Don't micromanage your coding agents

An agent should just make edits to your files. It shouldn't ask at every turn, "Hey, I want to make this change, should I apply it?" If it's asking while it's already done the wrong thing, it's already wasted your time.

We need to move out of the inner loop and sit on top of it - still steering and guiding, but not micromanaging every change.

Anthropic people call yolo mode as --dangerous-skip-permissions and flags this risky. Still, almost everyone I know runs yolo mode all the time, because it's shockingly effective.

So far it hasn't done anything malicious on my machine. It has the permissions to do some wild things, but for whatever reason they did a really good job and it just works.

There are obvious risks. Prompt injection is not solved. But in practice the trade-off seems acceptable for the kinds of projects you typically use coding agents.