AI Isn't Coming for Your Job. It's Coming for Your Position.

There is a clip of Jimi Hendrix at Monterey where he plays the guitar behind his back, with his teeth, sets it on fire, and none of it matters as much as the part right before - where he just plays. Something comes out of him that he did not learn in a classroom. It arrived from somewhere inside the accumulation of everything he had ever done with the instrument.

Steve Vai practiced scales eight hours a day. Mapped every mode on paper. Built technique like an engineer builds a bridge. One of the greatest players alive. But nobody confuses what Vai does with what Hendrix did. Vai's mastery was built. Hendrix's was found - after years of dive bars and backup bands that nobody remembers.

Every field has both kinds of work. Work where the breakthrough is a leap and the scaffolding is mechanical. And work where the breakthrough is the scaffolding. AI makes the first kind trivially easy. The second kind it cannot touch - because it requires the person to have been changed by the process of doing it.

The invisible compression

The decade of unglamorous work a junior does before becoming a senior was not inefficiency. It was training. The junior architect drawing walls she could have auto-generated was building a pattern library through thousands of micro-failures. That library is what we call judgment. It does not come from being told things. It comes from doing things wrong long enough that your nervous system rewires.

AI compresses that. A junior with the current tools produces senior-looking output from month one. The client can't tell. But the person behind the output has not built the substrate the output is supposed to represent. Fluency without machinery.

This is not automation. Automation removes a task. This removes the developmental function of a task while leaving the task in place. The junior is still working. She is just not being trained by the work anymore.

The market cannot see this. It optimizes on output. Output looks fine. It will look fine right up until the conditions exceed the design parameters, and by then it is too late to train the people who understand the fundamentals.

Two compressions

It goes deeper. Even when you try to preserve what the apprenticeship produced - capture institutional knowledge, build systems that retain expertise - the record of a decision is not the decision. An email says "go with vendor B." It does not capture the thirty minutes of staring at the ceiling weighing pricing against a gut feeling from a handshake three years ago.

A surgeon's case notes record the procedure. They do not record the moment she looked at the tissue and felt something was wrong before she could say what. That feeling is fifteen thousand cases compressed into a reflex. It will never be in a system.

We are compressing in two directions. AI compresses the time to produce expert-looking work. And the best record we can keep of actual expertise - the documents, the decisions, the communication trail - is itself a lossy compression of the reasoning, which is itself a lossy compression of the experience. Three generations removed from the signal.

That is not a reason to quit. It is a reason to be honest about what layer you are working on. A system that captures every decision and communication pattern across twenty years knows something about how a practitioner thinks. It is not the apprenticeship. But it is categorically different from zero.

The barbell

Commoditize the middle tier of a profession and two things happen at once.

The floor rises. The 80th-percentile practitioner with AI produces what used to require the 95th. More people get better work. This is real and dismissing it is dishonest.

But the pipeline that produces the top tier thins. The middle was a conveyor belt. People entered mediocre and exited expert through years of doing work that was genuinely hard for them. Compress that band and the economic incentive to stay in the uncomfortable developmental phase evaporates. You can skip it and still get paid.

The average goes up. The ceiling comes down. For most things, the average is what matters. But the frontier - unsolved problems, paradigm shifts - lives at the ceiling. And it requires a tolerance for extended confusion that atrophies fastest when fluent answers are always thirty seconds away.

The point

This is not a eulogy. When everyone has access to competent output, competence stops being the scarce resource. Judgment becomes the scarce resource - knowing which competent output is wrong, which edge case the tool missed, which confident answer is confidently broken.

The tools solve the easy problems. What remains is the hard stuff - where depth cannot be compressed, where the human needs to be a real human who has done the work. The organizations that see this build systems that capture and compound institutional knowledge structurally. Not to replace lived experience. To make sure the next person's starting point is not zero, and to be honest about what is absent rather than letting a model fill the gap with fluency.

The apprenticeship is changing form. The people building for what replaces it are building the most important systems of the next decade. Not because they are nostalgic. Because they understand what was actually happening in the process everyone else is racing to skip.