You don't audit and review all $1k worth of tokens!

The AI might write ten versions. Versions 1-9 don't compile, but it automatically makes changes and gets further each time. Version 10 actually builds and seems to pass your test suite. That is the version you review!

—and you might not review the whole thing! 20 lines in, you realize the AI has taken a stupid approach that will obviously break, so you stop reading and tell the AI it messed up. This triggers another ~5 rounds of producing code before something compiles, which you can then review, hopefully in full this time if it did a good job.

▲

beefnugs 2 days ago | parent [-]

Thank you for this honesty. I mean if i try something 5 times and its a total failure every time, I would never in a million years think to try it another 5. (maybe i will give it another try when home hardware capable of a coding model is anywhere near affordable)

I guess I see why the salesmen dont mention this... but it seems really important for everyone to know?

	▲	Wowfunhappy a day ago \| parent [-]
		The scenario I described is still what I would consider "two shot" though—between attempts 1–10 I did not need to intervene. But it's true that I'm always surprised when people talk about using Claude on the beach or whatever, I love Claude Code but I have to test and test and test again per each incremental feature.