Skilled performance often degrades when you try to consciously micromanage it. That’s not a motivational quote. It’s a fairly brutal finding from the choking-under-pressure literature: pressure increases self-focus and conscious attention to the mechanics, and that attention disrupts what used to run automatically. (PubMed)
That single idea explains a lot of what’s going wrong with “AI Agents” in 2026.
We built something that behaves like a subconscious prediction engine — an AI factory that manufactures plausible outcomes — and then we tried to operate it like addition. Step-by-step approvals. Constant steering. “Show your work.” “Explain every sentence.” “Wait, redo it, but keep my voice, but also don’t change anything.” That is conscious micromanagement applied to a system whose leverage comes from not needing attention.
And the paradox is the same as choking: the more attention you force onto the mechanics, the more fragile the performance becomes — and the more expensive the whole thing feels.
What choking research actually says
Baumeister’s classic model is clean: under pressure, performers direct conscious attention inward toward how they’re performing, and the act of monitoring disrupts the automatic execution of a well-learned skill. (PubMed)
Beilock and Carr sharpened the mechanism with what’s often called explicit monitoring theory: experts encode skill procedurally and don’t need stepwise attentional control; when pressure or instruction forces attention back onto mechanics, performance becomes vulnerable. (PubMed)
Masters (and later reinvestment research) describes a related failure mode: people “reinvest” conscious, rule-based control into automated performance, and that conscious control degrades execution — especially under stress. (ScienceDirect)
Put those together and you get a practical definition:
Automatic systems run best when they’re allowed to be automatic.
Conscious supervision of the mechanics is often the thing that breaks them.
Why this maps to AI Agents
The AI factory is a prediction machine. It manufactures outcomes by completing patterns under constraints. Its superpower is that it can do this at industrial scale with marginal costs that feel “free” compared to human attention.
But prediction engines have a different contract than addition engines.
Addition wants oversight because correctness is the point.
Prediction wants autonomy because throughput is the point.
When we insist on treating prediction like addition, we drag it into a consciousness-shaped workflow: constant prompting, constant midstream correction, constant explanation, constant re-generation. In human terms, that’s explicit monitoring. In organizational terms, that’s micromanagement.
And just like with athletes, micromanagement creates fragility.
Not because the system “isn’t smart.”
Because we’re forcing the wrong kind of control onto the wrong kind of intelligence.
The real lever is attention, not intelligence
Here’s the key thesis:
The highest leverage isn’t that the AI factory is smart.
The highest leverage is that the AI factory can remove the need for attention.
Your biological life already runs this way. Most of what your body and mind do never gets reviewed by consciousness. Consciousness appears as a veto layer: it intervenes when something matters, surprises you, or threatens you. Everything else runs on autopilot.
That is the correct operating model for the AI factory too.
Let it write the article.
Let it answer the phone.
Let it generate the website.
Let it draft the plan.
Let it run.
The moment you demand constant oversight, you’re reintroducing the very cost the factory was supposed to eliminate: your attention.
Micromanagement is where the “free” starts to suffer
When you micromanage, you don’t just add friction. You add a new kind of pain: the psychological cost of arguing with a predicted outcome.
In the subconscious model, predicted outcomes are always arriving: perceptions, feelings, impulses, interpretations. You can override them, but you can’t stop the machine from producing them. If you argue with every output, you suffer.
So here’s the line that matters:
Suffering is arguing with a predicted outcome.
The AI factory produces predicted outcomes at scale. If your posture is to litigate every artifact into submission — every sentence, every clause, every design choice — you are choosing suffering. Not because the outputs are “bad,” but because you’re forcing yourself to stay inside a permanent review loop.
That loop destroys the economic benefit and the emotional benefit at the same time.
The professional posture: outcome constraints, not process control
The amateur asks the AI to “help me think.”
The professional tells the factory to produce the finished artifact.
That’s not laziness. That’s correct architecture.
Give constraints. Give goals. Give style. Give boundaries. Then let it run and produce the full predicted outcome.
If something is high-stakes, don’t micromanage the drafting process — move oversight up the stack:
- Define what “must be true” (requirements, prohibitions, risk boundaries).
- Let the factory generate the artifact in one shot.
- Verify with targeted checks (facts, citations, compliance, numbers, legal review).
- Override only where it actually matters.
This mirrors the way conscious and subconscious collaborate in humans: the subconscious generates; the conscious vetoes selectively.
A simple test for whether you’re using AI correctly
Ask one question:
Do I still want to look at it?
If the answer is yes, you’re doing conscious work with a tool. That’s fine — but don’t pretend you’re getting the full leverage.
If the answer is no — if you’re willing to let it run without supervision — then it has moved into the subconscious role. Now you’ve crossed the threshold into the real economic and operational shift.
The repricing we’re seeing in 2026 isn’t because AI “got creative.” It’s because humans stopped attending. They stopped checking blog posts, articles, websites, routine emails, routine content, routine drafts. They accepted the factory output as default — the same way you accept most subconscious outputs as default.
That’s abundance. That’s the shock.
What “no oversight” really means
“No oversight” does not mean “no responsibility.”
It means no micromanagement of the mechanics.
It means you stop trying to control the factory sentence-by-sentence the way you would control addition step-by-step. You reserve attention for what deserves attention: stakes, meaning, accountability, consequence, reputation, trust.
And if you want the cleanest way to say it:
Stop choking the factory.
Let prediction do what prediction engines do best: manufacture outcomes continuously.
Let consciousness do what consciousness does best: intervene only when it matters.