williamliu.ai
EN中文
← Blog

AI May Reshape Who Gets to Build Judgment

Automating junior knowledge work may remove the apprenticeship tasks that develop expertise, forcing organizations to rethink how future experts gain judgment.

4 min read
  • Future of work
  • Learning

The real job question is deeper than “replacement”

Most AI job debates are still too crude.

The lazy framing is:

AI will replace humans.

I think the sharper framing is this:

AI may first replace the work people used to do in order to become useful.

That is why Claude Code and OpenClaw matter beyond productivity.

Anthropic is explicit that Claude Code is built to automate development tasks. OpenClaw is built to keep an assistant available from the chat surfaces people already use.

Those two ideas together tell you where the disruption starts: not necessarily with top-level strategy, but with the routine and semi-routine layers of knowledge work.

What work gets absorbed first

That includes:

  • first drafts

  • basic synthesis

  • repetitive research

  • junior coding

  • routine task execution

  • low-risk coordination

For years, that work was the apprenticeship layer.

It was how people learned:

  • how to judge quality

  • how to recognize edge cases

  • how to structure decisions

  • how to ask better questions

  • how to see failure patterns before they got expensive

It was not glamorous.

But it was developmental.

That is why the issue matters so much.

If agents absorb those layers, they are not just taking work.

They are starting to shape who gets to build judgment in the first place.

What the current labor signals suggest

That is why the labor data worries me more than the AI hype cycle.

Anthropic’s January 2026 Economic Index says 49% of jobs in its combined reports had AI used for at least a quarter of their tasks.

Anthropic’s earlier work on productivity found large task-level time savings in Claude conversations, while noting those gains do not fully capture refinement work needed for finished outputs.

The World Economic Forum says 40% of employers expect workforce reductions where AI can automate tasks.

Its January 2026 entry-level briefing says routine tasks in early-career roles are increasingly exposed and role expectations are changing.

You can read all of that two ways.

One way is the optimistic version: AI removes boring work and frees humans for higher-value tasks.

That is partly true.

The other way is the more dangerous version: AI removes the exact work people used to do to become capable of higher-value tasks.

That is the part too many leaders skip.

Why the apprenticeship layer matters so much

A society cannot just consume senior judgment forever.

It has to keep producing it.

And that is the real labor question:

If the apprenticeship layer gets compressed, how do we produce future experts?

This is why I do not think “learn prompting” is a serious answer.

Prompting does not magically create judgment.

Judgment is built through:

  • exposure

  • repetition

  • failure

  • correction

  • increasingly difficult decisions

If we automate too much of that too early, we may get short-term productivity and long-term weakness.

The hidden asymmetry of the AI era

This is where Bundle 2’s argument about encoded judgment stops being purely economic.

Because once judgment becomes valuable enough to encode and scale, another question follows immediately:

Who still gets the chance to build that judgment in the first place?

This is the hidden asymmetry.

A small number of people may gain enormous leverage by encoding their thinking into systems.

But if the junior layers disappear, fewer people may ever reach the point where their thinking is worth encoding.

That means AI is not just changing how work gets done.

It is starting to shape:

  • who gets to learn

  • who gets to develop taste

  • who gets to accumulate judgment

  • who gets to become expert

And once that happens, the next fight is not just about productivity or employment.

It is about authorship of human value itself.

The bridge to the next question

If agents begin to absorb the work that used to produce judgment, then the next question becomes unavoidable:

When that judgment is finally encoded into systems, who owns it — and who is responsible for what it does?

That is where the conversation has to go next.

Closing question

If AI eats the apprenticeship layer, who gets to become the kind of person whose judgment is worth encoding at all?