Who Are You in This Scene?

There's a moment in The Matrix everyone remembers. Most of us remember it wrong. And that mistake explains exactly why most organizations are using AI badly today.

The scene we all remember wrong

Neo sits in the chair, they plug him into the machine, and ten hours of combat training load directly into his brain. He opens his eyes and says the line:

"I know kung fu."

From there, in most people's memory, he becomes a fighting machine.

That isn't what happens. What happens next is that he steps into a sparring program against Morpheus — and loses. Repeatedly. Morpheus puts him on the floor while the entire crew watches from above.

And it's a strange scene, because Neo's knowledge is perfect. He didn't learn kung fu, he loaded it. There isn't a move he doesn't have, a technique he's missing, a gap in the material. And he still loses to someone who learned the same thing slowly, like everyone else.

Morpheus doesn't tell him he's missing knowledge. He tells him something else entirely:

"Stop trying to hit me and hit me."

And then:

"There is a difference between knowing the path and walking the path."

I went back to that scene recently, and not out of nostalgia. I went back to it because I watch it play out — almost line for line — in every organization I work with.

What Neo is actually missing

Neo isn't short on knowledge. He's short on an entirely different kind of knowing.

He knows what to do. What he doesn't have is knowing when. He doesn't have the sense that something is wrong half a second before it happens. He can't recognize that the move he's selecting is right in the book — and wrong in this fight, against this opponent, in this moment.

The philosopher Michael Polanyi called this tacit knowledge, and captured it in a single line: "We can know more than we can tell."

A physician who looks at a patient and says "something here doesn't add up" before she can point to what. A manager who reads a proposal and feels it won't close. A salesperson who hears a tiny shift in tone and knows the deal is gone. None of them can write it down as a procedure.

Which is exactly the point: if it could be written down, it could be loaded.

This kind of knowledge is built one way only — through a loop. Attempt, failure, correction, attempt again. And what that loop produces isn't more knowledge. It produces discrimination: the ability to tell good from not-quite-good, fast, without reasoning it through.

That's what we call judgment. And it's what separates someone who knows from someone who can.

Watts and the trap of the word "discipline"

At this point the obvious response is: "So it takes discipline. It takes putting in the work. There are no shortcuts."

Alan Watts argued that this response is precisely the problem.

"Now I know the word discipline isn't very popular these days, and I would like to have a new word for it — because most people who teach disciplines don't teach them very well. They teach it with a kind of violence... I would prefer to use the word skill."

The distinction looks semantic. It isn't.

Discipline is something you impose on yourself against your own will. It is defined in advance as unpleasant. And so — this is Watts's point — it breaks. People don't quit because they're weak. They quit because it was framed as a fight, and it's very hard to win a long fight against yourself.

A skill is something you acquire. Nobody "disciplines themselves" into learning an instrument. People come back to it because they're curious what will happen. And the repetition along the way stops being a price — it becomes the skill itself, in the act of forming.

Which changes the management question entirely. It isn't "how do I get my people to stick with it" but "how do I make the craft interesting enough that they want to come back to it."

Every organization that tried to force AI adoption by mandate — "everyone must use this" — did exactly what Watts describes: taught it with violence. And got the predictable result: performative compliance, shallow usage, and total abandonment the moment the pressure dropped.

What AI actually did

Here we need precision, because most AI discourse operates at a level of generality too high to be useful.

AI did not "replace expertise." It did something very specific:

Anyone in your organization can now produce, in two minutes, a marketing strategy, a competitive analysis, a requirements document, working code, a legal letter, an operating plan. The output will look professional. It will use the right vocabulary. It will be well structured.

What AI did not take to zero is the number of reps.

And here we have to stop, because we got the casting wrong

Up to this point I've assumed something, and you've probably assumed it with me: that we are Neo.

It's the obvious assumption. AI is plugged into the back of our heads, loading us with knowledge, and we walk into fights carrying knowledge we didn't earn.

But look at the scene again, and ask who in it actually fits the description.

• Who arrived in the world with all existing knowledge, without acquiring any of it?
• Who knows every move, in every style, from every school — and has never taken a hit?
• Who has read every account of failure ever written, and has never failed?

A language model is precisely Neo after the download. It was loaded, not trained. It has read everything and done nothing. It has perfect knowledge of consequences and zero experience of one.

And I'm not speaking in metaphor. This is a technical description of what a language model is: a system that learned from text. No body, no stakes, not one single instance where something hurt.

Once you see it, its behavior stops being mysterious. The total confidence in a wrong answer. The move that's right in the book and wrong in the fight. The reasoning that sounds perfect and falls apart on first contact with reality.

That isn't a bug. That's what happens when there's knowledge without hits.

So who are you?

There are four roles in the scene. Most of us are auditioning for the wrong one.

The role most of us try to take: Neo

This is what happens every time we compete with the model. When we try to know more, remember more, produce faster.

It's an unwinnable competition — not because we're bad, but because we picked the one axis where it genuinely beats us.

The role the ring actually needs: Morpheus

Notice the detail that's easy to miss: Morpheus doesn't know more than Neo. He knows less. He didn't get ten hours of training programs loaded into his head.

What he has is hits. He is the only person in the room who can look at a perfect move and say "that won't work on me" — and be right.

That is exactly your role opposite the model. You are there for the resistance — the knowledge already exists without you. You are the thing that turns it into capability, because you're the only one in the room who can say "no, that isn't good" and be correct.

Which has made discrimination — that tacit knowledge — the entry requirement. Without it you aren't Morpheus. You're a second, worse Neo.

The third role: the director — and also the trap

This is the part that's easiest to love, which is exactly why it needs care.

The director doesn't fight. He decides what the scene is supposed to achieve in the first place, who's in frame, what counts as a win, and when it's done. In an AI world that's precisely the role that has opened up: defining what the system is trying to achieve, what "good" means, and when good enough is enough.

It's a real role, and it's the opportunity of the decade.

But it carries a trap: the director is also the role that tempts you to skip the dojo. "I don't need to know how to do it, I orchestrate the process." A director who has never been in a fight can't tell a good take from an impressive one.

So the precise formulation isn't "we are the directors." It's:

You can only direct a scene you could have acted in.

Without the dojo, a director isn't a director. He's a spectator with opinions.

The fourth role: the spectators — that's the market

There's someone else in the scene who's easy to forget: the crew watching from above.

They can't see what was loaded into Neo's head. They don't know how many training hours went in. They see exactly one thing — who is on the floor.

That's your market. It isn't impressed by the download.

And the agent you built — it's the purest Neo of all

If you build AI agents, you are staging this scene literally.

You take a model, load it with knowledge — a knowledge base, prompts, playbooks, tone of voice, objection handling — and declare that it knows kung fu. Then you send it into its first fight against a real user.

And it loses. Always. The first time, it always loses.

Not for lack of knowledge — its knowledge is more complete than any employee you have. It loses because the real user isn't in the catalogue.

And what you do next — read the transcripts, find where it lost the conversation, correct, run it again — isn't "optimization."

You are running the agent through the exact loop that built your own judgment. The only difference is that this time you're standing on the other side of the mat.

(This is exactly what we do at e-Sharp. The overwhelming majority of the work on an agent isn't building it — it's the reps that come after.)

Which is why you cannot build a good agent in a domain you don't understand. Not because you'd lack knowledge — the model brings that. But because you wouldn't be able to tell that it's losing.

Three failures, now with the right names

1. Playing Neo instead of Morpheus

Shipping the output instead of applying judgment to it. The output is syntactically correct, professional in appearance, and wrong for the situation — and the person without discrimination doesn't know it's wrong, because it looks exactly like something that would be right.

This is a real break from the previous world: mediocre human work looked mediocre. Mediocre model output looks excellent.

2. Directing a scene you couldn't have acted in

Your ability to evaluate an AI's output is bounded precisely by your ability to recognize when it's wrong.

Outside your domain of expertise, AI doesn't give you expertise — it gives you confidence without expertise. Which is more dangerous than ignorance, because ignorance at least knows to ask.

3. The paradox: skipping the loop that builds the judgment needed to use AI well

This is the part that concerns me most.

AI is most valuable to people who have already done the reps — it's a speed multiplier for anyone who can judge. And it is most tempting to people who haven't, because it offers them exactly the skip.

The result is the opposite of what we're promised:

What to do about it

Separate tasks with verifiable output from tasks that require judgment

Code that passes tests, a translation you can cross-check, a summary of a document you can read — the output is checkable, and AI is immediate leverage at low risk.

Strategy, pricing, org design, positioning — there the output looks best exactly when it's wrong.

These two categories demand completely different policies. Most organizations treat them identically.

Protect the reps of your junior people

The grinding work a junior does in their first two years isn't waste to be optimized away. It is the manufacturing process for your Morpheuses.

If AI eats it, you get an immediate cost saving and a serious problem in five years: an entire cohort that can produce output and cannot judge it. A room full of Neos, with nobody who can put them on the floor.

The saving is immediate. The bill arrives in five years, and its name is continuity.

Measure reps, not adoption

"How many people are using AI" is a meaningless metric.

The real question is how many rounds they went through on a deliverable before shipping it, and how many times they rejected the first answer.

A team that accepts the first output is playing Neo. A team that argues with it is playing Morpheus.

And finally: make it a skill, not a discipline

Don't mandate adoption. Generate professional curiosity. People return to a craft when they're curious what will happen, not when they've been told to.

It sounds like the softer approach. In practice it's simply the one that works, and mandates don't.

Back to the dojo

There's one detail in the scene that's easy to miss: the download wasn't useless.

Neo genuinely needed it. Without it he'd have had nothing to fight with at all. It gave him the vocabulary — every possible move, instantly available. What it didn't give him was the ability to choose between them under pressure, against a real opponent, in a situation matching nothing he'd loaded.

The dojo gave him that.

But the question we started with — "how do we avoid becoming Neo" — was the wrong question from the start.

We aren't Neo. We never were. Infinite knowledge without experience isn't our condition — it's the condition of the tool we built.

The right question is whether we've earned the second role.

Because against a machine that knows everything and has experienced nothing, the only thing we have — the only thing that cannot be loaded — is that we've taken hits. That we tried, failed, corrected and tried again, enough times to know something is wrong before we can explain why.

That isn't what's left to us after AI took the rest.

It's the only thing that was ever worth anything in the first place.

AI isn't the enemy, and it isn't the dojo.
It's Neo — and you need to be good enough to train it.

I'm Shahar Peer. At e-Sharp I help B2B companies build AI systems that hold up against reality — not just in the demo. And at Pele I built what this article describes: instead of a large AI program that attempts everything at once, you pick one process, prove measurable value on it, and only then expand.

One rep at a time.

— Shahar Peer · e-Sharp