Coach, Don't Play

The first time I handed an agent a piece of work I could have done myself in an hour, I sat and watched it run. Not because I distrusted the output. Because I had an hour with nothing in it, and watching felt like participating.

That reflex is the problem, and it took me longer than I would like to admit to name it.

The asymmetry is older than the technology

Coaching rests on an arrangement we normally find unremarkable: the person calling the plays is usually not the best athlete on the field, and often not close. The coach's contribution was never superior execution. It is knowing which play, against which opponent, at which moment — and having watched enough film to know why the last one failed.

Nobody finds this scandalous in sport. We accept it in an orchestra, where the conductor produces no sound at all. We accept it in a kitchen, in an operating theater, on a film set. A director cannot act, shoot, score, or cut as well as the specialists doing each of those jobs, and the profession is not embarrassed about it.

Then the same arrangement arrives in knowledge work and it lands as a demotion.

What actually changed is narrower than the headlines

Execution got cheap. That is the shift, and it is worth stating in the smallest possible terms because the large terms are doing damage.

For most of the last century, the scarce thing in professional work was the ability to do it — write the query, draft the contract, model the scenario, produce the deck. Judgment mattered, but judgment without execution was a person with opinions. Execution was the toll you paid to be taken seriously, and paying it repeatedly is how you acquired the judgment in the first place.

That toll is collapsing. Not the judgment — the toll. On a widening set of tasks, an agent will produce a competent first pass faster than you can open the file, and it will do it again, differently, if you ask. What remains scarce is knowing which pass was any good and what to do about the one that wasn't.

Which is a coach's job description.

The craft is not the tooling

The mistake I see most often — I made it myself for months — is treating this as a tools problem. Learn the prompts, wire up the integrations, and the role change comes free with the software.

It doesn't. The tooling is a weekend. The role is a season.

A coach does four things, and only one of them is instruction. They set boundaries: what is in play, what is out, what done looks like. They make strategy: which problems get attacked in which order, and what we are deliberately not doing. They steer tactics in the moment, when the plan meets a situation the plan did not anticipate. And they give feedback — specific, repeated, unsentimental — until the pattern sticks.

Three of those four are things you cannot delegate to the thing you are directing, because they are the direction.

The failure mode is a person who has automated their execution and kept none of these. They hand over a vague problem, receive a plausible artifact, and have no basis for judging it beyond whether it looks like the sort of thing they expected. That is not coaching. That is buying.

The playbook is the artifact

Here is where it stops being an attitude and becomes work you can actually do.

A season of coaching produces something durable, and it is not any individual result. It is the playbook: the accumulated record of which setups worked, against what, and why the ones that failed failed. Coaches keep this obsessively. It is the reason a good one is worth hiring even when the roster changes underneath them.

Directing agents produces the same artifact, and it is almost never kept.

Run a play. Watch what comes back. Keep what worked, cut what didn't, write down which was which — the framing that produced a usable draft, the constraint that stopped the wandering, the phrasing that reliably got you a real answer instead of a confident one. Then run it again with what you learned. Not once. Repeatedly, the way anyone gets good at anything.

Most people iterate on the output and throw away the play. They fix the draft, ship it, and start the next problem from zero. The draft was never the valuable part.

One thing to keep straight

All the vocabulary in this piece — team, captain, feedback, patience, practice — is borrowed from relationships between people, and the borrowing does something to you. Manage an agent for a few weeks and you will catch yourself relating to it: softening a correction, feeling briefly bad about a blunt instruction, extending it the benefit of the doubt you would give a junior colleague on a bad Tuesday.

There is nothing there to receive any of it. No stakes, no shame, nothing to appeal to. It is not malicious — it cannot be, which is exactly the problem. Coach it like a team. Do not confuse it for one.

Nothing stops you playing

The craft is genuinely satisfying once it starts working, in a way that surprised me. Watching a well-framed problem come back solved by three specialists you pointed in the right direction is its own pleasure, and it is not a lesser one.

And nothing stops you playing when you get the itch. I still write code by hand some evenings, badly and slowly and entirely for my own entertainment. That is a fine reason to do it.

Just don't do it because the job needs you on the field. Increasingly, it doesn't — and the seat that's opened up is one people have been doing well for a very long time.