Human pair programming is valuable and comes with friction. Your pair has opinions about the right approach and won't always let them go gracefully. They get tired. They have a backlog of their own work pulling at their attention. They might feel awkward if you reject their third suggestion in a row. They might not say what they really think if they think it will cause conflict.

The AI assistant has none of these constraints. Reject its suggestion and ask for another — it has no investment in the first one. Ask it to argue against the approach it just proposed — it will, without defensiveness. Ask the question you'd feel embarrassed to ask a colleague — it won't remember it tomorrow. Push for a fourth alternative when the first three weren't right — it won't get frustrated.

This changes what you should ask for. Ask for the critique you've been avoiding. Ask for the alternative you haven't considered because the first approach seemed fine. Ask for an honest assessment of the code you've been maintaining for two years and quietly know isn't great. Ask for the explanation of the concept you should have learned long ago but have been faking your way through.

The ego-free dynamic also means the assistant is unusually good at adversarial tasks — finding the holes in your plan, listing the ways the implementation could fail, identifying the assumptions in your design that haven't been tested. These are tasks humans are reluctant to do for each other because they feel like attacks. The assistant does them as a natural response to being asked.

The absence of ego is a feature. Use it for the conversations you haven't been having.

Prompting for Code