But for the last few months I stopped learning new things, because it stopped making sense to me. And AI had something to do with that, because why learn new things when AI can summon everything for you.

Spoiler alert: well boi, was I wrong… Yes sorry, I know this sounds like a duh moment for most of you.

So for the last couple of months I felt lost. For those who do not know me, I work for myself as a fractional product manager, helping startups navigate the hard process of building and setting up strong development teams. And for months, I kept adjusting. My framing. My positioning. My direction. Something was always missing.

At some point, my work and my passion started to feel like a constraint.

This was a hard thing to admit to myself. When you're working for yourself, your wellbeing directly reflects on how the business is doing. So not a good signal. So I had to figure out what was going on.

Was I tired of the AI noise? Maybe. Did the market stop making sense to me? Maybe. Was the field changing too fast? No. Did I miss my passion for learning? YES!

Diving deep into topics again. Going hands on with something I just learned. Doing experiments and making mistakes.

I found out I had simply stopped learning, and with it, lost the thing I loved most about my work. What made it worse: in the age of AI there is a new way of learning, and I had not even learned about it yet.

So here we are: my first step back into continuous learning, on how to learn in the age of AI.

Why bother teaching yourself anything, now AI can just do it

If you already know you should keep learning and never doubted it, congrats, go ahead and skip this section. But if you have been quietly telling yourself that AI makes independent learning less necessary, this is for you.

So why bother, you say? Well, let's not overcomplicate it: AI is quite dumb, and it all depends on context. It does nothing less than predicting the next word. It is only a powerful tool if you know how to use it. A very old saying in IT is more relevant than ever: garbage in is garbage out. So it comes down to two things: what you put in, and what you do with what comes out.

Expertise: what you put in. Knowing your domain is what lets you feed AI the right context in the first place, the details that actually matter, the constraints it needs to know about, the framing that gets you a useful answer instead of a generic one. Without that, you are not directing AI, you are just hoping it guesses right.

Judgement: what comes out. It works on the way out instead of the way in. AI is built to sound like it's right, whether it is or not, because it is optimized to sound fluent, not to be correct. And deciding is what matters. That is exactly why a lot of people say judgement is the best thing to have right now.

Subscribe now

Expertise and judgement do that job through four functions:

• Catching the error a novice would miss. A wrong answer that sounds confident is invisible to someone who does not know enough to question it. Expertise is what makes the wrongness visible in the first place.

• Asking the question that points AI somewhere useful. A vague prompt gets a generic answer. Knowing your domain is what lets you ask the specific, constrained question that actually gets AI working on your real problem instead of the average version of it.

• Judging whether the output actually fits your workflow. AI does not know your codebase, your client, or your market. It only knows whether an answer sounds plausible in isolation. You are the one who has to judge whether it fits the actual context it needs to live in.

• Telling good enough apart from genuinely good. This is the hardest one to fake. In anything creative or strategic, the gap between adequate and excellent is exactly where expertise shows up, and AI has no way to close that gap for you.

AI predicts the next word based on patterns in what it was trained on, it has a knowledge cutoff, and it does not browse the live web unless you turn that on. That is not a flaw you can patch by asking nicely, it is just the limitations of where we are.

Which is exactly why you sometimes have to distrust what comes out. Not because AI is lying to you, it cannot do that, but because it genuinely does not know what it does not know. It will answer a question past its cutoff with the same fluent confidence as one it actually knows.

Expertise and judgement help you navigate through this.

Continuous improvement and learning is what gives you resilience. It is also the one thing that compounds. Every skill you pick up now makes the next AI interaction sharper, and that stacking effect is what actually keeps you ahead.

And if you are not learning on a continuous basis, I highly recommend you start from today onward. Here is a guide to help you.

Old habits, new tools

Expertise and judgement are not grown overnight. You build both the same way people have always built them: by learning. Become an expert in a topic by learning about it. By just doing it.

The only thing that changed is the environment you can draw your knowledge from. You do not have to buy a book anymore, or hunt down the right source. The answer is always one prompt away.

The question is how you actually do it, in a world where AI can do most of the work for you if you let it.

You do it the way you always have: by doing it, through repetition, through getting things wrong until you stop getting them wrong. The only difference is that AI can help you understand it faster, cover more ground, and navigate it as you go.

Here are a few ways AI can help you learn:

• Learn with AI, not around it. Learn AI fluency, a framework for how you work with AI. I will dive deeper into this later on.

• Create a system that helps you learn. You do not become an expert by reading more, or by letting AI generate more. You become one by writing down what you decide and notice, so it stacks instead of evaporating. Let AI log this for you.

• Explain it to AI. The moment you put something into words for someone else, you find out exactly where your own understanding is thin. This article is a version of that.

And here will be a payoff.

What all that expertise actually gives you is context, and context is the one thing AI does not have and never will. It knows a lot in general, it knows nothing about your client, your codebase, your market, your specific mess, until you tell it.

Combined with your expertise, you can feed it quality, no more garbage in. Quality in is quality out. You also become better at judgement: the more you know, the better you can tell when AI actually nailed it versus when it just sounded like it did.

Skip the learning and you lose both ends at once: worse context going in, no way to catch a wrong answer coming out.

What AI Fluency looks like, so that it helps you get the best out of AI

This is not about learning AI as a technology, it is about learning to actually collaborate with it. That is what AI Fluency means: working with AI in ways that are effective, efficient, ethical, and safe. It helps you build context and judgement, both of which matter more than the tool you happen to be using.

There are three ways you actually work with AI. Augmentation, where you and AI go back and forth, building something together. Automation, where you hand over a fully scoped task and AI just executes it. And agency, where you set AI up to run on its own for a while, without you watching every step.

AI Fluency comes down to the 4D framework: Delegation, Description, Discernment, and Diligence. All four apply across all three modes, whether you are handing off a task, building something together, or letting AI run on its own.

Delegation

Delegation is deciding what you do yourself, what you do together with AI, and what you hand over completely, before you even open a chat.

• Problem Awareness. What am I actually trying to achieve, and what does the work involve? Figure that out before AI gets anywhere near it.

• Platform Awareness. What can this specific AI system actually do well, and where does it fall short? Trust it because you checked, not because it sounded confident.

• Task Delegation. Given both, what goes to you, what goes to AI, and what do you build together. Good delegation needs your own expertise and an honest read of the tool. Skip either one and you either hand over work AI cannot actually do, or you do work yourself that AI could have done faster.

Description

Description is how well you communicate with AI in a way that actually creates a productive collaboration, not just a query and a response.

• Product Description. Defining what you want in terms of output, format, audience, and style. AI cannot guess this, “guess what I’m thinking” gets you a generic answer every time.

• Process Description. Defining how AI should approach the request, step by step instructions instead of a single vague ask. “Act as a Socratic tutor” is process description, not just a fun instruction, it changes how AI gets to the answer.

• Performance Description. Defining how AI should behave while working with you, concise or detailed, challenging or supportive. This is the one people forget, and it changes the collaboration as much as the other two.

Discernment

Discernment is evaluating what AI produces, how it got there, and how it behaved while doing it, not just accepting the answer because it sounds right.

• Product Discernment. Evaluating the quality of the actual output: is it accurate, appropriate, coherent, relevant. This is the one that depends entirely on your own expertise, outside your domain you can only tell if it sounds fluent, not if it is true.

• Process Discernment. Evaluating how AI got there, checking for logical errors, dropped context, or reasoning that quietly went sideways.

• Performance Discernment. Evaluating how AI behaved while working with you, was the tone right, was the communication style actually useful for what you needed.

Discernment works hand in hand with Description, in a loop. You describe, it responds, you evaluate, you describe again. That loop, repeated enough times, is basically what learning through AI looks like in practice.

Diligence

Diligence is taking responsibility for what you do with AI, and how you do it, this is the one that is easy to skip when you are moving fast.

• Creation Diligence. Being thoughtful about which AI system you actually use, and how you interact with it, not just reaching for whatever is open.

• Transparency Diligence. Being honest that AI was part of the process, with whoever needs to know that. “In this document, Claude drafted the first version” costs you nothing and buys you trust.

• Deployment Diligence. Taking responsibility for what you actually verified before you used or shared it. What counts as enough verification changes by context, personal, academic, professional all expect something different, but “I did not check” is never an acceptable answer in any of them.

Now put it into practice

To maintain continuous development and learning habits, I always recommend building a system for yourself that helps you structure and support your needs. So the last piece is a system: a place where what you learn, decide, and figure out actually stacks instead of disappearing.

I cannot do that for you, because it is very personal. But I can show you how my own knowledge system works, so you can get inspiration from it and copy from it to build yours.

I documented my AI workflow, including my knowledgebase system, in my grimoire, you can find it here.

It describes the principles behind it, the five layers it is built from (knowledgebase, practice and experience, continuous improvement, sources, track record), the skills and tools that run it, and my routines that keeps it alive.

I’ve always advised starting with small teams, not only because it’s cheaper, but because it forces you to get the way you work right before you start scaling it. For years, that was simply the way I believed it should be done.

That advice hasn’t changed. What's changed is what a small team can actually do. When built right, small teams don't just keep up with larger ones, they actually win.

AI shifted that, and almost nobody, including me, has fully caught up with it yet. Nobody has laid out what it actually takes to build small teams that make full use of that shift.

No one connected the dots

I believe the best products still come from the same place they always have: teams with real autonomy, a short distance to the user, and the ability to learn fast. That part hasn’t changed.

What has changed is what happens when you apply AI to that foundation, at the organization level, the team level, and the individual level. Get that right, and small doesn’t just hold its own. It wins.

I keep running into a problem, though. I see a bunch of self-proclaimed AI consultants who claim to know how it's done, but don't know anything about the foundation of building strong, empowered teams. So all their advice becomes noise, tools without a team that can actually use them.

I want to change that.

I've seen the patterns, and nobody has laid out how to do both at once: structure the team, and apply AI, in a way that actually builds something small and high-performing.

So that’s what I’m doing.

I'm building one piece of my own AI workflow every single day, at the organization level, the team level, and the individual level, and publishing what actually works and what doesn't. And I'm doing it alongside the founders and teams I work with, testing what holds up in a real team, not just in my own.

No polished framework upfront. Just the work, documented in real time, in a stack that's moving just as fast, so you can see what holds up and what falls apart as I work toward the framework that actually gets this right.

And you can follow along, we're figuring this out together. So feel free to share your insights at any time, I'm happy to discuss.

Come along if

Here's how you'll know you're in the right place:

• If you’re about to hire your next developer because things feel slow, and you’re not fully sure that’s the actual fix, this is for you.

• If you’re building a SaaS or AI product with a small team and want to know what AI actually changes about how you should work, not the hype version, this is for you.

• If you'd rather watch someone figure this out in real time than wait for a polished framework two years from now, this is for you.

Small teams win, and now is the time to act. Let's build one.

Subscribe now

AI plays an ever larger role in our lives. And whether you are a fan or not, there is no escaping it anymore: it’s here to stay. But here is the thing: most people do not use it well. You see everyone shouting about what they do with AI and how it speeds everything up, and whether that is big talk or not, for me it always comes down to the same problem: nobody tells you how to use it properly.

Especially not on a personal level, applied to your own work. Because how do you get better at AI when the development move so fast that just keeping up is a full time job in itself, let alone building a healthy, workable workflow for yourself?

Like everyone else I use AI in my work, a lot and often. But do I always find it supportive? No. Do I use it well everywhere? Also no. Is it sometimes AI slop? Yes.

That is what I want to put a stop to. I no longer want to use AI casually, but professionally and workably, embedded in my daily workflow so that it actually supports and speeds me up. And to do that properly, I am going to do exactly what the rest does not: show how I apply it. Through a challenge.

The challenge

That is why I am challenging myself. 90 days, from 29 June to 27 September. Every day I post a note on Substack about what I did that day to build and professionalize my AI workflow: what I researched, built or learned.

I track my progress in a dashboard I built myself, based on my GitHub contributions and what I researched or learned:

One square a day, 90 days long. Every colored square is a day I recorded something; the darker it is, the more I did. Alongside the squares I track the numbers that matter: contributions, learnings, current streak, daily average and overall progress.

The rhythm is the same every day: I research, I build my workflow, and I document how I do it. That last part is the core, because it lays a foundation I can easily apply within my own work.

From ordinary AI user to someone who applies it professionally and methodically. A deadline forces me to take it seriously, instead of doing it casually on the side. Every day.

Why I am doing this

As a solo founder I switch between product, development, design and marketing all day, which means AI touches everything I do. Building one workflow that holds up across all of it is exactly the problem worth solving in the open, and exactly what I want to show instead of explain.

Because unlike all those people shouting about how AI improves their workflow, I do not believe in shouting, but in showing. Telling people how good you are at something is easy; actually doing it is something else entirely. That is why I am not building my AI workflow quietly for myself, but recording all of it: what works, what does not, and how I approach it.

This challenge keeps me sharp. By having to post every day, I make sure that by the end I have actually done what I set out to do, and I always need that kind of pressure to discipline myself. And if someone else wrestling with the same question, how do I really get better at AI, gets something out of it, that is a welcome bonus.

The grimoire: the manual for AI magic and sorcery

All those daily pieces need to come together somewhere, otherwise they evaporate. That is why I am setting up a living website: grimoire.esmeepeters.com. There I collect everything I learn about my AI workflow during this challenge, in one place.

The grimoire is my manual for how I apply AI in my workflow. The name is a metaphor: a grimoire is originally a book full of magic and spells. And with the right AI workflow it can sometimes feel like you are moving your work forward by magic: you type a few words and something happens that would have taken hours before. But there is no magic involved. These are just actions you can learn, repeat and pass on.

Over the 90 days I keep adding to it. It is not a finished product I reveal at the end, but a document that grows with me: the further I get, the more complete it becomes. It does not stop after the challenge either, because a workflow is never done.

Everything that ends up in it is yours to use and copy. You can take the whole grimoire at once, or pull out only the principles you can apply to your own way of working. No paywall, no conditions. What works for me might just work for you.

It will not be perfect, but that is better than making and sharing nothing at all. Hopefully we learn from each other, and I get feedback on what could be better or more efficient. Because if nobody shares how they do it, we never learn.

Finally

Over the next 90 days I work on the grimoire, with a daily update through notes and a more detailed checkpoint every 30 days: what was built, what worked and what did not. That follows in a part II, III and IV.

Want to follow along? Then follow me through this challenge. And feel free to check in on grimoire.esmeepeters.com while it grows. I hope others get something out of it, and in return I would love to hear what I can do better.

With over 10 years of experience in the IT sector and a background in digital product design, development and product management, I’ve helped dozens of companies deliver faster and prioritize better.

Currently I support various growing startups in developing profitable products and optimizing their teams as a fractional product manager. I’m also building my own tech startup.

Previously I worked at Ouwehands Zoo and Planon, among others, where I gained experience with diverse environments, technologies and methodologies. This broad foundation enables me to make quick impact in new contexts.

Website: https://esmeepeters.com/
LinkedIn: https://linkedin.com/in/esmeepeters/
Contact & questions: mail@esmeepeters.com