If AI Does the Work, Who Learns?

I found myself asking a question I did not like:

If I can do this faster with AI, then spend more time correcting your version, why do I need you?

We had hired smart interns and junior team members from some of the best schools in the world. They had access to tools that should have made learning faster than at any point in my career.

Yet I kept giving the same feedback: start with what the user needs, define what good looks like, and verify that the result solves the real problem.

The next draft often looked better. The judgment still lived with me.

Hiring experienced people did not automatically fix it. Strong titles could still produce polished work without enough framing, verification, or ownership.

The question was not junior versus senior. It was whether my feedback changed what the person could own next.

I was measuring the wrong thing

When I graduated into the financial crisis in 2008 and 2009, I was in New York trying to build a career in a terrible job market.

I worked constantly and treated every opportunity as something I might not get again. When I did not see the same visible urgency from people I managed, I read it as a lack of dedication.

That was too simple. I was comparing them with my memory of myself under different pressure. I was also using effort as a proxy for development.

Someone can work long hours and repeat the same mistake. Someone else can use AI, finish quickly, and show better judgment on the next assignment.

I needed to stop watching how busy people looked and start watching what they learned.

AI can improve output without building judgment

Junior work used to create both leverage and learning. Research, first drafts, small bugs, and bounded analyses forced a less-experienced person to encounter missing information, make choices, receive correction, and try again.

AI can now do much of that visible work. The artifact arrives faster, but it may hide what the person understood.

Research shows both sides. In a study of 5,172 customer-support agents, an AI assistant improved average productivity, with the largest gains going to less-experienced workers. Some gains appeared to persist when the tool was unavailable. The work had clear outcomes and strong examples people could learn from. Generative AI at Work

In a separate experiment, consultants used AI to perform data-science tasks beyond their existing skills. Their immediate performance improved, but afterward they did no better than the control group on technical questions without AI. Generative AI and the Temporary Upskilling of Knowledge Workers

That distinction changed what I look for:

A person can perform beyond their current skill without acquiring the skill.

The useful question is not whether AI helped someone produce a better draft. It is whether the person formed a view, checked the answer, saw the consequence, and carried the lesson into the next assignment.

I changed what I expect

I no longer judge a junior person's value mainly by the first draft.

I watch what happens next. Can they explain the user need and what good looks like? Can they show what they checked and where they are uncertain? Do they need the same feedback again?

I do not expect a junior person to already possess senior judgment. I expect to see judgment forming.

AI also raises my expectations for senior people. They should frame ambiguous problems, set quality bars, and make their judgment teachable. A senior person can be extremely productive and still become the bottleneck if every important decision stays inside their head.

The company has to create the next rung

There is no honest version of apprenticeship that costs no time.

Companies need real assignments that are small enough for a less-experienced person to own, with consequences they can observe and feedback they can use. Otherwise we will keep hiring people who already have judgment while removing the work through which anyone develops it.

I began by asking, “Why do I need you?”

Now I ask two different questions:

Is this person turning feedback into judgment? Are we giving them work through which that judgment can grow?