Why every agent action needs evidence

AI can now write a lot of code very quickly. The hard part has moved: someone still has to check it. If an agent hands you a diff and nothing else, you are left to re-do its thinking before you can trust it.

Finished work, not suggestions

In Koy IDE, an agent works in its own isolated copy of your project. When it is done, the change is staged with its evidence — the diff, the test run, and a screenshot for UI work. Only then does it come to you.

An agent that can't show its evidence hasn't finished its work.

You approve, then it commits

Nothing is committed until you approve it. If something is off, you reject it with feedback and the agent tries again. Over time, steps that pass become skills and failures become lessons, so the agents get better at your project.