Niklaus Gerber

reading the end of software engineering paper (arxiv 2606.05608). its own benchmark has agents above 80% on one-off tasks and at most 38% when they maintain a codebase over time, because early mistakes carry forward. writing code got cheap, keeping a system correct didnt. feels like the real job is now writing intent down clearly and checking agent output with a different method than the one that wrote it. are you spending more time writing code or reviewing what agents wrote?

agents made prompt edits scary for me. tweak one system prompt and a tool call 3 steps later starts failing. what helped: prompts live in git, every change runs against a small fixed set of real traces before shipping, and the last good version is always one revert away. curious what others do here, do you run evals on every prompt change or only the big ones

The growth example hits home because AI is terrifyingly good at agreeing with whatever hypothesis you feed it.

If your starting prompt is biased, the tool will just find data to make you feel smart about a wrong decision.

True collaboration only happens when you force the model to look for counter-arguments or blind spots you missed. Have you found a reliable way to get AI to challenge your strategy instead of just echoing your thoughts back to you?