You Don't Have a Knowledge Problem. You Have a Knowing–Doing Gap.

TL;DR — AI didn't just make doing easier; it made knowing dramatically cheaper too. When both get cheap at once, access stops being the bottleneck. The gap between what you know and what you turn into something is. The way through: run a loop — pick, ship, share, learn — and let each turn feed the next. I call it knowing–doing unity (knowing–doing unity). The long version is below.

1. The version of the problem you've been sold

You open your browser with three tabs: a course on AI workflows, a thread of "26 prompts that changed my life," and a half-finished draft you meant to write three weeks ago.

You watch the course on 2x. You favorite the thread. The draft stays half-finished.

By evening you feel productive — you learned so much. But if someone asked what did you make? the honest answer is nothing. It nags at you, quietly.

The common diagnosis is a knowledge problem: I don't know enough yet. If I just learn a little more, I'll finally be ready to do the thing.

I think you have a knowing–doing gap — and adding more knowing to the knowing side won't close it.

2. AI made "knowing" cheap — and that created a new trap

Everyone talks about how AI lowered the cost of producing: drafting, coding, designing, shipping. It did, and it's the whole reason "do more, faster" became the default advice.

AI did something quieter and more consequential too: it made knowing cheap.

• Understand a topic → get a synthesized answer in seconds.
• See how experts think → summarize a decade of books tonight.
• Learn a skill's map before practicing → here are the steps, the pitfalls, the templates.

Knowledge has never been this fast, this accessible, or this cheap.

That created a sneaky trap: accumulation disguised as growth. Learning now feels like progress — the little dopamine hit of a new idea, a new tool, a new mental model — without the discomfort of actually doing something uncertain with it.

Underneath, this is an identity problem dressed as a productivity one. We collect because collecting is safe. Doing is where you might find out you were wrong.

So: with AI, you can know your way into permanent inaction and feel busy the whole time. More knowledge does not shrink the knowing–doing gap. For some people it widens it, because watching gets easier to mistake for building.

3. AI also made "doing" cheap — enter the second trap

The other side: because doing got cheap too, the mirror-image failure exists — blind action.

Everyone can generate now. A landing page. A post. A report. A "brand." A flood of output is one prompt away.

Cheap doing without direction produces noise — yours added to everyone else's. When output is free, it stops being the differentiator. Judgment is: knowing what to make, for whom, and why it matters, before you generate anything.

That's why I keep returning to Position Before Productivity: speed (and now raw generation) only compounds whatever you're already pointed at. If you're aimed wrong, AI just gets you to the wrong place faster, and with volume.

Two mirrors face each other:

• Pure accumulation — know everything, do nothing. Feels smart. Changes nothing.
• Pure generation — do everything, aimed at nothing. Feels busy. Adds noise.

Both dodge the hard part: drawing the two into one act. knowing–doing unity is the name for closing that distance.

4. What I mean by "knowing–doing unity"

knowing–doing unity — the unity of knowledge and action — comes from Wang Yangming, a 15th-century Chinese philosopher. Its core claim: knowledge that doesn't move you to act isn't really knowledge yet. If you "know" something but never act on it, you hold a description of it, not the thing. Genuine knowing is inseparable from doing — not two steps in a sequence, but one continuous act, each completing the other.

I'm not a philosopher, and this isn't a history essay. I invoke it because it's the sharpest diagnosis I know for what AI is doing to us.

When knowledge gets cheap, the temptation is to treat knowing as the whole game. When action gets cheap, the temptation is to treat doing as the whole game. Both are half-truths. The value sits in the loop: let knowing emerge from doing, and let doing be reshaped by what you're learning. Don't let either drift far ahead of the other.

That loop is the thesis of this site and of Makerloop — the weekly note you're probably reading this inside. Learn by doing. Learn by doing, not and then do.

5. The trap most people actually live in

Most of us are neither pure accumulators nor pure generators. We're paralyzed in the middle. We know enough to know we could do something good — and that's exactly what stops us, because the moment you can imagine the good version, you can also imagine falling short of it.

So we stay comfortable: consume a little more, plan a little more, "research" a little more. We tell ourselves we're almost ready. Because AI made learning frictionless, "almost ready" can stretch into months of perfect comfort and zero discomfort.

Being direct, the way I'd want it: you will never feel ready. Ready only appears after you start. Confidence is what the loop produces on the far side of doing — you start with one small, slightly-too-hard act, not with courage you're waiting to feel.

Ask "what is the smallest thing I can make today that will teach me something I can't learn by reading?" — and go make it.

6. The loop you can steal: pick → ship → share → learn

Enough theory. Here's the engine I actually run — behind every project on this site, including the site itself. A weekly cycle with four beats. You can compress it to a day or stretch it to a month; the shape is what matters.

Beat 1 — Pick. Choose one thing you genuinely want to learn, small enough to make progress on in a week. Not "master machine learning" — "ship a tiny tool that reads my inbox and drafts a reply outline." Pick by asking: what would I be proud to point at a week from now?

Beat 2 — Ship. Make it real. Not a completed course, not a note — a concrete artifact: a working script, a landing page, a post, a one-page system, a dataset, a recording. It doesn't have to be big, just finished enough to show someone. The artifact converts your knowing into something the world can react to.

Beat 3 — Share. Put it somewhere semi-public, even if it's ugly and even if almost no one sees it. This is the beat most people skip, and it's where the loop becomes generous instead of private: sharing forces you to finish (not polish, finish), it lets one imperfect thing help someone else, and it opens the door to feedback from people who actually looked. A thing you keep to yourself can teach you; a thing you share can teach you and get corrected by the world.

Beat 4 — Learn. After you ship and share, do the real learning: notice where you got stuck, what you'd do differently, what the reaction taught you. This is the knowledge no course can give you — it's earned, produced by the gap between your plan and reality. Feed it into next week's Pick.

Then repeat.

The value is in the repetition, not in any single beat. Each cycle makes the next one smarter: shipping shows you what's actually tractable, sharing shows you what resonates, and both feed a sharper Pick. This is knowing–doing unity at the scale of a week — a rhythm you practice until knowing and doing stop being two separate things.

7. The non-native angle (because it's real, and it's mine)

English isn't my first language. For a long time — long before AI — that felt like a wall between what I knew and what I could do in public. I could think clearly in my own head, but expressing, showing work, and building in English carried a daily tax. For non-native speakers the knowing–doing gap has an extra layer: the fear is less "will I be wrong" than "will I be wrong in awkward English."

AI collapsed most of that for me. Drafting, editing, a second pass on tone — like having a patient native-speaker editor on call. The gap narrowed to what it always should have been: whether I had something real to show, not whether my grammar would survive contact with a reader.

So: don't let "I'm not ready / not fluent / not expert enough" keep you on the knowing side. The language barrier is largely a solved problem in the AI era. What's left is universal — are you willing to ship something imperfect and let the loop, share included, teach you?

If a non-native speaker can run this loop and build a public body of work from it — and I'm the evidence — the thing that was blocking you was never the language.

8. The judgment layer: when to do, and when to stop

"Just ship" can itself become a trap if taken literally.

Cheap doing means you can generate a lot, fast — and not everything deserves to ship. Part of knowing, the part most resistant to automation, is judgment about what matters: which of a hundred possible things is worth your loop this week. That's the skill in The Rarest Skill: Judging AI Answers and Knowing What Matters. What AI can't outsource is taste applied to your own life: what's worth building, what's worth abandoning, when an idea is finished rather than abandoned.

knowing–doing unity doesn't mean "do everything." It means: whatever you decide to know, let it become doing; whatever you do, let it feed what you know. Judgment chooses which loops to run. The loop makes the ones you run compound.

Run too few loops and you accumulate. Run loops on the wrong things and you generate noise. Run the right loops repeatedly and you build something with no shortcut: a body of work only you could have made, because only you climbed that particular learning curve by doing it.

9. The smallest loop you can start today

A post about the knowing–doing gap isn't worth much if it just adds to your knowing. So here's the ask, and it's small:

This week, close one loop — pick, ship, share, learn. Find the smallest version of that thing you've been "preparing for." Finish it (not master it). Share it somewhere semi-public. Then, in one honest paragraph, write what the doing and the sharing taught you that the reading never could.

That paragraph is worth more than the next ten hours of consumption. It's the first brick of your knowing–doing unity — not as an idea, but as a habit.

Want that loop handed to you weekly — one concrete AI workflow, one career lesson, one thing shipped, with zero noise? That's Makerloop. Learn by doing, weekly.

FAQ

Why do I keep learning AI skills but never actually using them? Learning is cheap and comfortable; doing requires facing uncertainty. Collecting knowledge feels like progress without the risk of being wrong. Close the gap by shipping a small, imperfect artifact — and sharing it — then letting that teach you.

How do I actually apply what I learn with AI? Run a pick → ship → share → learn loop: choose one small thing, turn it into a finished artifact within a week, share it somewhere semi-public, then write down what the doing and the feedback taught you and feed it into next week's pick.

What does knowing–doing unity mean in the AI era? Knowledge and action aren't two separate steps but one continuous act, each completing the other. With AI making both cheap, the value is in continuously closing the loop between them — let building sharpen what you know, and let learning (including from sharing) feed what you build.

How can a non-native English speaker turn learning into building? AI drafting and editing tools have largely solved the language barrier. The real barrier is universal: being willing to ship something imperfect. If a non-native speaker can run this loop and build a public body of work, the constraint was never language.

— Mason. Learn by doing, weekly — get it in Makerloop.

Did this article help you? If you're working through career direction, or want to use AI to work smarter, let's talk — I'm always glad to trade notes with people thinking about the same things.