Ask it what is wrong

The thing an agent is unusually good at is the thing nobody enjoys asking a person for. Tell it to find what is wrong with your work and it will, without minding, at eleven at night, having read every line of it.

It is the move that comes up more often on this show than any other, and it is almost never about code.

What comes out of a session

A three month build and the business model under it, put in front of four agents at once and torn apart. A second agent sent in to review a month of the first one’s work. A site audit installed as a separate project so it could not mark its own homework. A psychologist asking it, plainly, what she was doing wrong as a prompter. An architecture review that came back saying there were too many layers for a practice with one person in it.

What it costs

Nothing beyond the quota. That is the whole point: a second opinion has always been expensive and slow, and this one is neither, so you can ask for it early enough to still change your mind.

The measurement worth taking

Before you ask, write down what you think it will say. The gap between your list and its list is the value of having asked, and it is usually the items you left off on purpose.