There is a question about AI that we don’t ask often enough.

Not:

What jobs will AI replace?

Not:

How much faster will AI make us?

But:

How will people learn to become good at the work when AI does the work before they have learned how to do it themselves?

That question worries me more.

For decades, careers followed a relatively predictable pattern.

You started at the bottom.

You did the repetitive work.

You made mistakes.

Someone corrected you.

You watched someone more experienced.

You handled increasingly difficult problems.

Eventually, you developed judgment.

The boring work wasn’t always glamorous.

But it was often where the learning happened.

Now AI is coming for the boring work first.

And that sounds like progress.

It is.

But there is a paradox hiding inside it.

The work we most want to automate may also be the work through which humans learn.

The Apprentice Layer Is Disappearing

Consider software.

A junior engineer used to learn by:

Reading unfamiliar code.

Debugging something they didn’t understand.

Writing the wrong abstraction.

Breaking something.

Tracing the problem.

Asking someone senior.

Trying again.

Eventually, they developed intuition.

Not just the ability to produce code.

The ability to recognize good code from bad code.

Now an AI agent can generate the code.

Explain the code.

Refactor the code.

Write tests.

Debug the code.

Create the pull request.

And increasingly, execute the task from beginning to end.

OpenAI reports that by May 2026, more than 70% of sampled Codex users had asked it to complete a task estimated to require more than an hour of human work, while usage has expanded rapidly beyond engineering into areas such as legal, finance and recruiting.

The productivity story is obvious.

The learning story is not.

We May Be Automating the Wrong Part of Experience

Experience is not simply exposure to tasks.

Experience is what happens when you struggle with a task long enough to develop judgment.

That distinction matters.

AI can give you an answer.

It cannot automatically give you the experience of becoming the person who knows when that answer is wrong.

This is the emerging paradox:

AI can compress the time required to produce an outcome while simultaneously removing some of the experiences required to develop judgment.

That is a very different problem from job displacement.

It is a judgment pipeline problem.

The Junior Role Was Never Just About Output

This is where organizations may make a dangerous mistake.

If we measure junior employees only by output, AI makes them look increasingly unnecessary.

But a junior employee was never valuable only because of how much work they could produce.

They were also an investment.

An apprentice.

Future capability in formation.

If AI does all the low-level work, organizations have to become much more intentional about how people develop.

Otherwise we create an unusual workforce:

People with access to extraordinary tools…

but insufficient experience to judge what those tools produce.

The New Skill Isn’t Prompting

This is why I don’t think the future belongs simply to people who know how to prompt AI.

Prompting is becoming a commodity.

The harder skill is judgment.

Knowing what problem is actually worth solving.

Knowing what good looks like.

Knowing what constraints matter.

Knowing when an answer is technically correct but operationally wrong.

Knowing what question to ask next.

Knowing when not to trust the machine.

AI increases the value of context.

Not decreases it.

And this is already showing up in adoption.

Salesforce’s July 2026 India research found that 38% of surveyed workers said their organization had experienced an unsuccessful AI pilot in the previous year. Among Indian respondents whose pilots failed, lack of business context was the most cited reason, at 34%.

The lesson isn’t that AI doesn’t work.

It is that generic capability without context doesn’t create business value.

The New Apprentice Needs Different Work

If AI takes the old apprenticeship tasks away, we can’t simply tell young professionals:

“Use AI and figure it out.”

We need to redesign the apprenticeship.

Give people problems where judgment matters.

Let them own outcomes.

Make them explain decisions.

Have them review AI-generated work.

Let them challenge the machine.

Put them close to customers.

Expose them to failure.

Let them sit in uncomfortable meetings.

Let them see how decisions are actually made.

The new apprentice may write less code.

But they may need to understand more systems.

They may produce fewer documents.

But they may need to understand more context.

They may execute fewer tasks.

But they may need to make more decisions.

The Manager’s Job Changes Too

This is where leadership becomes interesting.

Managers can no longer simply ask:

“How much did you produce?”

They have to ask:

“What did you learn?”

“What did you decide?”

“What did you catch that the AI missed?”

“What would you do differently next time?”

The manager becomes less of a task allocator.

More of a judgment architect.

There Is Another Danger

AI can create the illusion of competence.

Someone can produce an impressive result without understanding how the result was produced.

The output looks senior.

The understanding isn’t.

That gap may remain invisible until the system fails.

And when it does, someone has to understand the system deeply enough to recover it.

This is why I keep coming back to something I’ve written about before:

Working code isn’t the same as understanding.

The same principle is now becoming true across knowledge work.

A finished output isn’t the same as developed judgment.

What Should Leaders Do?

Perhaps the answer isn’t to protect junior work from AI.

That would be like protecting calculators so students can learn arithmetic.

Instead, we need to redesign the path to competence.

Move people faster toward:

context → judgment → ownership → consequence

And use AI to remove the mechanical friction around it.

Let AI write the first draft.

But make the human defend it.

Let AI analyze the data.

But make the human explain the decision.

Let AI generate the architecture.

But make the engineer own the trade-offs.

Let AI execute the task.

But make the human understand the system.

The Human Layer

The question isn’t whether AI will do more of our work.

It will.

The deeper question is:

What happens to human development when the work that used to teach us is automated?

We may be entering an era where organizations can produce more with fewer people.

But that creates a responsibility we haven’t fully confronted.

Who is developing the next generation of judgment?

Because you cannot automate your way into a leadership pipeline.

You cannot outsource experience.

And you cannot delegate responsibility for becoming capable.

Leadership Reflection

Ask yourself:

• What work are we automating that people previously learned from?
• Are our junior employees developing judgment—or just producing AI-assisted outputs?
• What experiences do people need to become trustworthy decision-makers?
• Are we measuring productivity while accidentally starving capability development?
• If AI does 80% of the execution, what are we deliberately teaching humans to do with the remaining 20%?

The Human Layer Insight

The greatest risk of AI may not be that machines become capable enough to do the work. It may be that humans stop getting the experiences required to become capable enough to understand the work.

The future of work isn’t only about what AI can do.

It’s about what humans still need to learn.

— Majid Nisar

Read the full issue on LinkedIn →

THE HUMAN LAYER publishes on LinkedIn. Subscribe here.