Same AI, different hands

One person, then everyone else

Think of the person on your team who is brilliant with AI. They have built themselves a research assistant, a drafting workflow, maybe a small agent that preps their meetings. Now think of everyone else, using the same licenses to rewrite the occasional email.

Most organizations have one or two of those people for every hundred who have access. That single observation explains more about the AI payoff gap than any model benchmark.

Judgment is the scarce input

Modern AI is very good at task execution: reading, drafting, summarizing, reasoning through steps, writing code. It is not good at knowing your customers, your exceptions, or what actually matters in your work. A person has to supply the judgment.

Directing AI well looks like this:

• Knowing which parts of a job to hand to AI, and which to keep.
• Giving it the right context, and breaking the work into steps it can do reliably.
• Checking the output, and noticing when it is confidently wrong.
• Matching the shape of the solution to the shape of the problem: sometimes a simple prompt, sometimes a workflow, sometimes a small agent.

That judgment is a skill. It is unevenly spread, and it is learnable.

Why access programs plateau

Most AI programs are access programs: licenses, a portal, a lunch-and-learn. Access matters, but it plateaus fast. A license does not tell someone which of their weekly tasks is worth handing to AI. A course ends on Friday, and Monday’s work looks exactly the same.

A gap at work, and in life

AI raises the minimum skill that work demands. People who can direct it go further; people who can only consume it fall behind. We saw the same pattern with computer literacy. Inside a company, the risk is that the workforce splits into two groups and the people who can direct AI are the ones who leave. Outside work, it shows up as the gap between people whose lives get easier with AI and people who conclude it is not for them.

That is why Lucy Labs builds Lucy, an AI coach. You can’t easily redistribute wealth, but you can redistribute capability.

Winners redesign the work

McKinsey’s data points the same way: the companies capturing real profit from AI are nearly three times as likely to have redesigned how the work gets done. The redesign can only be done by people who understand the work from the inside. A strategy deck cannot do it. A vendor cannot do it. The people doing the job can.

That idea has a long history.

Next: We’ve been here before.