Insights

Practice used to need a person. Now it doesn't.

The thing that actually makes you ready, practice that pushes back, always needed a person. That's why it was rationed. Not anymore.

Think about how you actually got good at anything that mattered. Odds are it wasn't a video. It was someone on the other side: a manager who let you sit in, a mentor who asked the awkward question, a colleague who ran the role-play with you in an empty meeting room before the real one.

That person was the secret ingredient, even if it never felt like a system. They could respond. They pushed back, adjusted to you, and told you the truth while it still mattered. And they were almost always in short supply.

The bottleneck was always human

Static content scales beautifully. One video teaches a million people; one article answers a question forever. That's the whole strength of information, and it's why there's so much of it. But a video can't notice you've misunderstood. It can't make the scenario harder because you're coasting, or easier because you're drowning. It can't answer the specific question you actually have at 11pm the night before it counts.

Everything that could do those things ran through a person. And a person comes with the oldest constraint there is: time. Their attention doesn't copy and paste. An hour they spend coaching you is an hour they're not doing their own job, which is why that hour was always rationed.

So the most valuable kind of learning was also the least scalable, and we organised the whole world around that fact without quite noticing. Mentorship became a perk. Good training became a budget line. Rehearsal became something you got if you were lucky enough to sit next to the right person.

What only a person could do

Strip it back and the scarce ingredient was two abilities, not one.

The first is to respond: to react to what you actually said, not what a script expected, and to change the next move accordingly. The second is to tell you the truth: to watch you do the thing and say, honestly, what was off, while you could still feel why. Content can do neither. A person could do both, which is exactly why a person was the bottleneck.

The scarce thing was never information. It was someone who would answer back.

Readiness became a lottery of access

Follow that through and you get a quietly unfair outcome. If you joined a firm with a strong apprenticeship culture, you got ready fast, because responsive practice was all around you. If you didn't, you learned on the job in front of real stakes, picking it up slowly and paying for the lessons in mistakes.

The same was true for individuals. The people who could afford coaches, or who happened to know someone generous and experienced, had a rehearsal advantage that had nothing to do with talent. Readiness tracked access, not ability. We just called it experience and moved on.

What actually changed

AI didn't make information cheaper. Information was already free; we were drowning in it. What AI made cheap is the responsive part, the bit that used to need a human.

A model can hold a real conversation, take a role and stay in it, raise the difficulty when you get comfortable, and give feedback that's specific rather than polite. For the first time, the thing that used to require someone's scarce hour can happen on demand, at two in the morning, fifty times in a row if that's what it takes. The bottleneck didn't get wider. It got removed.

  • Available, not scheduled. The rehearsal is there the moment before it counts, not next Tuesday when your mentor's free.
  • Patient and tireless. You can be bad at it repeatedly without burning a favour or a colleague's afternoon.
  • It scales its difficulty to you. The good mentor's real trick was calibration: easing off when you were drowning, turning the screw when you'd got comfortable. That adjustment, not the raw hours, is what made their time worth so much, and it's the part that can now happen automatically.
  • The same standard for everyone. Not just the people lucky enough to sit beside the right person.

What this is not

It's worth being precise, because the hype isn't. None of this means a model is better than a brilliant human coach. A brilliant human coach is still one of the best things that can happen to a career. But most people, most of the time, never had one. The honest comparison isn't AI versus the best mentor you can imagine. It's responsive practice versus the far more common alternative: no practice at all, and a real customer or a real interview as your first attempt.

Why it matters more than it sounds

When a capability stops being scarce, who gets to be good at things changes. Take away the human bottleneck and the old link between readiness and access starts to loosen. The rehearsal that used to belong to people with mentors and budgets becomes something anyone can have.

That shift is the thing we're built on, and it's why we point a single engine in two directions: at the individual and at the team. The individual gets the rehearsal that used to require a tutor. The team gets the onboarding that used to require its best people to stop and shadow. Same move, made possible by the same change.

The coach on the other side was always the point. For most of history, there just weren't enough of them to go round. The interesting thing about now is that, finally, there can be.