The Director's Chair

Most people I talk to about AI describe it the same way: it does the work for me. Which sounds liberating, until you actually try it for a week and discover that doing the work was never the hard part.

The hard part was always deciding what to do, knowing why, holding it to a standard, and steering when it went off track. Those are the things AI cannot do, and those are the things you are now responsible for, full-time, without the cover that being busy with execution used to provide.

The job is not easier. The job has changed.

The metaphor I keep coming back to is the film director. The director doesn't operate the camera, light the set, design the costumes, or deliver the lines. They do almost nothing that ends up on screen. And yet the film is their film. A different director shoots a different movie from the same script and the same cast.

What does the director actually do? They hold the vision. They know what the scene is supposed to feel like. They cast the right people. They notice when a take isn't quite right and they call for another one. They keep the whole sprawling production aligned to an intention that lives, mostly, in their head.

That is the job working with AI is becoming. And like film directing, the people who think it's the easy chair are the ones who fail at it.

Five capabilities make a good director — of a film, or of an AI-leveraged business — and only one of them is comfortable.

Clarity of intent. You have to know what you're trying to make. Not in vague terms — in specific, articulable terms. "Better customer experience" is not intent. "Reply to every customer inquiry within ninety seconds in a tone that sounds like our best account manager" is.

Standards under pressure. When the model produces something that's almost-but-not-quite right, you have to be the person who says "do it again". Most failed AI deployments are failed because someone shipped the almost-but-not-quite version. The standards were unenforced and the system drifted to mediocre.

Pattern recognition across many outputs. Directors watch a hundred takes and see the through-line. AI users see a hundred outputs and have to do the same: notice the systematic biases the model is making, the recurring shape of its mistakes, the places it consistently misses the mark.

Trust calibration. You have to know what the AI can be trusted with unsupervised and what still needs a human eye. That calibration is not static — it changes per task, per model, per stakes. Maintaining it is most of the work.

The willingness to say no. Directors get told "this take is good" by people whose careers depend on the take being good. They have to be able to override the room. AI users get told "this output is good enough" by their own fatigue and the small voice that wants the task to be over. They have to override that voice, every time it matters.

If you take the director framing seriously, the practical implications change everything about how you organise your day.

Time spent on execution shrinks. Time spent on intent grows. Time spent reviewing, redirecting, and re-running grows. Time spent thinking — yes, just thinking, with no immediate output — grows.

I notice this in my own work. The amount of code I write personally has gone down. The amount of time I spend deciding what is worth building, for whom, by what date, to what standard has gone up. The output has gone up. The leverage has gone up. The work has gone up too, because direction is harder than execution and there is now no escape into the comfort of "being busy".

It's worth saying out loud: a meaningful number of people will hate being directors of AI work, and the reason they hate it is interesting.

Execution was a hiding place. While you were doing the work, you could not be asked harder questions. You could not be asked why you were doing it, or whether it was the right thing to do, or whether the standard was high enough. The work was its own justification. I'm slammed is the most socially acceptable answer to any harder question.

Strip out the execution and the hiding place is gone. The harder questions sit on the table, every day, and there's nothing to be busy with except answering them.

Some people thrive in that. Others find it deeply uncomfortable. Both responses are honest. But the comfortable answer — go back to the way it was — is not available, because the market is going to reward whoever can sit in the director's chair, and punish whoever cannot.

If you have spent your career being good at doing things, the next several years will require a deliberate change. You will spend less time doing and more time deciding. You will have to articulate things you previously held implicitly. You will have to defend standards that previously defended themselves through the labour required to produce the work.

Done well, the trade is enormously in your favour. The leverage is real. The output multiplies. The work gets more interesting, not less. But the chair is harder to sit in than it looks from the outside, and the people who imagine it as a recline are the ones who slide out of it first.