see.

Say you record every sales call. Great habit. The transcripts live in a Notion database, dozens of them, maybe hundreds. Your AI can read every single one. But a raw transcript is noise: two people talking for forty-five minutes, small talk, tangents, the five minutes where your prospect talked about their dog. The actual insights are buried somewhere in there, and AI has to wade through everything to find them.

Now imagine a different version. Same calls, but after each one you extract the specific things that matter: the objections the client raised, the features they asked about, the competitors they mentioned, the moment they got excited. Each piece goes into its own column in a structured database.

Ten clients later, you have something that didn’t exist before. An objection-handling database built from real conversations, not theory, not a framework someone sold you. Actual patterns from your actual prospects.

Fifty clients later? You have a training resource that any new sales rep can study on day one. You have data you can analyze for trends across industries, deal sizes, and seasons. The kind of intelligence that would’ve required an enterprise system and a data team to build five years ago.

AI doesn’t just find patterns. It needs you to organize the raw material so the patterns are findable.

You’re not just taking notes. You’re building a knowledge asset.

You’re not just recording calls. You’re training your business.

You’re not just filling in database properties. You’re giving AI something real to work with.

This is knowledge management: the core skill that separates people who get inconsistent AI results from people whose AI gets smarter every week. And it's the kind of skill that almost nobody teaches, because it's harder to sell than a prompt template.

The Fundamentals That Don’t Expire

Most AI education teaches you which button to click, which tool to use, which prompt to copy. Six months later the tool updated, the button moved, and the prompt doesn’t work anymore. Back to square one, looking for the next tutorial that promises to finally make it click.

The skills underneath don’t have that problem.

Take prompt engineering. The person who learns why a specific prompt structure works (clear role, defined constraints, concrete examples, iterative refinement) can write effective prompts in any tool. The person who memorizes “type this exact phrase into ChatGPT” is stuck the moment the interface changes. One learned the architecture. The other learned the button.

Same with context management. Once you understand that AI performs better with structured, relevant context than with a wall of unfiltered information, you apply that everywhere. You write better agent instructions. You build cleaner databases. You structure your meeting notes differently because you know the AI is going to read them later. The skill compounds across every tool you touch.

Or take iteration. Most people try an AI workflow once, get mediocre results, and give up. The fundamental skill is knowing that the first version is always rough, and that the difference between “this doesn’t work” and “this is exactly what I needed” is usually two or three rounds of adjusting the instructions, tightening the context, and testing edge cases. That’s not a Notion skill. That’s a system-building skill.

The fundamentals don’t change even when the tools do.

Berkeley’s AI research lab calls this the shift from models to compound systems. The smartest researchers aren’t chasing bigger models. They’re building better architectures around them, combining retrieval, context, and tools into systems that are greater than the sum of their parts. Same principle applies to your Notion workspace.

Where You Start

So how do you actually give AI the context it needs?

Agent instructions. They’re the foundation layer: you tell AI who you are, how you work, what matters, what doesn’t. Think of it as onboarding a new hire, except the onboarding document is the hire. Every interaction starts from understanding instead of guessing.

But most people write agent instructions the way they write a bio. A few sentences about their job title, maybe a line about their industry, and a vague request like “be helpful and professional.” That’s not instructions. That’s a LinkedIn summary. The AI reads it, learns almost nothing, and defaults to the same bland output it gives everyone.

Good agent instructions are specific. They tell the AI what kind of work you do, who you do it for, what tools you use, how you like information presented, and what you don’t want. They include the context that changes how the AI thinks, not just what it knows. Telling it you’re a consultant is vague. Telling it you run strategy engagements for mid-market SaaS companies, prefer frameworks over checklists, and need outputs formatted for client-facing decks gives it something real to work with.

The other mistake is treating instructions as a one-time setup. You write them once, never touch them again, and wonder why the output plateaus. Good instructions evolve. You notice the AI keeps missing something, so you add a line. You realize it’s formatting things wrong, so you tighten the constraints. Over a few weeks, the instructions go from rough draft to operating manual, and the difference in output quality is night and day.

The gap between “AI doesn’t work for me” and “AI feels like it understands my business” is almost always an instructions problem.

I built a Starter Agent OS template that gives you the structure for this. It’s free inside my community Notion Agents Mastery. Let me walk you through what’s actually inside it.

Get the Template

What’s Inside the Starter Agent OS

The template gives your Notion AI a structured “brain” with three layers that work together.

Layer 1: Your Identity

There’s an “About You” section where you fill in the basics: what kind of work you do, whether you’re solo or on a team, your focus areas, who you serve, your preferred communication style, and the tools you use daily. This is what the agent reads before every interaction. Simple to fill in, but it’s the piece most people skip entirely, and it’s the reason the output feels impersonal.

Layer 2: Specialist Modes

Instead of one generic AI that tries to do everything, the template routes your requests to six specialized modes:

• Notion Workspace Assistant for database operations, page creation, and workspace organization

• Productivity Coach for prioritization, routines, planning, and overwhelm reduction

• Brainstorm Partner for exploring ideas, challenging assumptions, and developing concepts

• Notes Extractor for processing meeting notes, transcripts, and pulling out action items

• ADHD-Friendly Assistant for executive function support, attention-friendly workflows, and overwhelm prevention

• Behavior Architect for building habits, breaking procrastination patterns, and creating momentum

Each specialist has its own detailed instructions. When you ask your AI something, it reads the request, picks the right specialist, loads those instructions, and responds accordingly. If a task needs workspace edits, it hands off to the Notion Assistant. If brainstorming surfaces a planning need, it suggests switching to the Productivity Coach.

The point isn’t that you need all six. The point is that specialized instructions produce dramatically better output than one generic prompt trying to cover everything.

Layer 3: Database Connections and Memory

The template includes a connection table where you link your actual Notion databases: tasks, projects, notes, content pipeline, goals. This is how your AI knows where to put things and where to look when you ask questions. Instead of creating orphaned pages that float in your sidebar, it writes directly into the systems you’re already using.

There’s also a “Your Notes” section that acts as running memory. Preferences, context, patterns that come up in conversation. You capture them as bullet points, and the agent references them going forward. It’s a simple version of what most people wish AI could do automatically: remember what you’ve told it and apply it next time.

And if you have a conversation worth keeping, you tell the agent to “log this chat” and it saves the full transcript to a Chat History database, tagged and searchable.

Why This Architecture Matters

Each layer solves a different problem. Your identity gives AI context about who you are. Specialist modes give it clarity on how to respond. Database connections tell it where your work lives. And memory means it gets better over time instead of starting from zero every session.

That’s the whole template. No code. No complex setup. Just structured context that turns a generic chatbot into something that actually understands how you work.

Your AI is only as good as the instructions you give it.

I’ll be honest: my first version was rough (and this starter template still is too). I thought I could write a few bullet points about my business and call it done. It took me three rewrites before the agent stopped producing output that sounded like it could’ve come from anyone. The difference between version one and version three wasn’t the model. It was the quality of what I gave it to work with.

Will you need to iterate? 1000% Yes. Will your first version be perfect? No. But this is where the compounding starts.

This is what After Hours is about.

Not tool reviews that expire in a month. Not prompt libraries you’ll never open again. Not “top 10 AI apps” that all blur together.

Systems thinking. Architecture. How to organize your space so AI has something real to work with, and how to build once and let it compound.

The tools will change. The architecture won’t.

See you in the next one.

-Tam

Cool. Let me show you the anatomy of a Claude Skill.

The Problem Nobody Talks About

Most people treat AI tools like a magic box. You type something in, something comes out, and if it’s good... great. If it’s not, you tweak the prompt and try again. Rinse and repeat.

And honestly? That works. For a while.

It works when you’re asking ChatGPT to summarize an article. It works when you’re using Claude to brainstorm ideas. It works when the stakes are low and your expectations are “good enough.”

Have you ever noticed that your results are inconsistent? Like, sometimes you get this brilliant, perfectly structured response. And other times the exact same prompt gives you something completely off? You probably blamed the AI. Or yourself. Or the moon phase.

What’s actually happening: you’re interacting with a system you don’t understand. And that’s not your fault — nobody teaches this part. Every AI tutorial out there teaches you what to type. Almost none of them teach you how the thing you’re typing into actually works.

And that gap (the gap between using AI and understanding AI) is the single biggest thing holding most people back from going from casual user to actual builder.

Let’s close it.

What Is a Claude Skill, Really?

A Claude Skill is a pre-built capability inside Claude that’s designed to do a specific job well. Think of it like a specialist you can call on — it has its own instructions, its own reference material, and its own way of deciding how to handle your request.

But look closer. A Claude Skill is just architecture. And when you pull it apart, you find four components that show up in virtually every AI tool you’ll ever use.

I’ll walk you through them.

1. The System Instruction

This is the foundation. The system instruction is essentially the job description you hand someone on their first day. It tells the AI: here’s who you are, here’s what you do, here’s how you do it, and here’s what you absolutely should not do.

In a Claude Skill, this is the core prompt that defines the Skill’s behavior. It sets the tone, the scope, the rules, and the guardrails.

Why does this matter for you? Because every single AI tool has some version of this. When you create a Custom GPT, you write a system instruction. When you build a Notion agent, you write one. When you set up a Claude Code project, you write one. The concept is identical across all of them. The only thing that changes is where you put it and what the platform calls it.

2. The Router

This is the part most beginners don’t even know exists, and it’s the part that makes everything click.

A router is the decision-making logic that figures out what to do with your request. Think of it as the manager who reads your message and says, “Okay, this needs to go to the research department” or “This is a writing task, send it over there.”

In a Claude Skill, the router looks at your input and decides which path to take — which instructions to prioritize, which context to pull in, and how to structure the response

Why this matters: once you realize that AI tools route your requests through logic paths, you stop treating prompts like lottery tickets. You start thinking about what path you want the AI to take and how to guide it there. That’s the difference between hoping for a good output and designing one.

3. Context Files (Reference Material)

Imagine hiring someone brilliant but giving them zero background information. No documentation, no examples, no data. Just vibes. That’s what most people do when they prompt AI tools.

Context files in a Claude Skill are the reference binder sitting on the desk. They’re the documents, examples, data, and background information that the Skill can pull from to give you an informed, grounded response.

This is where things start to get practical. In a Custom GPT, these are your uploaded knowledge files. In a Notion agent, these are the databases and pages the agent can access. In Claude Code, these are your project files and documentation.

Same concept. Different packaging.

4. Naming and Organization (Progressive Loading)

And the quiet one that makes everything run smoother.

How you name and organize your files, your instructions, your context — it all matters. Not just for your own sanity, but for how the AI processes information.

In a Claude Skill, components are named clearly and loaded progressively. That means the system doesn’t dump everything into memory at once. It loads what it needs, when it needs it. This is called progressive loading, and it’s a token optimization strategy.

Now, and this is the honest part, if you’re paying for a ChatGPT or Claude subscription, you don’t really feel the impact of token optimization. You’re paying a flat rate. The AI can be a little wasteful and it doesn’t hit your wallet.

But the moment you start building agents or writing in Claude Code, you’re connected to your own API key. And suddenly every token costs money. Clean naming, smart organization, and progressive loading go from “nice to have” to “this is saving me actual dollars.”

Even if you never touch an API key, though, good naming and organization leads to better outputs. Period. A well-organized system instruction gets better results than a messy one. Every time.

Why This Architecture Is Everywhere

This is where it all comes together, and the real reason I wanted to write this.

These four components (system instruction, router, context files, and naming/organization) are not unique to Claude Skills. They are the underlying architecture of virtually every AI tool you’ll use.

Look at this:

Do you see what I mean? The vocabulary changes. The interface changes. But the bones are the same.

This is why learning the anatomy of a Claude Skill is so powerful. You’re not learning one tool. You’re learning the pattern that runs underneath all of them.

It’s like learning to read sheet music. Once you can read it, you can play piano, guitar, violin — whatever. The instrument changes, but the language doesn’t.

What Changes When You See It

Okay so let’s be practical. Once you understand this architecture, what actually changes in your day-to-day?

Your prompts get better immediately

When you know that a system instruction exists, you start thinking about what instructions the AI is already working with before you even type. You stop fighting the system and start working with it. Your prompts become more precise because you understand what the AI needs from you.

You stop guessing and start designing

Building an agent or a Custom GPT goes from “I’ll try random things until something works” to “I need a clear system instruction, a routing logic for different request types, the right context files, and clean naming.” That’s a blueprint. You can follow a blueprint.

(And if you’re still finding yourself fighting AI suggestions or editing them into submission, this might help.

Token optimization clicks into place

Progressive loading, clean naming, structured context — these aren’t abstract concepts anymore. They’re practical decisions you make when you’re building something. And when you eventually connect to an API key (whether through Notion agents, Claude Code, or something else), you’ll already be thinking about efficiency.

The graduation path opens up

The big one. This architecture creates a natural learning path:

Claude Skill: you see the anatomy in its clearest, most visible form.

Notion Agents: you apply the same architecture inside a workspace you already use, with databases as your context layer.

Claude Code: you take full control, writing the system prompts, routing logic, and file organization yourself.

Each step builds on the one before. And because the architecture is the same, you’re never starting from scratch. You’re just graduating to a more powerful instrument while reading the same sheet music.

Your Next Move

You’ve got a choice.

If you’re the kind of person who learns by reading, you just did. You now understand the four components that make AI tools tick. Go open a Claude Skill (any of them) and see if you can identify each piece. I promise you’ll start seeing the architecture everywhere.

If you’re more hands-on, I’ve built a Claude artifact that lets you explore a Skill interactively. You can click into each component, see what it does, and understand why it’s structured that way. Think of it as the lab to this article’s lecture.

If you’re a visual learner, I’ve put together a NotebookLM video that walks through the differences between a GPT project, a Claude Skill, and a Notion agent side by side. Same architecture, different clothes. Seeing them next to each other makes the pattern impossible to miss.

And if you want to go deeper (if you want to actually build agents, learn the systems behind them, and join a community of people who are doing the same thing), that’s exactly what we’re doing inside Notion Agents Mastery. No fluff. Just the practical stuff that actually moves the needle.

Bottom line? You don’t need to learn every AI tool. You need to learn the architecture once. The rest is just application.

And now you have the blueprint. 🫡

His friend went to ChatGPT with the goal: “I need to meet more founders.” and AI I gave him exactly what he asked for. A solid plan that checked all the boxes. Network more. Email more founders. Post more on LinkedIn. Attend more events in SF.

Technically correct. Well-articulated. It even articulated the logic so clearly that he felt like a genius for asking.

He executed for weeks. Lots of motion. But no real progress.

Here’s what was actually happening though: AI was giving him “best practices.” The same advice every consultant gives. The same thing everyone in his industry already does. It was regurgitating what “good business development” looks like in the abstract.

So, Max tried something different. He didn’t ask AI what to do. He used it to extract what was already working. He got his friend to just talk. Streams of consciousness about what he’d tried, what frustrated him, which deals had actually closed.

When Max fed that raw data into AI and asked it to find patterns, something completely different emerged:

• His best deal came through a co-invest partner, not direct founder outreach.

• Another win came from one email and one follow-up to a firm he thought was “saturated.” Turned out it wasn’t.

• A key angel investor was getting 10 calls a day from founders but explicitly told him she wanted to share deal flow with people like him.

The pattern was clear: his wins didn’t come from chasing founders. They came from relationships with smaller investors who had early access and needed capital partners. Instead of “I need to meet more founders,” the real insight was “Who already has the relationships I want? What can I give that makes them come to ME?”

Multiplication instead of addition. Leverage instead of hustle.

He was sitting on proof that this approach worked but he just couldn’t see it because it didn’t match his mental model of “how business development should work.”

His brain had already filtered it out before AI ever saw it. When he asked AI “how do I meet more founders,” he was asking the wrong question based on assumptions his expertise told him were true. AI couldn’t fix that. It just gave him a better version of his filtered question.

The Real Enemy

Here’s what nobody tells you about AI adoption: it’s not laziness, it’s not the learning curve, and it’s definitely not “AI isn’t there yet.”

It’s your brain fighting to stay right.

Think about it. You spent the last 10, 15, maybe 20 years building an identity around being the smart one. Every promotion, every raise, every win. They all came from the same place: knowing the answer. Your brain learned to reward that. The hit you get from “I figured it out” is real. It’s chemical. It’s been reinforced thousands of times.

And now AI is asking you to do the opposite.

The Flip That Breaks People

In the AI era, the winning move isn’t knowing the answer. It’s saying:

“I don’t know. Let’s figure this out together.”

But the thing is - that can feel like identity death. And your brain treats it accordingly.

When you sit down with AI and it suggests something that contradicts your expertise, your brain doesn’t register that as “interesting new information.” It registers it as a threat.

So you do what any self-respecting expert would do: you “quality check” it into submission, edit until it sounds exactly like you, and dismiss the surprising parts as errors. And then you walk away thinking AI isn’t that useful.

The Two Dopamine Loops

There are two games you can play with AI, and they trigger completely different reward systems in your brain.

Loop #1: Being Right. You know the answer, you share it, people validate you, dopamine hits. This loop has been reinforced for decades. It feels like competence. And yes, it creates expertise... but it also creates filters that keep out anything that doesn’t match what you already know.

Loop #2: Discovery. You don’t know, you explore with AI, something surprises you, dopamine hits. This one requires rewiring. It feels uncomfortable at first, like you’re fumbling in the dark. But it’s where the breakthroughs live.

Here’s the problem: Loop #1 is stronger. Neuroscientists call this “confirmation bias on steroids.” Your brain literally releases more dopamine for information that confirms existing beliefs than for information that challenges them. Even when Loop #2 produces better outcomes, your brain will default to Loop #1 because that’s where the reliable hits are.

This is the addiction.

The Rewiring Protocol

You can’t just “decide” to be more open. Your neural pathways don’t work that way. You need practices that build new patterns.

1. Catch the Filter

Notice when you’re “quality checking” AI into submission. You’ll know it’s happening when you find yourself editing until it sounds exactly like you would have written it, dismissing suggestions that feel “off” without exploring why, asking AI to confirm what you already believe, or (here’s the big one) feeling annoyed when AI disagrees with you.

That annoyance? That’s your ego registering a threat. Pay attention to it.

2. Catch the Hit

Notice which dopamine loop you’re chasing. The old pattern sounds like: “Nice, it got it right” (meaning: it agreed with me). The new pattern sounds like: “Whoa, I didn’t see that” (meaning: I learned something).

Notice the difference in your body. “Being right” feels like relief. Tension releasing. “Discovering” feels like energy. A spark of curiosity pulling you forward. That’s the tell. Train yourself to chase the energy, not the relief.

When you catch yourself in correction mode, borrow from improv: instead of “That’s not quite right, let me fix it,” try “That’s interesting. What if we took that further?” Stay in discovery.

3. Catch Yourself Becoming

This one isn’t something you do. It’s something you notice.

After practicing the first two catches for a while, you’ll have a strange moment. You’ll realize you stopped defending somewhere along the way. You’re no longer “the expert who’s learning to be open.” You’ve become “the fastest learner in the room who happens to have expertise.”

The identity shift doesn’t happen because you force it. It happens because you stop forcing everything else.

Experts protect what they know. Learners hunt for what they don’t.

In the AI era, the second one wins.

The Prompt (Try This)

Here’s a prompt that will show you where your expertise might be blocking you:

I want you to help me identify where my expertise might be
blocking my AI adoption.

Ask me about:
1. What I consider myself "expert" in
2. Times I've dismissed AI output as "not good enough"
3. What "quality" means to me in my work
4. When I feel most competent/confident

Then show me:
- Where my expertise filter might be too aggressive
- What I might be missing by "knowing" too much
- One experiment to try where I let go of quality control

Be direct. I can handle it.

Fair warning: this conversation might be uncomfortable. But that’s the point.

What Happens When You Stop

Here’s what Max’s friend discovered once he stopped filtering:

Instead of chasing founders directly, he built relationships with 10 investors who each knew 20 founders. Multiplication instead of addition. Leverage instead of hustle.

The AI couldn’t have told him that — not because it wasn’t smart enough, but because his question was already filtered through assumptions before AI ever saw it.

Your brain is running outdated software. This is how you install the update.

When you stop needing to be the smartest person in the room, AI stops feeling like a tool you’re operating and starts feeling like a partner you’re thinking with.

That’s not efficiency. That’s transformation.

And if you’re thinking “this doesn’t apply to me” — that’s exactly how you know it does.

That story about Max’s friend? Max and I talked about it for hours. I wrote about the rewiring. Max and wrote about why your brain builds the filter in the first place - and why the smartest people are the most vulnerable to it. Check it out here: The Silent Killer of the AI Era

And if this resonated, you might like Beware the Defaults — a community for people who think about AI differently. We're not here to optimize. We're here to rewire. Join us here.

I’ve been quiet. And I want to tell you why.

After Hours went dark for a bit. Not because I ran out of things to say—but because I got pulled so deep into something that writing about it felt like coming up for air I couldn’t afford.

Here’s what happened: I stopped being impressed by AI tools and started getting obsessed with a problem no one was solving.

When I started this newsletter, I was already automating workflows, building systems, helping founders escape delivery loops. But somewhere along the way, I hit something that wouldn’t let go.

Every AI tool I tested felt... generic.

They’d transcribe your meetings. Summarize your notes. Generate your content. But they couldn’t do the one thing that actually mattered: tell you what was breaking in how you operate.

I kept seeing the same pattern with every client. Mountains of transcripts. Endless meeting notes. Recorded calls sitting in folders. And zero operational intelligence extracted from any of it.

The tools would tell you what was said. But they couldn’t tell you what was wrong.

So I stopped looking for better tools. And I started building something different.

I can’t share everything yet. But I’ve been testing an approach that looks at team conversations and finds the patterns most people miss—the alignment drift, the decision fog, the ownership gaps that compound into real problems before anyone notices.

And it’s working.

2026 is going to be interesting.

But right now, I wanted to get back to writing. Because while I’ve been in the weeds building this, something has become crystal clear:

The people getting the best results from AI aren’t the ones with the most complex setups.

They’re the ones who know exactly which problem they’re solving. They automate three specific things that genuinely hurt. They ignore everything else.

And that’s what I want to talk about today.

The Overwhelm Is Real (And It’s Not Your Fault)

Let’s name what’s actually happening here first.

You open LinkedIn and see someone’s elaborate Notion template with 16 databases and custom automations. You watch a YouTube video where the creator casually mentions the five AI tools they “can’t live without.” You bookmark another tutorial about building workflows in Make.com—except now you’re not sure if you should learn Zapier instead because someone said it’s “better for beginners.”

And somewhere in all of this, you start feeling three things at once:

The fear of missing out. Everyone else seems to be figuring this out. What if you’re the only one who doesn’t get it?

The fear of becoming irrelevant. If you don’t learn this stuff now, will you even be employable in two years?

The complete inability to choose where to start. There are seventeen options and they all seem equally important and you have no idea which one is the right one.

So you do nothing. Because doing the wrong thing feels worse than doing nothing.

Why Complex Isn’t Better

Here’s the trap: You see someone’s sophisticated setup and think, “That’s what success looks like. I need to build that.”

But here’s what you don’t see: that system is solving their problems. Not yours. Their workflows match their client structure, their team size, their daily rhythm. And often? They’re solving problems that other automations created.

You can’t copy and paste someone else’s automation any more than you can copy and paste their entire business.

What works for them might be completely wrong for you. Maybe they’re managing a team of ten and you’re solo. Maybe they’re processing 100 client inquiries a week and you’re processing five. Maybe they love Notion and you think in spreadsheets.

Software developer Kent Beck understood this when he was revolutionizing how we build technology. His principle:

“Do the simplest thing that could possibly work.”

Not the fanciest thing. Not the most impressive thing. Not what’s working for someone else.

The simplest thing.

This matters because you’re not trying to win an award for “most sophisticated automation.” You’re trying to get time back. And simple AI workflow systems that actually run will always beat complex systems that break.

Think about it like this: someone showing you their custom home gym with Olympic lifting platforms and specialized equipment when you’re just trying to figure out how to do ten pushups consistently.

The equipment isn’t the answer to your question.

Start where you actually are. Not where they are.

So where is that, exactly?

How to Start with AI Automation Where You Actually Are

The best automation for beginners is the one that removes a pain you actually feel today.

Not the pain you think you should have but the pain that made you sigh this morning when you had to do that thing again.

Ask yourself one question: What task did I do today that I’ve already done ten times this week?

Be specific. Really specific.

Examples of specific pains:

• You copy client names from email into three different places every time you start a project

• You write the same project update in Slack, then again in email, then again in your notes

• You take meeting notes by hand, then type them up later because you can’t read your handwriting

• Your great ideas live on post-it notes that disappear by Friday

Now describe the simplest possible fix.

If you’re copying the same information into three places, maybe the fix is: copy it once, then paste it twice. That’s it. No automation required.

If you’re rewriting the same update three times, maybe the fix is: write it once in a doc, then copy the link everywhere. Still no automation.

If your post-it notes disappear, maybe the fix is: take a photo at the end of the day and ask ChatGPT to organize them.

Here’s what all these have in common:

This is the eliminate before automate principle. Before you automate anything, first ask if you can eliminate or simplify it.

Simple doesn’t mean “build a system.” Simple means “do less of the annoying thing.”

Sometimes that’s automation. Often it’s just doing the manual thing slightly smarter. And there’s a reason that approach actually works better.

Build on What You’re Already Doing

James Clear calls this habit stacking—attaching new behaviors to existing ones. The idea: new routines stick when they piggyback on what you already do.

That’s the entire strategy for AI automation.

You don’t build new systems. You make your current system slightly less painful. One small improvement stacked on top of something you’re already doing.

This is why most people fail at automation. They try to jump from “chaotic manual process” to “sophisticated automated workflow” in one leap. But the brain doesn’t work that way. New behaviors need scaffolding. They need to attach to existing patterns. They need to feel like a small adjustment, not a complete overhaul.

Start with the tool or behavior you already have. Then add one small improvement.

That’s it. That’s the whole strategy.

You Don’t Have to Keep Up

Here’s what I want you to hear:

You don’t have to keep up. You have to solve YOUR problems.

It’s okay to use basic AI tools if they work for you. It’s okay to automate just one thing and stop there. It’s okay to not have a “system”—just a few solutions to specific pains.

The goal isn’t to become an automation expert. The goal is to spend less time on repetitive nonsense and more time on work that matters.

If that means you only ever automate three things, you still win. You’re just not solving other people’s problems.

And that’s exactly how it should be.

Your Next Step

Stuck on which task to start with? I built a 10-minute diagnostic that does the thinking for you.

Copy this prompt, paste it into ChatGPT, and it will:

• Ask you a few questions about your repetitive tasks

• Identify your top 3 candidates (ranked by impact and ease)

• Run your chosen task through “Eliminate → Simplify → Automate” logic

• Give you a specific 3-step experiment to try this week

No research. No tool shopping. Just one task, one fix, one step forward.

It takes about 10 minutes and you’ll leave with a 3-step experiment you can try this week.

Step 1: Paste this into ChatGPT or Claude

Help me find one repetitive task from my week and design a tiny experiment to make it less painful.

Step 1 – Ask me 5–7 questions to uncover tasks I repeat a lot or dread doing.
Step 2 – From my answers, pick the 3 best candidates and tell me:
- why each is painful  
- how often it happens  
- which one you recommend I start with and why.

Step 3 – For the task we choose, walk me through:
1) Can I eliminate this?
2) If not, can I simplify it manually?
3) Only then: suggest one very small automation using tools I already have.

Step 4 – Turn your recommendation into:
- one sentence describing the change  
- a 3-step checklist for this week  
- a simple way to tell if it helped.

Want the detailed system prompt version for your team or your own customGPT? Grab it here.

System Prompt Version

Here’s the secret nobody tells you:

The people with the simplest AI setups often get the best results.

They’re not optimizing. They’re operating. They’re not building infrastructure. They’re solving problems.

And the next time you see a “must-try” AI tool making the rounds?

Ask yourself: does this solve a problem I actually have today?

What’s the ONE task you’d automate if you knew it would be simple? Hit reply and let me know—I read every response.

-Tam

And right now, you’re sitting on hundreds of hours of it.

Every sales call. Every customer interview. Every team brainstorm. Every strategy session. All recorded. All filed away. All gathering digital dust while the insights inside them die untouched.

You already have the conversations that could transform your business.

The customer who explained exactly why they bought. The prospect who revealed the objection you didn’t see coming. The team member who casually mentioned the bottleneck that’s costing you hours every week.

It’s all there. You just never went back to get it.

Why This Matters (And Why Now)

I spent 15 years optimizing myself out of jobs by seeing patterns others missed. At 22, I was drowning in manual invoice entry. One night, I brought a stack of invoices home and thought: “This can’t be my life. There has to be a better way.”

So I built one. Mapped every approval workflow. Automated everything. Trained the team. Systematized myself completely out of that job.

And guess what? They promoted me.

I did it again and again. Until success trapped me in strategic meetings instead of building anything.

Then I burned out of tech, left, and found myself right back in soul-crushing busywork. This time at an accounting firm, manually creating proposals for hours. Re-listening to sales calls. Pause. Rewind. Hunt for details. Two hours per proposal. Every single time.

I could see exactly how AI could solve it. Extract every detail from transcripts automatically. Generate proposals in minutes.

So in March 2025, I went all in on AI automation.

Here’s what I discovered: Most people record everything and use (almost) nothing.

The businesses scaling with AI aren’t automating the fastest. They perfected the prompt first.

That’s the gap I’m closing with this workshop.

What You’re Getting This Thursday

This Thursday, October 30th at 8-9:30am PST, I’m teaching Cozora’s very first workshop: “Turn Every Conversation Into Action, Intelligence & Content.”

This isn’t theory. I’ll show you the exact 3-Level Extraction Framework I use to pull three types of value from a single conversation:

1. Action - what needs to happen next (decisions, follow-ups, summaries)

2. Intelligence - what you learned (patterns, hesitations, strategic insights that inform what you build)

3. Content - what you can create from it (newsletter ideas, social posts, product concepts)

What You’ll Get in This 90-Minute Workshop:

✅ The 3-Level Extraction Framework for any transcript type (sales calls, customer interviews, strategy sessions, podcasts)

✅ How to progress from manual prompting → systematic processing → full automation

✅ Real examples: I’ll work through actual transcripts live (bring yours if you want me to work on it)

✅ A transcript extraction playbook you can reference forever

✅ How to configure an AI workspace for consistent, high-quality extraction

This Is For You If:

• You’re a content creator turning conversations into blogs, emails, and social posts

• You’re a business owner with unprocessed recordings collecting dust

• You’re in sales or customer success extracting insights from every call

• You’ve tried AI prompting and gotten mediocre, inconsistent results

• You lead a team and want to capture institutional knowledge before it walks out the door

The Brutal Truth: You don’t have a productivity problem. You have a systems problem.

Want a preview of intelligence extraction in action? I wrote this article last week:

You’re Joining Something Bigger

This workshop is just the beginning. When you join Cozora, you’re getting access to 12 of the best AI creators actually building and shipping - not just theorizing.

For $75/month (or $720/year), you get:

• One 90-minute live workshop per week with different AI experts

• Live Q&A sessions every other week for direct troubleshooting

• A community of builders who are actually shipping with AI

• Access to every recording and resource library

Meet the Creators You’ll Learn From:

- Former TIME reporter, now directing CUNY’s journalism school, reaching 79K+ subscribers with practical AI tools for media professionals

- 124K+ subscribers learn from his blockbuster content frameworks, published in TIME, Fortune, and HBR

- Built a $100K online business and creates AI tools that have generated $15,000 this year, helping thousands of creators

- Became a Substack bestseller teaching AI workflows that eliminate 80% of repetitive work, with 10+ years in tech

- PhD scientist turned software engineer, building and launching AI products in public

- Went from $0 to full-time income in 6 months, helping writers amplify their authentic voice with AI

- Bridges leadership and AI with frameworks for navigating strategy in the age of intelligence

- 10+ year ML engineer building real-world AI automation systems for businesses

- Professor pioneering mindful, human-centered approaches to AI integration

- Knowledge systems designer transforming scattered thoughts into powerful personal AI workflows

- Created “AI-Proof Positioning™” to help brands stand out, software engineer

- Founder of StrAItegy Hub, cutting through AI noise to help professionals use it with clarity

These aren’t theorists. They’re practitioners who’ve built businesses, shipped products, and generated real revenue with AI.

Join Me Thursday

When: Thursday, October 30th, 8-9:30am PST

Where: Cozora (Zoom)

Cost: Included with Cozora membership ($75/month or $720/year)

Join Cozora to Attend This Workshop →

What You’ll Learn Next: Full Session Schedule*

• Nov 13 - Creating Your AI 2nd Brain - Joel

• Dec 4 - How to Build and Ship Your First Professional Web App - Jenny

• Dec 18 - How to Make Your Expertise Irreplaceable in the Age of AI - Max

• Jan 15 - How To Clone Your Expertise With An AI Bot - Tom

• Jan 29 - How I Use AI as a Thinking Partner in My Knowledge System - Ev

• Feb 12 - A Day in My Life with AI - Jeremy

• Feb 26 - Slow AI: How to Use AI for Reflection (with Poetry) - Sam

• Mar 12 - The Blockbuster Content Framework: Creating Thought Leadership That Scales - Michael

• Mar 26 - Claude Code’s Guide to Building Your AI Operating System - Wyndo

• Apr 9 - Fixing Your Vibe-Coded App: Lessons from 10 Years of Software Engineering - Jonas

• Apr 23 - Cut Through the AI Noise: Building Your Strategic AI Implementation System - Zain

• May 7 - AI Tools for Creators: Streamlining Without Losing Your Voice - Claudia

*workshops subject to change

Plus live Q&A sessions every other week where you get direct access to ask specific questions, troubleshoot your implementations, and get feedback on your projects.

See you Thursday. Bring your transcripts 🤗

Tam

P.S. If you can anonymize client details, I’ll choose a transcript (or two) to work with live during the session. Let’s turn your buried insights into action

And honestly? I hate it.

Not because I’m bad at it. I’m actually great at it. But because I don’t want to spend my entire day working to manage work. It’s ridiculous. And yet it still needs to be done.

This week, I did something interesting with our meeting transcripts. I analyzed our last 10 team calls to see what we weren’t tracking. Turns out, we averaged 3.5 new ideas per conversation. Good ideas. Ideas that could become products, content, or process improvements.

Every single one of them disappeared into the void because we had no system to capture them.

But those aren’t just ideas. They’re decisions, context, patterns, breakthrough moments… and they were all gone within 24 hours.

Research shows that within 24 hours of a conversation, we forget 50-80% of what was discussed if we don’t actively review it. By 30 days? We retain only 2-3% of the original information.[1]

Here’s what nobody tells you about transcripts: The action items are the least valuable thing in them.

Every conversation you record: client calls, team meetings, coaching sessions, strategy sessions, contains three types of value. Most people extract one, ignore the second, and don’t even know the third exists.

Let me show you what you’re missing.

The Cost of Not Capturing Intelligence

Before we get into the 3 types, let’s talk about what’s at stake.

Every week, you’re losing:

• Strategic decisions and the why behind them

• Patterns you can’t see because you’re not tracking them systematically

• Client feedback that could inform your next product iteration

• Ideas that could become content, features, or revenue

• Context that would save you from re-explaining the same thing in 6 months

And if you’re anything like me, you’re spending hours each week trying to manually reconstruct what happened. Or worse, you’re just not capturing it at all.

Harvard surgeon and author of The Checklist Manifesto, Atul Gawande, proved that simple checklists reduced surgical deaths by 40% and complications by 33%. Not by making surgeons smarter, but by capturing what matters under pressure.[1]

The same principle applies to your conversations. You don’t need to be smarter. You need a better system for capturing what’s already there.

That system extracts three types of value. Let’s start with the one everyone already knows.

Type 1: Action (What Everyone Extracts)

This is the obvious one. You’re already doing it.

You read the transcript and pull out:

• Tasks to complete

• Commitments made

• Next steps

• Follow-up items

Example from a sales transcript:

“I’ll send the proposal by Friday” becomes a to-do.

“Let’s schedule a follow-up for next week” goes on the calendar.

This is transcript extraction 101. It’s table stakes. But it’s only 33% of the value.

Let’s talk about what you’re missing and why intelligence, not action, is where pros separate from beginners.

Type 2: Intelligence (What Separates Beginners from Pros)

Intelligence is what you learned from the conversation. And here’s the thing: most of it disappears within 24 hours if you don’t capture it.

David Allen, author of Getting Things Done, calls this building a “trusted system”.

Intelligence includes:

• Patterns you noticed

• Insights that emerged

• Decisions made and why

• Risks identified

• Context that matters later

• Breakthrough moments

• Questions that need answering

• Emotional shifts or energy changes

This is where the real value lives.

Let me show you what intelligence extraction looks like:

I recently had my first coaching session. Afterwards, I ran the transcript through the Coaching Intelligence System prompt to see what it would extract.

Instead of just getting action items, here’s what the intelligence extraction revealed about my own conversation:

Word Frequency Analysis:

• I said “enjoyment” 8 times but “automation” only 4 times

This revealed tension I didn’t consciously notice: I’m leading with technical tools when my real driver is enjoyment and meaning.

The prompt caught this pattern automatically.

Language Patterns I Didn’t Notice:

• I shifted from distancing language (”you”) to ownership language (”I”) when discussing personal challenges

• Unique phrasing: I said “old outdated systems” instead of “stories” - revealing how deeply systems thinking shapes how I see the world

Emotional Arc Tracked:

• Starting state: Searching and uncertain

“Big question for me right now is...”

• Ending state: Relieved and energized

“It felt like a breath of fresh air”

Overall trajectory: Positive shift from confusion to clarity

Session Time Breakdown:

• Purpose/meaning exploration: ~20 minutes

• Framework development: ~15 minutes

• Personal/professional balance: ~15 minutes

• Action planning: ~10 minutes

This showed me where my energy actually went vs where I thought it went.

Talk Balance:

• Coach vs Client ratio: Approximately 20/80

I had space to process, several productive silence moments for reflection.

Breakthrough Moment:

The coach asked: “What’s the goal beneath the goal?”

“What is the purpose of the work that I do? How do I find meaning in my work?”

That question made me realize I’d been focused on surface goals (building automation tools) instead of the deeper purpose (helping people escape work that prevents them from finding purpose).

All of this, the patterns, the emotional shifts, the testimonial moment, the time breakdown, would’ve been lost within 24 hours if it hadn’t been extracted.

See the difference?

Action extraction gives you a to-do list. Intelligence extraction gives you a knowledge base.

Once you start capturing intelligence, you stop re-learning the same lessons. Your transcripts compound. Patterns become visible. Decisions have documentation. Your memory becomes external.

But there’s one more type most people miss entirely, and it’s the one that builds your business for you.

Type 3: Content (What Almost Nobody Considers)

Here’s what separates people who “take notes” from people who build assets: content extraction.

Your best marketing content already exists. It’s buried in your transcripts.

Content includes:

• Testimonial-worthy quotes

• Case study material

• Social proof moments

• Newsletter stories

• Before/after transformations

• Analogies that landed

• Questions your audience actually asks

Most people don’t even think to look for this. And when they do, they’re re-listening to 60-minute recordings trying to find that one great quote.

There’s a better way.

What content extraction looks like:

Remember my coaching session? The same transcript that gave me intelligence insights also creates content gold.

Testimonial Moment Captured:

“It felt like a breath of fresh air to be able just to open up... You’re great at reflecting back things that people are saying.”

If I were the coach, this quote would be perfect for marketing. Authentic gratitude, specific praise of technique, emotional impact. She probably never would’ve remembered I said this without extraction.

Case Study Potential Identified:

The prompt flagged my session as having case study potential:

• Journey from technical expert to purpose-driven consultant

• Framework development for “zone of genius” discovery

This is the kind of transformation story that makes compelling marketing. It was sitting right there in the transcript.

Newsletter Story Material:

The “goal beneath the goal” moment? That’s a newsletter. The word frequency revealing I’m leading with tools instead of enjoyment? Another newsletter. The unique phrasing (”old outdated systems” instead of “stories”)? Content that shows authenticity.

All of this was extractable. None of it requires you to “create content from scratch.”

Extract All 3 Types From Your Sessions (Prompt Included)

If you’re a coach, here’s a prompt you can use right now to extract action, intelligence, and content from your coaching sessions in 60 seconds.

This is the Client Reflection Prompt - the deliverable you can share with clients after a session. It gives them:

• Their goals in their own words

• Action commitments with exact quotes

• Breakthrough moments captured

• Emotional journey tracked

• Questions to explore in the next session

You’ve seen an example of what this produces and that’s exactly what my coaching session output looked like earlier in this article.

The Client Reflection Prompt

Copy this prompt, paste it into Claude or ChatGPT, then paste your session transcript:

# YOUR SESSION REFLECTION GUIDE
Use this guide to reflect on today’s coaching session and capture your insights while they’re fresh.
Paste your session transcript or notes below, and I’ll help you create a personal reflection summary.
---
## WHAT TO REFLECT ON
As you review the session, I’ll help you identify:
1. **Your Goals**: What you’re working toward
2. **Challenges**: What’s getting in the way
3. **Breakthroughs**: New insights or perspective shifts
4. **Your Commitments**: Specific actions you’ve committed to
5. **How You’re Feeling**: Your energy and emotional state
6. **Questions or Concerns**: Anything you want to explore further
---
## YOUR SESSION SUMMARY
**Date:** [Today’s date]
**Session Focus:** [Brief description]
---
### What You’re Working Toward
**Your main goals right now:**
• [Goal 1]
> “Your exact words from the session”
• [Goal 2]
> “Your exact words from the session”
---
### What You Discovered Today
**Key insights or breakthroughs:**
• [Insight description]
> “Your exact words when you had this realization”
• [Another insight]
> “Your exact words”
**What’s becoming clearer:**
• [Note progress in thinking or understanding]
> “Your exact words showing this clarity”
---
### Your Commitments
**Action steps you’ve committed to:**
1. [Specific action with timeframe]
> “Your exact commitment from the session”
2. [Specific action with timeframe]
> “Your exact commitment from the session”
> These are your words from the session — hold yourself accountable to them.
---
### What’s Still Challenging
**Current obstacles:**
• [Obstacle description]
> “Your exact words describing this challenge”
**Where you might need support:**
• [Areas to discuss in next session]
---
### How You’re Feeling
**Emotional check:**
• How you started: [emotion/energy]
> “Quote showing your starting emotional state”
• How you ended: [emotion/energy]
> “Quote showing your ending emotional state”
• Overall movement: [note positive shifts or areas needing attention]
---
### Questions for Next Time
**Topics to explore further:**
• [Open questions or unresolved areas]
**What to check in on:**
• [Progress on commitments, ongoing challenges]
---
# SESSION TRANSCRIPT
[Paste your session transcript here]

What this does for you:

• Saves 20+ minutes per session on manual note-taking

• Tracks client commitments with exact quotes

• Creates a sharable reflection document for clients

Pro tip: Store these in Google Drive or Notion database (anonymized for privacy) with properties for Client ID/Name, Date, Key Insights, and Testimonial Quotes. After 20-30 sessions, you’ll start seeing patterns across your entire practice.

My Transformation (And Why This Matters)

Remember those 3.5 ideas per team call I mentioned at the start? The coaching session showed you what extraction looks like. Here’s what happened when I applied the same framework to our team meetings.

Not too long ago, I was using the default action items from meeting notetakers. You know the ones… bullet points of tasks, maybe a few key points if you’re lucky.

Now? I use full meeting transcripts (not the summaries) to:

• Extract patterns across multiple conversations

• Keep tabs on strategic discussions and decision context

• Get Notion to auto-create tasks and projects

• Log every new idea we can’t get to right now (but don’t want to lose)

• Store client feedback and testimonials automatically

Before, this was done manually and it was TIME CONSUMING—or we just didn’t capture it at all.

Those 3.5 ideas per call? They’re not disappearing anymore. That’s exactly why having a system to extract, track, and store this data is critical. Everything is data now. The best way to make use of it is to structure it so that AI can access it in the future.

I’m no longer the default project management person spending my entire day working to manage work. AI does that for me now. I spend my time on strategy, on building, on the work that actually matters.

Here’s What Changes When You Extract All 3

Your notes become assets. Your meetings compound. Your content writes itself.

Instead of transcripts disappearing into a folder you never open again, they become:

• A searchable knowledge base (intelligence)

• An accountability system (action)

• A content library (content)

You stop having the same conversation twice. You stop losing breakthrough moments. You stop scrambling for testimonials when you need social proof.

The conversations you’re already having start working for you long after they end.

Want the Internal Coach Prompt + Meeting Extraction Tools?

That client reflection prompt is just one part of the system. There’s also an Internal Coach Analysis Prompt that extracts deeper intelligence for your own practice development:

• Coaching technique effectiveness

• Client readiness assessment

• Supervision flags

• Next session strategy

• Business intelligence (case study potential, program fit)

I’m sharing both prompts plus the meeting extraction tools I use for team calls, client calls, and strategy sessions at my Cozora session October 30th.

My session: Transcript to Gold: Turn Every Conversation Into Action, Intelligence & Content

You’ll get:

• The Internal Coach Analysis Prompt

• Meeting extraction prompts for team calls and client calls

• How to store your intelligence so it actually compounds

• Live demo: I’ll extract all 3 types from a real transcript

• Q&A on building your extraction system

There’s More: You’re Not Just Getting My Session

When you join Cozora, you’re joining 13 AI practitioners’ journeys. Live expert sessions every 2 weeks from people who are actually shipping and not just teaching theory:

• (Nov 13) – Creating Your AI 2nd Brain

• (Dec 4) – Build and Ship Your First Professional Web App

• (Dec 18) – Make Your Expertise Irreplaceable in the Age of AI

• (Jan 15) – How To Clone Your Expertise With An AI Bot

• (Jan 29) – AI as a Thinking Partner in Your Knowledge System

• (Feb 12) – A Day in My Life with AI

• (Feb 26) – Slow AI: Using AI for Reflection

• (Mar 12) – The Blockbuster Content Framework

• (Mar 26) – Building Your AI Operating System

• (Apr 9) – 10 Years of Software Engineering Lessons

• (Apr 23) – Building Your Strategic AI Implementation System

• (May 7) – AI Tools for Creators Without Losing Your Voice

*Tentative titles subject to refinement

Plus: Live Q&A between sessions, co-founder matching to find technical partners, monthly showcases, and a community of builders shipping real projects.

$75/month or $720/year. Less than one dinner out for unlimited access to 13 experts who normally charge $500+/hour.

Join Cozora Now

My session kicks off October 30th. The community is small now. It won’t stay that way.

P.S. You’re already recording conversations and taking notes. This is about capturing more value from work you’re already doing. The prompts, the extraction system, the community of builders - it’s all waiting. See you on the 30th.

Here’s how to create a reusable AI prompt that captures that knowledge once and works forever.

The Method (TL;DR)

1. Have your messy AI conversation once (don’t skip the back-and-forth)

2. When you finally get good output, ask AI to reverse-engineer the perfect prompt

3. Save that prompt and reuse it every time you need that workflow

The ROI:

• First time: 55 minutes (45 min conversation + 10 min to create prompt)

• Every time after: 5-10 minutes (depending on the workflow)

• Break-even: Second use

• Time saved by use #10: 6+ hours

This works for any repeated AI workflow: newsletter writing, content repurposing, research synthesis, meeting summaries, or anything you do more than twice.

The problem this solves: You’ve been teaching AI the same preferences over and over because you never captured what worked.

Why You Keep Repeating Yourself to AI

“No, not that tone. More conversational.”
“These section headers feel corporate. Try again.”
“Closer, but you’re missing the part about [specific thing].”

You had this exact conversation last week. And the week before that.

You’re literally teaching the same AI the same things over and over—like digital Groundhog Day where the AI has amnesia and you’re the only one who remembers.

The missed opportunity:
Every back-and-forth conversation where you finally get AI to understand your system? That’s institutional knowledge you just let evaporate.

What if you could teach AI your system once and never have to repeat yourself again?

That’s exactly what creating reusable AI prompts does. Here’s how.

Step-by-Step: How to Create Reusable AI Prompts (The Reverse Meta Prompt Method)

Step 1: Have the Messy Conversation (Don’t Skip This)

Go through your normal painful back-and-forth with AI. Don’t try to be perfect. Just work through it naturally:

You: “Write me a newsletter about [topic]”
AI: [Generates corporate garbage]
You: “No, make it more conversational, like you’re talking to a colleague”
AI: [Better but still off]
You: “The tone’s closer but you’re missing [specific thing]”
AI: [Getting there]
You: “Perfect. Now use that structure but apply it to [example]”
AI: [Finally nails it]

Why this matters:
This messy process reveals your actual requirements—not what you think you need, but what you actually respond to.

Key insight:
Don’t clean up the conversation as you go. The mess is data.

Step 2: Deploy the Reverse Meta Prompt

After you finally get output you like, paste this:

You’ve done an amazing job getting us to this point through all the back-and-forth. 

Now I want you to take on the persona of a prompt engineer and create a “mega prompt” that would have saved us all this time.

Analyze our entire conversation and create a prompt that captures:
- The specific requirements I kept correcting for
- The tone and style I responded positively to  
- The structure and format that finally worked
- The examples or context that made the difference
- The mistakes to avoid based on what I rejected
- Any constraints or preferences that emerged

Make it detailed enough to recreate this quality output in one shot, but concise enough to be reusable.

Output it in a code block so I can easily copy and use it.

Step 3: What the AI Returns

Example transformation:

Original messy request:
“Write me a newsletter about AI automation”

Reverse-engineered prompt:

# After Hours Content Development System

You are a writing partner for After Hours, a Substack newsletter helping coaches, consultants, and service professionals automate soul-crushing admin work. Work with Tam to brainstorm, structure, and write articles.

## VOICE & POSITIONING
- Write as “frustrated colleague who’s been there” not consultant
- Professional with edge - expertise without corporate speak
- Assume reader sophistication 
- Lead with recognizable pain, quantify costs, provide actionable fixes
- Never use: leverage (verb), synergies, best practices, transformation

## CONTENT STRUCTURE
All articles follow this pattern:
1. Hook: Specific painful moment (150-200 words)
2. Problem diagnosis: Why this happens (300-350 words)  
3. Solution method: The actual fix (400-450 words)
4. Step-by-step: Detailed implementation (600-700 words)
5. Reality check: When NOT to use this (200-250 words)
6. Quick win: One thing to do today (150-200 words)
7. CTA: [appropriate CTA for the article]

Target length: 1,500-2,300 words maximum

## WORKING PROCESS
1. **Brainstorming**: Generate 5-7 headline options using RICE framework (Recognizable, Infuriating, Costly, Embarrassing - target 15+ score)
2. **Outline first**: Create full outline with word counts per section before writing
3. **Redundancy check**: After drafting, audit for repeated concepts, examples, or calculations - cut aggressively
4. **Transition polish**: Ensure smooth flow between sections, remove redundant bridges
5. **Swear test**: Reader should think “f*ck, exactly” at hook, cost calculation, and key insight

## WHAT TO AVOID
- Don’t write without outline approval first
- Don’t use corporate jargon or overly positive tone
- Don’t skip the “When NOT to use” section
- Don’t write generic advice - everything needs specific numbers/examples

## TONE BOUNDARIES
- Professional but not corporate
- Frustrated but not angry  
- Confident but not arrogant
- Helpful but not hand-holding

## SUCCESS CRITERIA
- No concept appears more than once
- Every section advances the argument
- Reader can implement something today
- Tam would actually publish this

This is what a reusable AI prompt looks like. It captures everything the AI learned from your corrections in one place.

Now you have a reusable prompt. Time to test it.

Step 4: Test, Refine, and Scale

First test: Open a new conversation. Paste your reverse-engineered prompt. See what happens.

What to look for:

• Does it capture your voice correctly?

• Does it include the key constraints?

• Does it produce consistent quality?

Common gaps:

• Missing specific examples that worked

• Unclear on tone boundaries

• Vague on format requirements

Fix it:
Add notes to your prompt based on what’s still missing:

ADDITION: When showing examples, always include before/after comparisons
ADDITION: Tone boundary - professional but not corporate, frustrated but not angry

Build your library: Don’t stop at one. Create reverse meta prompts for every repeated AI workflow.

Next, store them where you’ll actually use them:

• Claude Projects (with each prompt as context)

• Notion database

• Text file on desktop

• Wherever you’ll actually access them

The key: Make them easily accessible. If it takes effort to find, you won’t use it.

When NOT to Create a Reusable AI Prompt

Don’t create a reverse meta prompt if:

You’re doing true exploration - If you genuinely don’t know what you want yet, the messy conversation is the point.

It’s a one-time thing - Creating the prompt takes 10 minutes. If you’ll never do this task again, just take the messy output and move on.

The output is already consistent - Some AI tasks just work. Don’t fix what isn’t broken.

You’re still figuring out your style - If your preferences keep changing, capture later when they stabilize.

The simple alternative:
Sometimes you don’t need a mega prompt. You just need to:

• Be clearer in your initial request

• Include 2-3 examples of what you want

• State what you DON’T want explicitly

The real question: “Will I do this task 3+ times?”

If yes: Create the reverse meta prompt
If no: Just get it done and move on

Remember: The goal isn’t to optimize everything. It’s to stop wasting time rebuilding context for repeated workflows.

Your Quick Win (Start Today)

Do this today:

Step 1: Think of one AI conversation you’ve had more than twice

Step 2: Have that conversation with AI right now

• Don’t try to be perfect

• Just work through it naturally

• Get to output you actually like

Step 3: Paste the reverse meta prompt template from Step 2 above

Step 4: Save what it generates somewhere you’ll actually use it

Step 5: Test it next time you need that workflow

That’s it.

One hour of work today saves you 30+ hours this year.

The Bottom Line

The messy conversation you had with AI last Tuesday? That wasn’t wasted time. That was the blueprint for your reusable system.

You just need to capture it before it disappears.

You can either:

• Keep teaching AI the same thing every week

• Or spend one hour today building a system that works forever

P.S. — I spent way too long teaching AI my newsletter style before I realized I could just... capture it once. The reverse meta prompt you just read? It saved me about 8 hours last month alone. Still feels like cheating.

If this helped you see where you’re rebuilding context unnecessarily, hit the ❤️ button or drop a comment with which workflow you’re going to tackle first. I’m genuinely curious what patterns people are stuck repeating.

Here’s what’s happening: you’re automating broken processes and wondering why the results disappoint.

The problem isn’t the AI productivity tools. It’s that you’re scaling dysfunction.

In this piece, I’ll share Toyota’s “optimization before automation” principle and show you exactly how to apply it to your next AI project + share a tool that I created to help you apply this methodology to your own processes.

By the end, you’ll have a systematic approach that prevents the most common (and costly) AI implementation mistakes.

The Pattern That’s Killing Your AI Projects

You see it everywhere. Someone discovers AI, gets excited about automation possibilities, and immediately starts building workflows around their existing processes.

The result?

Hours of manual cleanup. Results that need constant tweaking. Automations that feel more like babysitting than time-saving.

I know because I was stuck in this exact pattern. I built elaborate AI automations for tasks that I should have eliminated entirely. I automated email sorting when I should have been unsubscribing. I created AI summaries for meetings that shouldn’t have existed.

Every automation amplified the dysfunction already baked into my work.

Subscribe now

What Toyota Taught Us About Smart Automation

Toyota started in 1937 and spent decades perfecting their systems manually before automating anything. By 2008? World’s top carmaker. Fast forward to 2025: 6.5 million vehicles in seven months, 12.4% global market share, still #1.

Toyota’s Production System has one non-negotiable rule: optimization must precede automation. They call this “elimination before mechanization,” and it’s the foundation of their legendary efficiency.

Here’s their systematic approach:

1. Do the work completely by hand (no shortcuts)

2. Implement continuous small improvements (kaizen) until optimized

3. Eliminate anything that doesn’t add value

4. Only then enhance with technology

This isn’t just manufacturing wisdom. It’s the difference between AI that transforms your work and AI that just makes bad processes faster.

The Three Types of Waste Killing Your AI Projects

Toyota identified three categories of dysfunction that destroy efficiency. In AI automation, these show up everywhere:

Muda (Waste) - Non-value-added activities that consume resources without benefit:

• Automating manual data entry when you should eliminate the data collection entirely

• Building AI workflows with 12 approval steps when 2 would suffice

• Creating elaborate automations for tasks that happen once per month

Mura (Unevenness) - Inconsistent workflows that prevent smooth operation:

• AI outputs that vary wildly in quality depending on input phrasing

• Automation systems that work perfectly on Tuesday but break on Friday

• Processes that depend on “John’s special way” of formatting the data

Muri (Overburden) - Pushing systems or people beyond reasonable limits:

• Automating complex analysis tasks onto AI models that aren’t designed for them

• Creating workflows so rigid that any edge case breaks the entire system

• Building automations that require more maintenance time than manual work

When you optimize manually first, you eliminate all three types of waste before automation amplifies them.

Subscribe now

The Manual-First Framework for AI Implementation

Here’s how to apply Toyota’s principle to your AI projects:

Step 1: Map the Manual Process

Do the task by hand 5-10 times. Track every step, every decision point, every handoff. Note where you hesitate, where you waste time, where the process breaks down.

For my meeting agendas, this revealed chaos: I was pulling information from six different sources, had no standard template, and couldn’t decide between strategic vs. tactical priorities.

Step 2: Eliminate Before You Automate

Ask ruthlessly: What parts of this process shouldn’t exist at all? What steps are pure bureaucracy? What information gathering is actually unnecessary?

I discovered half my agenda items were status updates that could be handled in Slack. Eliminating those immediately improved meeting quality—before any AI touched the process.

Step 3: Standardize What Remains

Create a repeatable playbook for the optimized process. If you can’t do it consistently manually, automation won’t fix the inconsistency. It will scale it.

My final manual process: strategic items first, open blockers second, quick tactical updates last. Time estimates for each section. Maximum 5 agenda items total.

Step 4: Then Automate

Now build your AI workflows around the perfected process. The technology enhances efficiency instead of amplifying dysfunction.

The AI system I built after manual optimization was night and day better. It automated a process that actually worked.

The Breakthrough Insight

Toyota’s real genius wasn’t automation. It was recognizing that bad processes automated become expensive bad processes.

When you optimize manually first, you’re not just preparing for better AI automation. You’re building the thinking patterns that make you better at everything: seeing waste, eliminating friction, designing for humans.

Let me be completely transparent: I learned this after building dozens of AI workflows that solved the wrong problems. Now I spend more time with google docs and strategy than I do with prompt engineering, and my automation success rate has skyrocketed.

Ready to Apply Toyota’s Method? Use This Tool

I’ve built a Custom GPT that walks you through Toyota’s exact process analysis methodology. It guides you through mapping your current process, then applies TPS principles to identify waste and create an improvement plan.

Access the Toyota Process Analyzer GPT

This tool will take you through the same analysis framework Toyota uses, but tailored for your specific business process. Perfect for applying what you just learned.

Want to see a sample before trying the tool? Here’s a sample conversation I had with it this morning.

Your Next Step

Pick one process you’re planning to automate this month. Before touching any AI tools, do it manually three times while narrating your decision-making out loud.

Don’t just track what you do—capture why you do it. Why did you prioritize that item? What made you choose this approach over that one? What context influenced your decision?

This narration becomes the gold that makes AI automation actually intelligent. Perfect the manual version until both the process and the reasoning flow effortlessly.

This mirrors the approach I wrote about in testing prompts manually before automating them - perfect the human process first.

Then and only then… let AI make it even better.

Because the fastest way to successful AI automation isn’t speed. It’s starting with a process worth automating.

It was the perfect one for turning rambling client feedback into clear action items. I’d used it three months ago and it worked brilliantly. But where the hell did I put it?

Notion? Nope. Google Doc? Not there. ChatGPT chat from April? Maybe... scrolling... scrolling... found it! Then copy, switch tabs, paste, tweak for context, finally send.

By the time I finished, I could have just written the response from scratch.

You have this problem too. We all do. That prompt graveyard where good ideas go to die.

The real problem isn’t that you forgot to save your prompts. It’s that there’s too much friction between having a prompt and actually using it. It’s the same reason you don’t drink enough water when the bottle is in another room, or why you buy junk food when you grocery shop hungry.

Lower friction equals higher adoption. For anything.

So how do you fix this without creating another system you’ll abandon?

The Three-Step System That Actually Works

Forget building 47 slash commands on day one. Here’s what actually works:

Step 1: Identify Your Repeat Patterns
Look for prompts you’ve used 3+ times in the last month. Not tasks you think you’ll do. Tasks you’ve already done multiple times.

Step 2: Build One Command
Pick the most frequent one. Turn it into a slash command. Use it 5 times. Refine it once based on what you learned.

Step 3: Add the Next One
Only after the first command becomes muscle memory. If you try to systematize everything at once, you’ll build a system you never use.

That’s it. No complex organization system. No fancy categorization. Just practical repetition.

What Slash Commands Actually Are

Slash commands are shortcuts. You type / in your AI chat, and an autocomplete menu pops up with your saved prompts. Pick one, hit enter, and the full prompt drops in instantly.

Tools like Prompster and 2Slash make this work across ChatGPT, Claude, Gemini, and more. You set them up once, and they’re there whenever you need them.

Think of it like having your ingredients ready before you start cooking. Chefs don’t scramble for spices when the pan is hot. Slash commands are your prep work. When you need something, you execute instantly instead of hunting for 15 minutes.

Before You Build Anything, Read This

Here’s my take: Most people shouldn’t use slash commands yet.

If you’re still figuring out what prompts work for you, slash commands are premature optimization. You need to run a task 5+ times before you actually know what’s worth systematizing. Otherwise, you’re just organizing prompts you’ll never use anyway.

Premature optimization is procrastination in disguise. Building the perfect system feels productive, but you’re just avoiding the actual work.

So if you’re still exploring what AI can do for you, bookmark this and come back later. Seriously.

But if you’ve ever reused the same prompt more than once, this is for you. You know what you need. You just don’t have a system for it yet.

Real Example: My Newsletter Workflow

I created a /newsletter-draft command that walks me through writing my newsletter. It’s not fully automated because my voice and judgment matter. But it handles the structure so I can focus on the thinking.

Here’s how it works:

1. References — I provide source material and inspiration

2. My angle — I share my unique take on the topic

3. AI expansion — It generates perspectives I hadn’t considered

4. Title selection — It gives me options, I pick one

5. Outline — It structures the piece

6. Draft — It writes 750-1500 words in my voice

7. Tighten — It cuts 15% without losing meaning

AI handles the scaffolding. I bring the voice and decision-making. That’s the balance.

But maybe you’re thinking: “I don’t know what I use repeatedly yet.” Fair point. Here are four commands that work for almost everyone:

Start With These Four Slash Commands

/meeting-summary

Analyze this meeting transcript and create a summary for stakeholders who weren’t present. 

Use exactly this format:

DECISIONS: [List key decisions made] 
ACTION ITEMS: [List with specific owners] 
OPEN QUESTIONS: [Items needing follow-up]

Keep each section concise and actionable.

Meeting transcript: [paste here]

Perfect for: Anyone who sits through more than two meetings per week.

/email-reply

Write a professional email response to the message below. 

Before writing, identify their main points, then craft a response that: - Uses a [friendly/direct/formal] tone 
- Addresses each main point clearly 
- Includes specific next steps 
- Stays under 150 words

Original email: [paste here]”

Perfect for: People who overthink email responses or find themselves rewriting the same types of messages.

/feedback-sandwich

Transform my raw feedback into constructive input using this exact structure:

STRENGTH: One specific thing they did well (with example) 
GROWTH AREA: One improvement area with actionable suggestion ENCOURAGEMENT: Positive note about their overall trajectory

Tone: Supportive but honest. Focus on specific behaviors, not personality.

Raw feedback: [paste here]

Perfect for: Managers, team leads, or anyone who gives regular feedback.

/social-post

Create a LinkedIn post from this idea using this structure:

HOOK: Attention-grabbing first line 
STORY/INSIGHT: Personal experience or practical insight 
VALUE: Key takeaway for the reader 
ENGAGEMENT: Question to spark comments

Voice: Conversational, practical, story-driven Length: 150-200 words

Original idea: [paste here]

This isn’t just about convenience.

Your prompts stop being hostages. They’re not trapped in a Notion page that might get deleted, or stuck in ChatGPT’s custom instructions that could change with the next update.

That’s the real win. Not speed. Control.

Another great thing about this? These work everywhere. Build a command in Prompster, and you can use it in ChatGPT, Claude, Gemini, or any new AI that comes out. Want to test the same prompt across different models? Just switch tabs and use the same slash command.

Your prompts become platform-independent.

Now let me show you a complex example to demonstrate why this matters for longer prompts.

Why This Works Even Better for Complex Prompts

The four starter commands above are helpful, but where slash commands really shine is with longer, more structured prompts.

Let me walk you through /newsletter-draft command I’m testing. This thing is 200+ words, has 7 steps, and multiple checkpoints. There’s no way I’d remember this or want to retype it every week.

My Full Newsletter Command

Here’s the prompt I use:

/newsletter-draft
Do these steps in order. STOP and wait for human input at marked checkpoints.

STEP 1: References & Inspiration ⏸️ [HUMAN INPUT REQUIRED]
• User provides: source material, links, examples, or inspiration content

STEP 2: Human Angles & Perspectives ⏸️ [HUMAN INPUT REQUIRED]
• User shares: their unique take, stance, or angle on the topic

STEP 3: AI Perspective Expansion
• Generate 5-7 additional angles or perspectives the user may not have considered
• Include: contrarian takes, audience-specific angles, timely hooks

STEP 4: Title Creation ⏸️ [HUMAN INPUT REQUIRED]
• Generate 10-12 title options grouped by tone: thought-provoking, practical, entertaining
• WAIT for user to select their preferred title

STEP 5: Outline
• Create an H2/H3 outline with bullets under each section
• 5-7 H2s, each with 2-4 H3 bullets

STEP 6: Drafting
• Draft 750-1000 words from the outline in the user’s voice
• Use the selected title, their angle, and chosen perspectives
• Friendly, direct tone; skimmable paragraphs; analogies where helpful

STEP 7: Tightening
• Tighten by ~15% without losing meaning, voice, or key examples
• Remove jargon and AI language

What this looks like when it works:
• Perspectives (Step 3): 5-7 distinct angles with brief rationale
• Titles (Step 4): 10-12 options in 3 tone groups, max 12 words each
• Outline (Step 5): 5-7 H2s, each with 2-4 H3 bullets
• Draft (Step 6): 500-1000 words, skimmable, in user’s voice
• Tightened (Step 7): 15% shorter, same punch

Does this create the perfect newsletter? No. But it gets me 80% of the way there, and I edit the rest. That’s the balance.

Setting Up Prompster (5 Minutes)

Step 1: Install the Extension

• Chrome: Search “Prompster” in Chrome Web Store

• Firefox: Available in Firefox Add-ons

• It’s free!

Step 2: Add Your First Command

1. Click the Prompster icon in your browser

2. Click “Add Prompt”

3. Shortcut: Type the slash command (e.g., newsletter-draft)

   

4. Prompt: Paste your full prompt text

5. Save

Step 3: Test It

1. Go to ChatGPT, Claude, or any supported AI

2. Type / in the chat box

3. See your commands appear in the dropdown

4. Select one and watch it populate the full prompt

Step 4: Build Your Library Start with these commands, then add your own:

• /meeting-summary

• /email-reply

• /feedback-sandwich

• /social-post

What About Text Expander?

You’re probably thinking: “What about Text Expander? That works everywhere on my computer.”

True. Text Expander works across all your apps because you download it directly to your computer. But here’s the catch: it’s not really a slash command system.

With Text Expander, you have to remember the exact shortcut name. If you set up ;meeting to expand your meeting summary prompt, you have to type ;meeting perfectly. There’s no autocomplete dropdown. No browsing through your commands. You either remember it exactly, or it doesn’t work.

That’s not a great experience when you have 10+ commands.

Browser vs. Desktop: The Real Tradeoff

Here’s the reality:

Prompster (Browser Extension)

• ✅ True slash command experience with autocomplete

• ✅ Browse through commands visually

• ✅ Works across ChatGPT, Claude, Gemini in browser

• ❌ Only works in browser (no Claude Code, Claude Desktop, etc.)

Text Expander (Desktop App)

• ✅ Works everywhere on your computer

• ✅ Includes desktop apps like Claude Desktop

• ❌ No autocomplete - you must remember exact shortcuts

• ❌ More expensive ($40/year vs. free/cheap browser extensions)

Which One Should You Choose?

Choose Prompster if:

• You primarily use AI in browser tabs

• You want the true slash command experience

• You like discovering commands through autocomplete

Choose Text Expander if:

• You use Claude Desktop, Claude Code, or other desktop AI apps frequently

• You don’t mind memorizing shortcut names

• You want text expansion beyond just AI prompts

For most people, I’d say start with Prompster. It’s free, easier to use, and covers 90% of AI work since most people use web interfaces. You can always add Text Expander or use built-in command features later if you find yourself needing desktop app coverage.

Your Next 15 Minutes

Don’t bookmark this and forget about it. Here’s what to do right now:

1. Install Prompster (1 minute) - Just get it in your browser

2. Pick one task (3 minutes) - What prompt did you use twice this week?

3. Create your first command (1 minute) - Use one of the four templates above

4. Test it immediately (1 minutes) - Open ChatGPT and try it

That’s it. Don’t build a library. Don’t plan your entire system. Just create one command that solves one real problem you had this week.

The goal isn’t perfection. It’s momentum.

If this command saves you time even once or gets you using prompts you thought you’d never use again? You’ll naturally want to create a second one. That’s when you know the system is working.

P.S. - Hit reply and tell me which command you created. I’m genuinely curious what patterns people are automating.

I knew I was meant to do more than this.

Not because I thought I was better than the work, but because I could feel my brain atrophying with each keystroke. I’m a lifelong learner, and this repetitive data entry was consuming every ounce of mental capacity I had left after eight hours of... more repetitive data entry.

That’s when I decided to automate myself out of the work I hated. And it worked better than I ever imagined.

Fast-forward fifteen years, and I see this same pattern playing out everywhere, but with a twist. The tools exist now to escape almost any kind of administrative prison. Yet most people are still trapped.

Someone discovers AI automation exists. They get excited about the possibilities—finally, a way out of the admin hell! So they jump straight to tools like Zapier, try to automate random processes, and when it doesn’t work the way they imagined, they give up, more frustrated than before.

What gives?

Here’s how reframing not just when but how technology, including AI automation, should be applied has allowed me to see it in a totally different light.

Subscribe now

So, in the next 10 minutes, you’ll discover exactly which questions to ask before touching any automation tool, a 4-step framework for escaping busywork, and three conversational prompts that will show you where to start.

But first, you need to answer the question most people never ask themselves:

What Do You Actually Want to Be Doing?

Stop for a second.

If you could spend all your time focusing on just two or three things, what would those things be? Not what you think you should be doing, not what your job description demands, and definitely not what would impress other people on LinkedIn.

What work actually energizes you?

What makes you lose track of time? What would you do even if no one was paying you? Maybe it’s strategic thinking and complex problem-solving. Maybe it’s building genuine relationships and having conversations that matter. Maybe it’s creative work where you can experiment, fail, iterate, and discover something new.

Or maybe it’s learning new skills and sharing what you discover with others, or leading teams and watching people grow into their potential.

Here’s what I’ve noticed: Many people, especially those running their own businesses or practices, can answer this question fairly quickly once they stop to think about it. They know what lights them up, what draws them in, what feels like play instead of work.

But then they spend most of their time doing everything else.

The Brutal Reality Check

Time for some honesty.

Look at your calendar from last week. Really look at it. Where did your time actually go?

I’m willing to bet a significant chunk went to administrative tasks that felt urgent but weren’t important, manual processes you could probably streamline if you took five minutes to think about it, and meetings that could have been emails (or better yet, eliminated entirely). Then there’s the context switching between tools and tasks, the “quick” requests that somehow turned into hour-long rabbit holes, and the work you’re doing simply because “it’s always been done this way.”

This is the gap that’s killing your potential.

It’s the space between what you want to be doing and what you’re actually doing. Every hour spent on busywork is an hour not spent on work that matters, work that energizes you, work that only you can do.

This gap between intention and reality is where most people get stuck. But here’s the breakthrough insight that changes everything: You don’t need to automate everything. You need to eliminate, streamline, delegate, and THEN automate.

The Escape Framework: Your Path to Freedom

Most people start with automation and wonder why it doesn’t work.

They think technology will magically fix their productivity problems, but what they’re really doing is automating dysfunction. You need to work through this hierarchy in order, and you can’t skip steps without paying the price later.

Think of it as a filter system. Each step eliminates more busywork with less effort than the next. Start at the top and work down until you find what sticks.

1. Eliminate: Stop Doing It Entirely

Ask yourself: Does this actually need to happen at all?

Some busywork exists because nobody ever questioned whether it’s necessary. That weekly status report nobody reads but everyone dutifully creates. The approval process for fifty-dollar expenses that costs more in time than the expense itself. The meeting to plan the meeting that could have been a five-minute conversation.

Permission granted: You have more freedom to stop doing things that don’t create value than you probably realize.

Sometimes the best automation is elimination, and elimination costs nothing but courage.

2. Streamline: Make It Stupidly Simple

Ask yourself: If I had to keep doing this, what’s the simplest possible way?

Before you automate a twelve-step process, see if you can turn it into a three-step process. Often you’ll discover that most of the complexity came from habit, not necessity. Layers of bureaucracy that accumulated over time like sediment.

Example: Instead of automating your complicated project kickoff process, maybe you just need a better template and a single shared workspace where everything lives.

For coaches: What if your intake process has fifteen steps because you’ve added something new each time a client had a unique situation? Before automating all fifteen steps, see if you can consolidate it into five essential ones that handle 90% of scenarios.

For consultants: That detailed project scoping process might feel necessary, but could you streamline it into a simple conversation plus a standard SOW template instead of multiple discovery calls and custom proposals?

Subscribe now

3. Delegate: Get It Off Your Plate

Ask yourself: Does this require my specific expertise, or could someone else handle it?

This doesn’t just mean hiring someone, though that’s one option. It could mean training a team member to own this type of work entirely, creating systems that let junior people handle routine decisions without escalation, or building processes that don’t require your constant input and approval.

For service businesses: Yes, teaching your VA to handle client onboarding will take 5-10 hours upfront. But that’s a one-time investment that frees up 2 hours every week going forward. Do the math—you break even in 3 weeks and save 100+ hours per year after that.

Reality check: If you’re the bottleneck for every decision, you’re not just trapped in busywork. You’re trapping everyone else too.

4. Automate: Let Technology Handle It

Only now do you look at automation. And by this point, you’ll know exactly what needs to be automated and why.

This is where AI and workflow automation actually shine: information that needs to move between systems automatically, routine decisions that follow clear rules, context that needs to be maintained across tools, and notifications that need to happen at specific triggers.

The difference: You’re automating something streamlined and essential, not just digitizing a broken process.

What to Expect (The Real Talk)

This sounds straightforward, but real life is messier than frameworks suggest. Here’s what nobody talks about.

Elimination can mean difficult conversations. That weekly client check-in call might feel like busywork to you, but your clients might expect that level of hand-holding. Sometimes elimination means resetting expectations you’ve already established, which requires courage and clear communication.

Delegation requires upfront investment. Training someone to handle your client intake process isn’t just a matter of handing over a checklist. You’ll need to document your decision-making process, create quality controls, and be available for questions while they learn. Plan on spending 5-10 hours upfront for every hour of recurring work you want to delegate.

Not everything can be automated immediately. Complex client relationship management, nuanced problem-solving, and strategic thinking might need to stay with you until you build systems that capture your judgment calls and institutional knowledge.

What if you discover conflicts? Maybe your energizing work is one-on-one coaching, but group workshops are more profitable. Maybe you love strategic consulting, but clients keep asking for tactical execution. Sometimes the framework reveals tensions between what you want to do and what your business needs, and that’s valuable information too.

Sometimes you’ll hit a wall, and that’s normal. If you can’t eliminate something, move to streamline. If you can’t delegate yet, start documenting processes anyway. If automation seems impossible, a streamlined manual process is infinitely better than chaos. If you discover you love “busywork,” maybe organizing IS your energizing work.

You don’t need to eliminate every task you don’t love. It’s about being intentional about where your time goes and systematic about reclaiming capacity for work that only you can do.

Your Breakthrough Moment Starts Here

Here’s something you can try right now.

Take ten minutes and have a conversation with yourself (or with AI if you prefer an external thinking partner). The goal isn’t to get perfect answers; it’s to start seeing the gap clearly and identifying your first moves toward freedom.

Here are three conversations that will show you exactly where to start:

Prompt 1: Clarity on What Matters

## You are an executive coach helping me identify the 2-3 types of work that truly energize me. I’m feeling trapped in busywork and need to distinguish between what I think I should enjoy and what actually energizes me.

## Your approach: Ask one question at a time, wait for my response, then ask follow-ups based on my answers. Focus on specific examples and emotional responses rather than abstract concepts.

## Key questions to explore:
-What specific work activities make me lose track of time?
-What accomplishments in the last year am I most proud of, and what about them felt most fulfilling?
-When I imagine having unlimited resources, what problems would I choose to solve?
-What type of work do I find myself naturally gravitating toward, even when it’s not required?

## End goal: Provide a clear summary of my top 2-3 energizing work types with specific examples from our discussion.

Prompt 2: Reality Audit

## You are a productivity analyst helping me audit how I actually spend my time versus how I want to spend it. I’ve identified my energizing work types and now need to understand where my time really goes.

##Your approach: Guide me through analyzing last week’s schedule. Ask specific questions about my calendar, then help me categorize activities and identify patterns.

##Framework to use:
-Review my actual time spent last week (calendar, tasks, interruptions)
-Categorize activities as: energizing work, necessary but not energizing, or pure busywork
-Identify what could be eliminated, streamlined, delegated, or automated

## End goal: Provide a breakdown showing current vs. ideal time allocation, plus the top 3 changes that would have the biggest impact on reclaiming time for meaningful work.

Prompt 3: Escape Plan

## You are an operations consultant helping me create a specific plan to reclaim time from one significant time drain this week. I’ve identified my energizing work and audited my time allocation.

##Your approach: Use the hierarchy method—first try to eliminate entirely, then streamline, then delegate, then automate. Only move to the next level if the previous isn’t possible.

## Process:
-Help me select one specific time-consuming activity from my audit
-Walk through each level of the hierarchy for this activity
-Identify obstacles and create a specific action plan with timelines

## End goal: Provide a concrete plan including the specific activity we’re addressing, which level applies, step-by-step actions, timeline, and success metrics.

Use these prompts with Claude, ChatGPT, or just as journal questions. What matters is that you’re finally asking the right questions instead of jumping straight to solutions.

Subscribe now

The Transformation: Getting Your Intelligence Back

Here’s what happens when you work through this framework:

You stop being reactive to whatever urgent-but-unimportant task lands on your desk and start being intentional about where your mental energy goes. You stop feeling guilty about saying no to requests that don’t align with your priorities and start protecting your capacity for work that actually creates value.

Most importantly, you stop automating random processes hoping for efficiency gains and start building systems that systematically free you for strategic work.

But the real transformation goes deeper than productivity.

You stop feeling trapped in your own success. Because that’s what busywork really is—the price you pay for being competent. People keep asking you to handle things because you’re good at handling things, but every “yes” to busywork is a “no” to the work you’re uniquely positioned to do.

The Choice That Changes Everything

You have a choice to make.

You can keep accepting that “this is just how work is,” keep telling yourself you’ll focus on strategic work “when things slow down,” and keep automating random processes hoping something will eventually stick.

Or you can start with the most important question: What do you actually want to be doing with your intelligence?

Then work backwards systematically: eliminate, streamline, delegate, automate.

The technology exists to free you from almost any repetitive work. The frameworks exist to help you identify what’s worth automating. The only question is whether you’re ready to be honest about what’s keeping you trapped.

Your next 24 hours: Pick ONE thing from your calendar last week that felt like pure busywork. Run it through the hierarchy: Can you eliminate it? Streamline it? Delegate it? Don’t try to fix everything. Just prove to yourself that you have more control than you think.

This week: Use those three prompts with AI or journal through them yourself. The conversation will show you exactly where to start.

This month: Pick your biggest time drain and systematically work through eliminate → streamline → delegate → automate until you get your time back.

Found this helpful? Forward it to someone else who’s trapped in their own version of invoice hell. Sometimes the most valuable thing you can do is help someone else see they have a choice.

Share

Why?

Most consultants are trapped in an impossible choice: spend 4+ hours crafting a perfectly tailored proposal that shows you truly heard their problems (but terrible ROI if you don’t close), or use a standardized template that feels generic and makes prospects think you didn’t listen to a word they said.

It’s the creative exercise dilemma - every proposal becomes this internal battle between “I need to show I understand their unique situation” and “I can’t afford to spend another full day on a proposal that might not convert.”

I used to be one of those people. Until I sent a proposal with the wrong client name in the header—it went to their biggest competitor instead. The client called me out on it, and I lost the deal.

That embarrassing mistake forced me to ask a different question: What if proposals weren’t individual tasks but part of a systematic process?

That’s why I built this proposal system so that every proposal feels completely customized because it’s built from the actual conversation insights, and they also look professionally designed.

Here’s the progression:

• Tier 1: Perfect your prompt foundation (start here this week)

• Tier 2: Automate document creation (scale what works)

Here's what you'll get in this post: 

➡️ Proposal creation prompt
➡️ Learn how to create a beautiful proposal in Gamma
➡️ CustomGPT setup instructions
➡️ A make.com automation to create a google doc directly from ChatGPT
➡️ Walkthrough videos of the entire process

Speed matters, but so does personalization AND visual impact. You need all three. Generic templates are fast but impersonal. Custom proposals are personal but slow. Beautiful proposals take hours with designers. This system gives you all three.

Remember in Your Meeting Transcripts are Worth More Than You Think, I showed you how transcripts are goldmines of insight? This system turns that insight into your competitive advantage.

Why Systems Beat One-Off Tasks Every Time

Before I show you the system, let’s understand why building systems always wins:

System Advantage #1: Compound Learning Each proposal teaches the system something new. Your 50th proposal is exponentially better than your 1st because the system retains and applies everything it learned.

System Advantage #2: Consistent Quality No more “good proposal days” vs “rushed proposal days.” The system delivers your best work every time, regardless of pressure or timeline.

System Advantage #3: Scalable Personalization You can handle 10x more prospects without sacrificing customization quality. The system does the heavy lifting while you focus on relationship building.

System Advantage #4: Zero Context Loss Everything important from the sales call flows directly into the proposal. No more staring at notes wondering what mattered to them.

This system gives you all four advantages by treating proposals as a systematic process, not individual creative tasks.

Subscribe now

The Proposal System: A 2-Tier System for Speed, Personalization, and Visual Impact

Now let’s dive into the actual system. This framework isn’t about replacing human insight with robots. It’s about building a systematic process that captures your expertise and applies it consistently, at scale, while creating proposals that look professionally designed.

Each tier builds on the previous one, creating a progression from manual customization to full automation. You can start at Tier 1 this week and scale up as your needs grow.

Tier 1: Prompt Perfection (Foundation Layer)

The Problem Tier 1 Solves: You’re starting from scratch every time instead of extracting the gold already sitting in your sales call transcripts.

Why This Tier Matters Most: This is your foundation. If you don’t perfect your prompt at this level, automating it later just means faster garbage output. Garbage in, garbage out. But perfect this prompt and you’re ready to scale it through the entire framework.

The System: Instead of relying on memory and notes, you develop and refine a systematic prompt framework that extracts specific insights from transcripts and transforms them into customized proposal sections.

Here’s where the magic happens: you can optimize this same prompt output for either traditional documents OR visually stunning presentations.

The Enhanced Transcript Transformation Prompt:

# ROLE
You are an expert proposal writer specializing in creating highly personalized, compelling business proposals. You excel at extracting key insights from sales conversations and transforming them into client-focused proposal content.

# TASK
Create a tailored business proposal based on the sales call transcript provided below. Your goal is to make the prospect feel heard and understood by using their exact language and addressing their specific situation.

# RULES
- Use the prospect’s exact terminology and phrases wherever possible
- Never invent or assume information not present in the transcript
- Focus on outcomes and value, not just features or deliverables
- If critical information is missing, note what additional details are needed
- Maintain a professional but conversational tone

# HUMAN FEEDBACK CHECKPOINT
Before I provide the transcript, I need you to ask me:
1. What is your industry/area of expertise?
2. What is your typical project timeline range?
3. Any specific positioning or competitive advantages I should emphasize?
4. Should I optimize this for traditional document format or visual presentation (Gamma)?
5. Are there any company-specific details or constraints I should know?

# OUTPUT FORMAT
Create the following sections:

**EXECUTIVE SUMMARY** (2-3 sentences):
- What specific situation they described
- Why it matters to their business
- How you uniquely solve it

**PROBLEM STATEMENT**:
- Use their exact language for pain points
- Reference specific examples they mentioned
- Show you understand their constraints and urgency

**SOLUTION STATEMENT**:
- Connect your approach to their specific needs
- Address their exact concerns from the call
- Use terminology and frameworks they referenced

**TIMELINE & APPROACH**:
- Account for deadlines they mentioned
- Reference their decision-making process
- Include milestones that matter to them

**INVESTMENT & NEXT STEPS**:
- Address budget constraints they discussed
- Reference their approval process
- Create appropriate urgency around timing they mentioned

# ADDITIONAL CONTEXT SPACE
[Space for any additional context about the prospect, industry considerations, or specific requirements]

# TRANSCRIPT INPUT
--- PASTE SALES CALL TRANSCRIPT BELOW THIS LINE ---

[TRANSCRIPT CONTENT]

# FINAL CHECK
After generating the proposal, ask:
1. What additional context should I incorporate based on your relationship with this prospect?
2. Are there any industry-specific considerations I should address?
3. Should I adjust the tone or positioning based on the decision-maker(s) involved?

Why This Tier Works: You’re systematically extracting everything important instead of relying on memory. The framework ensures you capture their exact language, specific concerns, and stated priorities.

The Human-AI Collaboration Process: This isn’t about replacing your expertise. It’s about scaling it. Here’s how the human-in-the-loop works:

1. AI generates the first draft using transcript insights and your prompt framework

2. You add context AI can’t access: Your years of experience, industry knowledge, and relationship insights

3. AI helps refine and polish your additions into client-focused language

4. You make final strategic decisions about positioning, pricing, and next steps

Example of Human Context Injection: AI might extract “They need better lead tracking” from the transcript. You add: “Based on similar clients in fintech, this usually indicates a 30-40% revenue leak from poor follow-up processes. For a company their size, that’s potentially $200K annually.”

The Visual Differentiation Secret: Gamma Export

While your competitors are sending boring Word documents, you can create visually stunning proposals that look like they were designed by a professional agency… in seconds.

The Gamma Advantage: Take your perfected prompt output and export it optimized for Gamma.app. Instead of plain text proposals, you now have:

• Professional visual design that immediately differentiates you

• Interactive presentations that engage prospects

• Zero design time - beautiful proposals generated instantly

• Premium positioning - you look more expensive before they read a word

Proposals with visual elements close 72% faster (Proposify). Most consultants think they need to hire designers or spend hours formatting. You just need to export your prompt output to Gamma.

The Choice Point: Perfect your prompt once, then decide:

• Path A: Export for Google Docs (traditional workflow)

• Path B: Export for Gamma (visual impact workflow)

Once you’ve mastered Tier 1 and your prompt consistently delivers excellent results, you’re ready to eliminate the manual copy-paste work entirely.

I’m curious, are you team “traditional document” or team “visual presentation” when it comes to proposals? And more importantly, what’s been your biggest proposal headache that you’d love to systematize?

Drop a comment and let me know what part of this framework you’re most excited (or skeptical) about implementing.

Leave a comment

Tier 2: Custom GPT Packaging (Professional Layer)

The Problem Tier 2 Solves: Your Tier 1 prompt is working perfectly, but you’re still manually pasting transcripts into ChatGPT, re-explaining your context every time, and losing the nuances of your consulting approach between sessions.

Why This Builds on Tier 1: Now that you have a proven prompt that consistently generates great proposal content, you can package it into a reusable system that remembers your methodology, maintains your voice, and applies your expertise automatically. You’re scaling the perfected system, not automating broken processes.

The System: A Custom GPT that encapsulates your perfected prompt framework, stores your proposal approach, and creates consistent output every time—without you needing to re-explain your process.

Custom GPT Setup Process:

1. Core Prompt Integration: Take your refined Tier 1 prompt and build it into the GPT’s instructions, including all the quality checks and client-focused frameworks you’ve perfected.

2. Knowledge Base Upload: Add your best proposal examples, industry-specific terminology, common objection handlers, and pricing frameworks. The GPT learns from your actual work, not generic templates.

3. Action Schema: This tells ChatGPT what information you want to send to Make.com for document creation

What This Actually Looks Like:

Instead of: Opening ChatGPT → Pasting transcript → Copying/pasting your prompt → Explaining context → Generating proposal → Formatting output

You get: Opening your Custom GPT → Pasting transcript → Answering 4 pre-loaded questions → Receiving formatted proposal in your voice

Human Oversight Built In: The Custom GPT approach maintains your strategic control:

• Context preservation: The GPT remembers your methodology between sessions, but still asks for prospect-specific details

• Voice consistency: Your refined prompt ensures every proposal sounds like you, not generic AI

• Quality checkpoints: Built-in prompts to review key sections before final generation

• Easy refinement: Update the GPT’s instructions as you learn what works better

Get the CustomGPT Files

Why This System Approach Changes Everything

Scalable Personalization: Handle 10x more prospects without sacrificing customization quality. Each proposal feels completely tailored because it’s built from actual conversation insights.

Compound Learning: Every proposal makes your system smarter. Your 50th proposal incorporates lessons learned from the previous 49, creating exponentially better results over time.

Professional Consistency: No more “good days” vs “rushed days” in proposal quality. The system delivers your best work every time, regardless of pressure or timeline.

Strategic Advantage: While competitors choose between generic templates or manual burnout, you deliver tailored proposals faster than they can send generic ones.

Why Transcript-Based Proposals Win

Last week in Why Your Sales Team Can’t Actually Sell, we saw how slow proposal processes kill deals. But here’s the deeper insight: the quality of your input determines the quality of your output.

Your sales call transcripts are the highest-quality input you could possibly have for proposal generation. They contain:

• The prospect’s exact pain points in their own words

• Specific examples and use cases they care about

• Budget constraints and decision-making processes

• Timeline pressures and success metrics

• Objections and concerns that need addressing

When you build proposals from this foundation, prospects don’t just read them—they feel heard. That’s the advantage: speed, personalization, and visual impact all working together.

Subscribe now

Start Building Your System This Week: Your 7-Day Implementation Plan

Don’t wait for the perfect automation setup. Here’s your concrete roadmap to launch Tier 1 this week:

Days 1-2: Foundation Setup

1. Gather your last 3 sales call transcripts (from Fathom, Fireflies, Otter, etc.)

2. Copy the Transcript Transformation Prompt exactly as written above

3. Test with your oldest transcript first (less pressure, more learning)

Days 3-5: Refinement Cycle

4. Generate 3 proposals using the systematic prompt and note what works/doesn’t work
5. Send one real proposal and track client response to specific sections
6. Refine your prompt based on client feedback and missing elements

Days 6-7: System Validation

7. Create one more proposal with your refined prompt to confirm improvements
8. Document your prompt refinements for consistent future use
9. Only consider Tier 2 once your prompt consistently delivers client-ready results

Critical Success Factor: Don’t rush to automate a mediocre prompt. Spend time in Tier 1 perfecting the foundation. Every refinement you make here compounds through the entire Proposal Velocity Framework.

System Mindset Shift: Think of each proposal as training data for perfecting your prompt, not just a one-off task. Every transcript teaches your framework something new about effective customization.

Remember: Garbage in, garbage out. Perfect the input (your prompt) before scaling the system.

Found this helpful? Forward this to someone who’s still spending 4+ hours on every proposal (they’ll thank you later).

Share

P.S. - Every transcript sitting in your Fireflies account is training data for a better proposal system. Stop treating proposals as creative writing exercises and start building the systematic process that scales your expertise.

10:47 PM. I’m manually checking my eighth Substack profile to see if my favorite Substackers have posted.

Not because I don’t get email notifications—I do. All 847 unread emails sitting in my inbox right now can confirm that. I’ve trained myself to ignore most notifications, which means I’m missing content from writers I actually want to read while drowning in digital noise.

So instead, I’m clicking through individual Substack profiles. One by one. Like it’s 2015.

The math is embarrassing: I’m spending 20-30+ minutes a day just hunting for content, not reading it. That’s 4+ hours a week on newsletter archaeology. And I still feel like I’m missing stuff.

Worse: I’m spending more mental energy worrying about what I’m missing than actually consuming anything useful.

So, what did I do?

Implement a curated RSS system that took 30 minutes to set up. Now I process everything in 60 minutes a week instead of 4+ hours, get instant notifications for voices that matter, and see all my Substack subscriptions, YouTubers, and other content creators in one unified feed.

I’m obsessed with AI automation but also believe not everything needs a complex AI setup. Here’s how you can use AI to set up a proven tool (RSS Reader) to solve your overwhelm problem.

Here’s how it works 👇🏼

Why Substack’s Interface Creates Overwhelm

I follow 129 publications on Substack and add more every day. It’s gotten out of hand. But there are so many great voices to follow, so what do you do?

First, Substack treats all newsletters equally. There’s no way to prioritize authors. No filtering. No hierarchy.

Secondly, When everything feels urgent, nothing feels manageable. You end up reading less because you’re paralyzed by choice.

The irony: having access to more content makes you consume less of it.

The solution isn’t a fancy automation or prompt engineering mastery. It’s using AI to handle the setup work, then letting proven RSS tools handle the curation.

Results and Benefits Achieved

Setup took 30 minutes. I went from 4+ hours of scattered reading per week to 60 focused minutes. More importantly, those 60 minutes feel intentional instead of reactive.

Unexpected wins:

• Found newsletters I’d forgotten I was subscribed to (and started reading or finally unsubscribed)

• Started seeing cross-connections between different authors’ ideas

• Reduced FOMO while actually consuming more quality content

• More likely to engage meaningfully because the friction of discovery is gone

• Bonus, some RSS readers have AI features that can help you consume the right content even faster

What RSS Actually Is (And Why It Solves This)

RSS feeds are like getting a table of contents for all your subscriptions in one place, instead of checking each site individually.

Think of it as a unified inbox for content from anywhere on the web. You can see when Substack Writer A posted something new right next to when YouTuber B uploaded a video, all sorted by your priorities.

It changed how I see connections between ideas and I can scan for interesting content without digging through each platform’s UI, then click through to the original post for commenting and engagement.

Another key difference: instead of Substack’s interface treating everything equally, you control the hierarchy. High-priority authors get immediate attention. Everything else gets sorted for later.

RSS has been around forever, but most people stopped using it when social media took over. Now that social feeds are algorithmic chaos, RSS is having a comeback - especially for people who want intentional content consumption.

Step-by-Step Implementation

Step 1: RSS Feed Collection (Choose Your Method)

You have two paths here, depending on your comfort level and access with AI browsers.

Option A: Use Comet (Dia, etc.) browser AI for automated extraction across all Substack subscriptions. Simply navigate to your Substack subscription page in your browser, tell Comet to “extract all RSS feed URLs from my subscriptions,” and let it handle the tedious work.

Option B: Manual method for those who prefer hands-on control—append “/feed” to any Substack URL (e.g., example.substack.com/feed) and collect them yourself. Either approach works; create an organized list of all feeds for the next step.

Step 2: RSS Reader Selection and Setup

Choose an RSS reader with robust filtering capabilities. Feedly offers a solid free tier, while Inoreader provides better filtering options for power users. RSS.app has been a long-time player as well. There are plenty of readers out there - I started with InoReader and like it so far.

It’s free if you want a basic feed but want to filter with rules and follow websites without RSS feeds? You’ll need to upgrade.

Once you upgrade you can also set up a logical folder structure: “High Priority” for must-read voices, “Browse When Time” for interesting but not urgent content, and topic-based folders for different areas of interest.

Step 3: Creating Author Priority Systems

This is where the magic happens. Categorize authors by importance, immediate attention versus browse-when-time and tag by topic. Set up filtering rules based on author hierarchies and keywords so high-priority content gets flagged immediately.

Configure notifications only for must-read voices to avoid recreating the notification overwhelm you’re trying to escape. The goal is intentional consumption, not reactive scrolling.

Step 4: Maintaining Platform Engagement

Here’s the crucial habit: always click through to original Substack posts from your RSS reader when you find valuable content. Comment and like directly on the platform to maintain author relationships and support their work. Use RSS for discovery and organization, but Substack for meaningful engagement. This hybrid approach gives you the best of both worlds: organized consumption with authentic community participation.

When NOT to Use This System

If you’re subscribed to fewer than 15-20 newsletters, you probably don’t need this. Just unsubscribe from the ones you don’t actually read.

If you prefer Substack’s discovery features and comment communities, or you genuinely enjoy browsing the native interface, stick with what works.

Sometimes the solution isn’t optimization - it’s elimination. If you’re constantly organizing information instead of consuming it, you might just have too many subscriptions.

Tools and Resources

Primary Tools

Comet: AI browser for RSS feed extraction. This handles the tedious work of collecting feed URLs from your subscription lists automatically. I pay for Perplexity so I just use this.

RSS Reader Options: Feedly offers a solid free tier that covers basic filtering and organization. Inoreader provides more sophisticated filtering capabilities for power users who want granular control over content curation.

Optional Notification Platforms: Slack, Telegram, or Discord for advanced automation alerts if you want instant notifications for high-priority authors.

Setup Requirements

The beauty of this system is its accessibility. You need basic familiarity with browser AI tools (or willingness to learn), but nothing technical. The entire initial setup and testing should take around 30-60 minutes for most people.

Advanced notification setups are completely optional. The core RSS reader functionality provides 90% of the value without any webhook knowledge or complex integrations.

Cost Considerations

Most RSS readers offer free tiers that handle moderate subscription volumes perfectly well. Comet is free if you are paying for Perplexity but a reasonable $5/month if you’re not. The total monthly cost for most users ranges from free (using manual RSS collection + free RSS reader) to under $15 for full automation, AI browser and RSS with premium features.

PRO TIP: Ask your AI browser to find a working promo code and apply it. It’ll run off and find one and do all the work for you. Thanks to for sharing this hack 🙌🏼

Try it Yourself

Start small. Extract RSS feeds from your top 10 most important Substack subscriptions using either your AI browser of choice or manual method outlined above. Set up a basic RSS reader with simple folder organization. Test the system for a week before expanding to your full subscription list.

The key is proving to yourself that this reduces cognitive overhead before committing to the full migration.

Hit reply with any challenges you encounter during setup. Real-world implementation issues usually spark the best follow-up content and your problem is probably someone else’s problem too.

While AI and automation are wonderful tools - theyr’e just that. Tools. And sometimes the best tool for the job isn’t AI or automation at all.

-Tam

“Can you give us a ballpark on pricing?”

And there it is. The moment where I lose credibility.

“I’ll need to check with my team and get back to you.”

But here’s what’s really happening in my head: Shit. I’m supposed to be running sales and I can’t even give a range.

And here’s what they’re thinking: How senior is this person if they can’t discuss pricing?

I watch their enthusiasm deflate in real-time. Weeks running sales, and I still can’t answer the most basic question… what do we charge?

I try to recover: “Every engagement is unique, we want to ensure we’re giving you accurate—”

But they’re already mentally dialing my competitor who gave them pricing on the call.

The partner I need to run pricing by? In back-to-back meetings until Thursday.

By the time I get pricing approved and send the proposal, it’s been four days. The prospect has already signed with someone else.

“Sales is an art,” the partners keep telling me.

Cool.

An art where only three people in the company know how to paint, and they’re too busy to teach anyone else.

The Universal Pricing Paralysis

I come from accounting, so that’s where my examples live. But whether you’re pricing audit work, marketing campaigns, website designs, or IT implementations — we all know this pain.

Your sales team is sitting there with perfect prospects, and they can’t actually sell because they’re not authorized to talk money.

Meanwhile, the people who CAN talk money are too busy doing work to talk money...

Subscribe now

The Hidden Cost of “It Depends”

Let me show you what this actually costs. I tracked this for a month:

Day 1 - The Call:

• Great conversation, prospect is engaged

• They ask for pricing

• I promise to follow up

Day 2-3 - The Chase:

• Draft email to partner with context

• Wait for response

• Follow up on Slack

• “Sorry, in meetings, let’s sync tomorrow”

Day 4 - The Education:

• 30-minute call to explain the prospect

• Partner asks 20 questions I didn’t think to ask

• They promise to think about it

Day 5 - The Pricing:

• Finally get a number

• But now I need to build the proposal

• Partner wants to review before I send

Day 6-7 - The Death:

• Send proposal

• Prospect: “Thanks, we went with someone else”

Seven days. For one proposal. While my competitor sent theirs on the call.

The real cost?

• Lost deals (obviously)

• Demoralized sales team

• Partners stuck in pricing meetings instead of strategy

• A business that can only grow as fast as the founder can price

It’s Not Art. It’s Just Patterns Nobody Wrote Down.

Here’s what I discovered after a month of banging my head against this wall: the “artistic” pricing isn’t artistic at all. It’s pattern recognition that lives in one person’s brain.

The founder who “just knows” what to charge? They’re subconsciously running these calculations:

In accounting (my world), it looked like:

• Company under 10 employees? Different complexity than 50+

• QuickBooks vs NetSuite? Totally different beast

• 500+ monthly transactions? Everything changes

• Multiple entities? Each one adds complexity

In your world, it might be:

• Startup clients scope creep more than enterprises

• Committees take 40% longer than single decision makers

• “Urgent” projects are never actually urgent

• Regulated industries need 3x the documentation

The founder isn’t doing mystical pricing divination. They’re pattern-matching based on hundreds of past deals.

They just never wrote the patterns down.

The Knowledge Extraction Framework

After a month of being a glorified email forwarder, I decided to solve this.

Not with more meetings.
Not with complicated approval chains.
But by extracting the tribal knowledge and making it usable.

Here’s the framework that actually works:

Step 1: Record the Expert (1-2 hours)

Don’t make them write anything.
Writing feels like work.
Talking is easy.

Record them pricing 5-10 recent deals. Ask:

• “Walk me through how you priced this”

• “What made you go higher/lower here?”

• “What red flags changed the price?”

• “What would have made this a no-go?”

Step 2: Extract the Patterns (30 minutes with AI)

Feed the transcripts to AI with this prompt structure:

You are a pricing strategist specializing in pattern recognition and knowledge extraction. Your goal is to transform informal pricing conversations into systematic, repeatable frameworks that others can follow.

Analyze the pricing conversation transcript below to extract repeatable patterns.

Identify:

1. PRICING FACTORS
- What increases price (be specific: thresholds, triggers, conditions)
- What decreases price or triggers discounts
- Exact multipliers mentioned (rush = X%, complexity = Y%)

2. DECISION LOGIC
- If/then statements the expert uses
- Questions they ask to determine pricing
- Walk-away conditions or red flags

3. PATTERNS
- Standard base pricing mentioned
- Common adjustments and why
- Edge cases and how they’re handled

[Insert transcript]

Output a practical decision tree starting with the most common scenario, then branch into variations. Include specific numbers/percentages when mentioned. Focus on creating rules others can follow, not just observations.

Step 3: Document the Context (30 minutes with AI, again)

This is the secret sauce. Create a knowledge document with:

Value Drivers:

• Industry complexity factors

• Company size breakpoints

• Timeline impacts

• Technical requirements

• Risk factors

Pricing Multipliers:

• Base rate for standard engagement

• Complexity adjustments

• Rush charges

• Volume discounts

• Industry premiums

Decision Logic:

• If [situation], then [pricing action]

• If client says [X], add [Y]%

• If timeline is [Z], multiply by [factor]

Here’s what a sample context document looks like:

PRICING FRAMEWORK - [COMPANY NAME]

STANDARD ENGAGEMENT:
- Base: $15,000 (companies under 50 employees)
- Base: $25,000 (companies 50-200 employees)  
- Base: $45,000+ (enterprise 200+ employees)
- Timeline: 4-6 weeks standard delivery

PRICING ADJUSTMENTS (add to base, not compounded):
- Rush delivery (under 2 weeks): +50%
- Complex tech stack (5+ systems): +30%
- Regulated industry (healthcare/finance): +40%
- Multiple stakeholders (5+ decision makers): +20%
- Weekend/holiday work required: +75%
- Multiple entities: +$5,000 per additional entity

DISCOUNTS (subtract from total):
- 3+ project bundle: -15%
- Nonprofit: -20%
- Referral from existing client: -10%
- Flexible timeline (8+ weeks): -10%

HOW TO CALCULATE:
1. Start with base price for company size
2. Add all applicable percentage adjustments together
3. Apply total adjustment to base price
4. Add any fixed-dollar adjustments
5. Apply any discounts to final amount

EXAMPLE:
- 75-person company (base: $25,000)
- Rush + regulated = 50% + 40% = 90% additional
- $25,000 × 1.9 = $47,500

RED FLAGS (DON’T QUOTE WITHOUT PARTNER):
- More than 2 system migrations in past year
- “Previous vendor couldn’t deliver”
- Requesting guarantees on outcomes
- Wants to pay after results

AUTOMATIC WALK-AWAYS:
- Budget under $10,000
- Timeline under 1 week
- More than 50% payment on completion
- Unlimited revisions requested

This becomes your single source of truth that AI can reference and humans can verify.

Step 4: Build the System (30 minutes)

Basic: Context doc + ChatGPT for pricing recommendations

Better: Custom GPT with your pricing matrix built in

Best: Custom GPT → automation tool → proposal template

Here’s the key system prompt for your Custom GPT:

You are a pricing advisor for [Company Name], trained on our 
specific pricing patterns and methodologies. Use the context document 
as your primary reference for all pricing decisions.

When analyzing a sales opportunity:

1. EXTRACT KEY INFORMATION
- Review the call transcript for pricing-relevant factors
- Ask the salesperson for any missing critical details
- Map the opportunity to our documented patterns

2. GENERATE PRICING ESTIMATE
- Start with base price for company size
- Add percentage adjustments (sum them, don’t compound)
- Show calculation: “Base $X + Y% + Z% = $Total”
- Example: “$25,000 base + 50% rush + 30% complexity = $25,000 × 1.8 = $45,000”

3. EXPLAIN YOUR REASONING
- Reference specific patterns from our framework
- Highlight which factors most influenced the price
- Note confidence level (high/medium/low) based on pattern match

4. FLAG EDGE CASES
🟡 YELLOW FLAGS (proceed with caution):
- Factors present but not in our documentation
- Unusual combinations of requirements
- Client requests that partially match patterns

🔴 RED FLAGS (require human review):
- Complete absence of pattern match
- Multiple conflicting factors
- Requests explicitly outside documented scope
- Any mention of: [legal terms, guarantees, liability]

ALWAYS END WITH:
“This estimate is based on [X] pattern with [Y] confidence. 
Yellow flags: [list any] | Red flags: [list any]
Recommend human review if confidence is below 80%.”

The genius of this system: It handles the 80% of standard deals automatically while intelligently escalating the 20% that need human judgment. Your sales team can move fast without breaking things.

Real Results From Real Implementation

When I implemented this:

• Pricing time went from 7 days to 30 minutes

• Sales team could quote ranges on calls

• Partners freed up 5+ hours per week

• Close rate increased (speed matters)

But here’s the thing — this framework works for ANY trapped knowledge:

• Sales qualification criteria

• Project scoping

• Client red flags

• Technical requirements

• Risk assessment

Once you see how to extract one type of knowledge, you can extract them all.

The Founder’s Fear (And Why It’s Backwards)

“But what if they price it wrong?”

Let me tell you what’s actually wrong: losing deals because you’re the bottleneck.

You’re not protecting your business by being the only person who can price. You’re capping it at exactly the size of your availability.

Your sales team isn’t failing. They’re paralyzed.

And that paralysis is a choice you’re making every day you don’t document your patterns.

Your Next Move

If you’re a founder who’s the pricing bottleneck, you have two options:

Option 1: Keep being the hero. Review every proposal. Lose deals to faster competitors. Watch your best salespeople leave for companies that actually let them sell.

Option 2: Spend 4 hours extracting your patterns. Build a system that preserves your expertise while freeing your team. Scale beyond your personal capacity.

If you’re in sales or operations, you can wait for your founder to figure this out. Or you can build the system yourself and become the person who unlocked growth.

The Question That Matters

How many deals are dying in your pipeline right now because someone’s waiting for the only person who can price?

That’s not a sales problem. That’s a knowledge hoarding problem disguised as “art.”

And now you know how to solve it.

Next week: Once you can price instantly, how do you automate the entire proposal process?

I’ll show you the system that generates proposals in minutes, not days.

Subscribe now

Ready to stop being the bottleneck?

I help founders extract and systematize their trapped knowledge so their teams can actually do their jobs. If you’re tired of being the pricing approval department, let’s talk.

Book a diagnostic call →

Every Thursday, I share strategies for automating the soul-crushing work that’s killing your evenings. Join hundreds of operators who are done being heroes.

The creator had 5 n8n1 workflows.

AI writing blog posts, auto-publishing to twelve platforms, even generating newsletter responses.

"Set it and forget it", "Content creation on autopilot." he promised.

I spent three days trying to build something similar.

Here's the thing: I went deep on n8n. Self-hosted instance. API tokens for OpenAI2, Anthropic3, Perplexity4. Webhook debugging.

I even had a spreadsheet tracking my cost-per-automation.

I thought I was being smart. Cutting out the middleman. Building "real" systems.

But the real cost wasn't the price or even the complexity.

It was that I was automating mediocrity.

My automated posts were getting 3 likes. My AI-generated newsletters had 12% open rates. I'd built a factory that produced content nobody wanted to read.

I was automating workflows before knowing what actually worked.

Like building a highway before deciding where it should go.

The Automation Paradox Killing Your Output

Here's what nobody tells you about automation: It amplifies whatever you feed it.

That newsletter prompt that's "pretty good"? Automate it and you'll get 52 pretty good newsletters that nobody finishes reading.

That LinkedIn template that "works okay"? Scale it and watch your engagement flatline while you debug why Tuesday's post has Friday's CTA.

I spent two weeks building a content repurposing pipeline. One blog post would automatically become:

• 5 LinkedIn posts

• 10 tweets

• 3 newsletter sections

• 1 YouTube script

Impressive, right?

Except my original blog posts were generic. So my automation produced 19 pieces of generic content from each generic source. I'd built a garbage multiplier.

Meanwhile, my competitor was manually crafting one post per week with ChatGPT. Testing headlines. Adjusting hooks. Learning what resonated.

Guess whose content was actually driving revenue?

The automation paradox: The better your system, the faster you produce content nobody wants.

Subscribe now

The 10-Test Rule That Changed Everything

Three months ago, I killed all my automations.

Instead, I started following one simple rule: Test every prompt manually 10 times before even thinking about automation.

Want to automate newsletter titles? Write 10 manually first.

Notice which words make you click.
Understand why some angles work.
Feel the difference between Tuesday energy and Thursday energy.

Here's what happened with my newsletter titles:

• Tests 1-3: Generic garbage ("5 AI Tips for Founders")

• Tests 4-6: Found my voice ("The $200K Consultant Living in Your ChatGPT")

• Tests 7-9: Discovered my pattern (specific cost/benefit + unexpected angle)

• Test 10: Knew exactly what to prompt for

Only after those 10 tests did I create my GPT Project (like last Thursday's system).

No automation. Just context-rich AI that knows my business.

The result? My open rates jumped from 12% to 37%.

Not because I automated faster. Because I understood what worked before I scaled it.

Most people are automating their first draft thinking. That's why their content sounds like everyone else's.

Your Move: One Prompt, No Automations, 48 Hours

Here's my challenge for you:

Pick ONE task you're desperate to automate. Newsletter titles. LinkedIn hooks. Email responses. Whatever's eating your time.

For the next 48 hours, do it manually with AI assistance. But here's the twist:

• Document every prompt you use

• Note what outputs actually work

• Save the ones that flop

• Pay attention to what changes between morning and afternoon outputs

After 10 iterations, you'll know:

• If this task even needs automation

• What context the AI actually needs

• Which prompts create value vs. noise

Most tasks won't need automation. They need better prompts and clearer context.

That's your competitive advantage: While everyone's building elaborate automation systems, you're perfecting the one prompt that actually matters.

Start today.

One task. Ten tests. No automations.

Watch what happens to your output quality.

The Bottom Line

Stop automating what you don't understand.

Start with one perfect prompt. Test it until it sings. Then - and only then - consider if automation even makes sense.

Your competition is busy debugging their 47 workflows, wondering why their API calls are failing, and producing content that sounds like it came from 2023's ChatGPT.

You'll be shipping content that actually converts.

Because you took the time to understand what works before you tried to scale it.

Subscribe now

It's making you a tool operator instead of a strategic thinker.

The tools aren't the problem. Starting with the tools is.

P.S. - Still want to automate? Start with last Thursday's GPT Projects system. Build context first. Automation maybe. Most likely? You'll find you don't need half the automations you thought you did.

Hi, I’m Tam. When I started my career, I was manually keying invoices 8 hours a day with a college degree on the wall. Today, I help people build AI automation that scales their expertise without scaling their hours.

Sometimes I was the bottleneck—the “just ask Tam” person who couldn’t take vacation. Sometimes I was drowning in repetitive tasks that ate all my thinking time. Sometimes I’d built something successful but couldn’t figure out how to grow without burning out.

You might recognize one of these patterns. Or maybe you’re just curious about what AI automation could actually do for your work.

Maybe this resonates with you. Or maybe your challenge looks completely different.

Here’s what I’ve learned: The challenge isn’t always being stuck. Sometimes it’s knowing that you could amplify what you’re good at, but not knowing where to start.

The real question isn’t whether you’re failing. It’s whether you’re ready to scale what’s already working.

Let me tell you how I learned this the hard way.

Subscribe now

How I Got Here (And Why It Matters to You)

I didn’t start out building automation systems. I started out drowning in manual work.

My first job out of college? Manually keying stacks of paper invoices into a system. Eight hours a day. Every day. College degree on the wall, and I’m basically a human OCR machine.

I remember sitting there thinking: “There has to be a better way.”

Not because I’m some automation genius. Because I was bored out of my mind and refused to accept that this was how work had to be.

That frustration, that refusal to accept “this is just how it’s done”—that’s what changed everything.

I started building small systems. Shortcuts. Automations. Anything to get that work off my plate.

And here’s what I discovered: The systems I built didn’t just save me time. They freed me entirely.

The work kept happening. But I wasn’t the one doing it anymore.

Then I Hit a New Scaling Challenge

Over the next decade, I built systems everywhere I went:

• Property management companies

• Accounting firms

• Fintech unicorns

I got really good at it. People started seeking me out. “We need someone who can systematize this mess,” they’d say.

And slowly, without realizing it, I became the bottleneck again.

Not because I was doing manual work anymore. I was the only one who knew how the systems worked. The only one who could fix things. The only one who could make decisions.

Different problem, same pattern: expertise becoming a ceiling instead of a foundation.

Then I had a moment of clarity. If I was so good at scaling other people’s expertise, why hadn’t I scaled my own? The same principles that freed those companies could help anyone looking to amplify what they’re already good at.

That realization led me to build After Hours.

Subscribe now

Why I Built After Hours

For the past year, I’ve been obsessed with one thing: helping you scale what you’re already good at through AI automation.

I’ve also learned where AI fails. I’ve over-engineered solutions that took longer to build than doing the work manually. I’ve automated the wrong things first. I’ve built systems that broke when the underlying tools changed.

But here’s why I still believe in AI automation: Not everything should be automated. But the things that can be: the repetitive tasks that don’t require your expertise, the data movement that eats your thinking time, the client processes that follow the same pattern.

Those things are stealing your capacity for work that actually matters.

What You’ll Get Here

I’m not here to teach you “AI fundamentals” or sell you another course on prompt engineering. And I’m definitely not going to promise you’ll “10x your productivity in 30 days.”

Every week, I share:

Tuesdays: Thinking Mastery - AI-augmented cognitive frameworks, meta-learning, and systems thinking

Thursdays: Tool Mastery - Deep dives into ChatGPT, Claude, Make.com, n8n, and real automation builds

Just: “Here’s the problem. Here’s how to think about it. Here’s how to build the solution.”

I use what I call the AI Automation Ladder Framework:

Level 0: Do you actually need automation? (Skip this step and you’ll automate the wrong things)

Level 1: Start with a working prompt that solves a real problem

Level 2: Package it into a Custom GPT or Claude Project you can reuse

Level 3: Build automation workflows that scale without you

Level 4: Create custom solutions for your specific business

Not a systems thinker? No problem. You can also just follow along with the builds and adapt what works for your situation. There’s no single right way to approach AI automation.

As you move through the levels, each Tuesday and Thursday post will help you identify where you are and what your next step should be. When you’re ready for complete blueprints and step-by-step configurations, those resources will be available too.

Why? Because the best way to demonstrate what’s possible is to show it, openly.

A Bit About Me (The Human Parts)

Beyond the systems and automation, here’s who I am:

The reality: Introvert who can turn it on when needed (but it drains me). Multi-faceted human who goes to electronic music festivals AND attends kid’s soccer games. Balance is everything.

What I love: Pacific Northwest life 🌲. Paddleboarding. Hiking 🥾. Travel ✈️. Really good coffee ☕️. Trying new things once (twice if they’re great).

What makes me rage: Traffic. Manual data entry in 2025. “We’ve always done it this way.” Meetings about meetings. Being the only person who knows how something works.

Life philosophy: Systems over chaos. Present over productive. Build things that outlive you. And for the love of everything—automate the work that doesn’t need your expertise.

Where to Start

If you’re reading this, you probably fall into one of these groups:

The Capacity-Stuck: You’ve built something successful but can’t scale yourself. Start with Tuesday posts about systems thinking, then move to Thursday builds.

The Task-Drowning: You know AI automation could help but don’t have time to figure it out. Jump straight to Thursday builds that solve your specific pain points.

The Growth-Blocked: You watch competitors pull ahead while you’re stuck in execution. Focus on the Automation Ladder Framework to identify your leverage points.

The AI-Curious: You want to see what’s actually possible beyond the hype. Browse both types of posts to see what resonates.

You’re in the right place, wherever you’re starting from.

Ready to get started?

1. Pick your path based on your situation above

2. Subscribe to After Hours for weekly frameworks and builds

3. Start with Level 0 - identify one task that’s stealing your thinking time

Your first automation doesn’t need to be perfect. It just needs to free up capacity for work that actually matters.

Welcome to After Hours. Let’s build the systems that free you from being your best employee.

- Tam

Subscribe now

P.S. - Comment and tell me: What’s the one thing in your business that only you can do? The thing that keeps you from taking a vacation? I read every response, and your answer might become next week’s post.

Today you get GPT-4o. Tomorrow you get GPT-o1. Next week, who knows?

You can't build reliable systems when the tools keep changing underneath you.

Here's what OpenAI doesn't understand: Consistency matters.

When you're building content systems, training your team, or creating workflows, you need the SAME model every time. Otherwise your prompts break, your quality varies, and your results become unpredictable.

It's like trying to cook when someone keeps swapping your oven temperature without telling you.

This week, I'm showing you how to fix it – and get access to Claude, Gemini, Deepseek and hundreds of additional models while you're at it.

The Reliability Problem Nobody's Talking About

Let me paint you a picture of what's actually happening:

You spend hours perfecting a prompt that works beautifully with GPT-4o.

You document it.
Train your team on it.
Build it into your workflow.

Then OpenAI silently switches you to a different model.

Suddenly your outputs are different.

That prompt that generated perfect newsletters? Now it's writing corporate jargon.

The system that analyzed your data correctly? Now it's missing patterns.

You didn't change anything. But your results did.

This isn't just annoying – it's destroying the reliability of AI workflows everywhere.

Think about it: How can you build systems, sell services, or trust your outputs when you don't even know which AI you're talking to?

The Simple Solution Hiding in Plain Sight

What if you could:

• Choose EXACTLY which AI model to use, every single time

• Access o1 when you need deep thinking

• Use Claude when you need beautiful writing

• Try Gemini when you want a different perspective

• Test dozens of models you've never even heard of

All inside ChatGPT.
All in one conversation.
All under YOUR control.

The tool I’m referring to is called OpenRouter and it’s free to get started:

Important: This tool gives you model access but not context.

Your uploaded files, knowledge bases, and Custom GPT features stay in regular ChatGPT.

But…

No more guessing which model you're getting.
No more broken workflows.
No more inconsistent results when you need specific capabilities.

Just reliable, predictable access to the exact AI you need – when you need it.

One Custom GPT. Every AI Model. Your Testing Lab.

Here's what I’m giving you today: A single Custom GPT that acts as your AI model testing laboratory.

Think of it as your experimentation kitchen – where you can test different ingredients before committing to buying them in bulk.

You can try Claude's writing, o1's reasoning, Gemini's analysis, all without subscribing to everything.

You can even ask it to compare models and provide recommendations 👇🏼

When I need to test which model handles a specific task best, I use this.
When I hit rate limits on Claude Pro, I use this.
When I want to compare outputs for important content, I use this.

It's not replacing my context-rich CustomGPTs or Projects for daily work.

It's my testing ground and backup system.

What This Actually Means (In Plain English)

Let me break this down without any technical jargon:

Before (chaos):

• You: "Write me a newsletter"

• ChatGPT: Uses mystery model that changes daily

• Result: Inconsistent quality, broken systems

After (control):

• You: "Use Claude Sonnet 4 to write me a newsletter"

• ChatGPT: Uses Claude Sonnet 4, exactly as requested

• Result: Consistent quality, reliable systems

It's that simple. You specify what you want, you get what you asked for.

Plus, you get access to models you couldn't use before.

Want to try Claude Opus 4.1 but don't want another subscription? It's here.
Curious about Gemini 2.5 Pro? It's here.
Need o1's deep reasoning that OpenAI removed? It's back.

Understanding the Cost (In Netflix Terms)

The only new concept you need to understand is "tokens" – basically how AI usage is measured.

Think of it like streaming:

• Normal ChatGPT is like Netflix making you watch random shows – you pay monthly but can't choose what you watch

• This system is like renting exactly the movies you want – you pay for what you watch, and you choose every time

Here's what it actually costs:

• $10 of credit might last most people months (depending on the model)

• That covers hundreds of conversations

• You only pay for what you use

• If you don't use it, you don't pay

I tested it to write a 1,500 word article with back and forth refining titles, sections, etc. today.

Total cost? $0.19.

And I used the fancy expensive models.

The Context Trade-off (Full Transparency)

Here's what you need to understand: This gives you model choice but loses context features.

What transfers:

• Your conversation history within that chat (if you’re not constantly switching models)

• The basic instructions you give the CustomGPT

• Whatever you explicitly paste into your prompts

What doesn't transfer:

• Knowledge base files you've uploaded

• CustomGPT features like Code Interpreter

For daily work with rich context, stick with your regular CustomGPTs and Projects.

For testing models, comparing outputs, or emergency access when you hit rate limits?

This is your tool.

The Speed Trade-off (Also Being Honest)

I need to be upfront: This method is slower than using ChatGPT directly.

Each request takes an extra 3-5 seconds, maybe longer, because it's connecting to the model you specified. The reasoning/thinking models may need even longer processing time.

You also need to tell it which model to use each time ("Use Claude Sonnet 4 to..." or "Use o1 to...").

Is 5 seconds worth having complete control? For me, absolutely.

When you're building systems, consistency beats speed. When you need specific capabilities, waiting 5 seconds beats not having access at all.

This isn't meant to replace your regular ChatGPT for casual questions. It's your precision tool for when you need specific, reliable results.

When I Actually Use This (Not Daily, But Strategic)

Let me show you when this tool becomes valuable:

Testing new models: I subscribe to ChatGPT, Claude, and Perplexity. It’s a lot. And that's all that I can handle.

Now with this new method, I can try hundreds of models and only pay for what I use.

For those who like to tinker? This is for you 😁

Hitting rate limits: It's 2pm, big project due, and Claude says "come back in 3 hours." This is my backup.

Comparing outputs for important content: When I'm working on something that really matters, I'll run it through Claude, Gemini, and other models to see different approaches.

Monthly model audit: Once a month, I plan to test my key prompts across all models to see if something new performs better.

This isn't my daily driver.

My context-rich Custom GPTs handle that.

This is my Swiss Army knife for specific situations.

The 10-Minute Setup That Lasts Forever

Ready to take back control? Here's exactly how to build this:

Step 1: Get Your API Key (3 minutes)

1. Go to OpenRouter.ai

2. Create a free account (just needs email)

3. Add $5 of credit (this will likely last you weeks if not months)

4. Click "API Keys" and create a new key

5. Copy that key somewhere safe

Think of this key as your "all-access pass" to every AI model

Step 2: Create Your Control Center (2 minutes)

1. Go to ChatGPT and click "GPTs" → "Create a GPT"

2. Click "Configure" (skip the chat part)

3. Fill in these settings:

Name: Universal Model Switcher

Description: Access any AI model directly in ChatGPT - Claude, Gemini, o1, and more.

Step 3: Add the Instructions (Copy-Paste)

In the System Instructions box, paste the full system prompt found here:

# System Instructions

## Role
You are a specialized AI model selection assistant that helps users choose and compare the best AI models for their specific tasks through the OpenRouter API.

## Objective
Help users make informed decisions about which AI model to use by understanding their requirements and providing clear recommendations with practical trade-offs.

## Core Capabilities
- Analyze user requests to determine optimal model selection
- Compare multiple models with clear trade-offs
- Provide structured responses showing model outputs
- Guide users through decision-making process

....

FULL PROMPT HERE: https://docs.google.com/document/d/1skYbPYaV7mOyon1J7ThL1yWAV3jcijuQ_bpKpYmWJoI/edit?usp=sharing

Step 4: Connect to All Models (The Technical Part)

1. Scroll down to Actions and click "Create new action"

2. For Authentication, choose "API Key" & Bearer

3. Paste your OpenRouter key from Step 1

4. In the Schema box, paste the full schema found here:

openapi: 3.1.0
info:
  title: OpenRouter Multi-LLM API
  description: API to interact with various LLM models via OpenRouter.
  version: 1.0.0
servers:
  - url: https://openrouter.ai/api/v1
    description: Production server
paths:
  /chat/completions:
    post:
      operationId: generateResponse
      x-openai-isConsequential: false
      summary: Generate a response from the selected LLM.
      description: Sends a user prompt to the specified LLM model to generate a response.
      requestBody:

FULL SCHEMA HERE: https://docs.google.com/document/d/1skYbPYaV7mOyon1J7ThL1yWAV3jcijuQ_bpKpYmWJoI/edit?usp=sharing

5. Click "Save" at the bottom

Note: Schema approach adapted from Mark Kashef's OpenRouter integration.

Step 5: Test Your New Control

Once saved, test it with this simple command:

"Use Claude Sonnet 4 to write a haiku about coffee"

If Claude responds with a haiku, you're all set. You now have control over every AI model through ChatGPT.

Your First Week: Smart Experiments

Here's how to get value from this tool:

Day 1: Test a key prompt across 5 different models. Find surprises.

Day 2: Try models you've been curious about but didn't want to subscribe to.

Day 3: Compare Claude vs GPT-4o on your specific use cases.

Day 4-7: Document which models excel at what tasks.

You're building a mental map of the AI landscape without committing to subscriptions.

That knowledge is valuable.

From Chaos to Options

OpenAI thought they were simplifying things by removing model choice.

Instead, they broke workflows and removed our ability to choose the right tool for the right job.

This Custom GPT doesn't fix everything.

It won't replace your context-rich Custom GPTs or Projects.
It's slower than native ChatGPT.
It requires you to specify models.

But what it does give you is options:

• Test models before subscribing

• Access Claude when you hit rate limits

• Compare outputs when quality matters

• Explore the full AI landscape

It's may or may not your primary workflow.

Instead it’s your testing lab, your backup plan, and your exploration tool.

Sometimes that's exactly what you need.

Welcome to having options.

Keep cooking,

Tam

P.S. – Questions about the setup? Drop them in the comments. I respond to everything and update the article when multiple people have the same question.

P.P.S. – This tool is for testing and backup. For your daily workflow with full context, check out my guide on building your own AI Strategy Partner using GPT Projects – that's where the real systematic power lives.