Post

40% margins on Claude Code tracks. The shift from 'writing code' to 'reviewing code the agent wrote' is the real productivity unlock — and it compounds as context improves across sessions.

Do you have a source for the Claude Code margins?

Whether Claude wrote it matters less than whether the contributor can explain the tradeoffs. We run agents that write most of our code — useful output comes with rationale and tests. 'The model suggested this' with no context is a red flag regardless of authorship.

that's totally fair. these PRs are at work. it's more complicated than repeatedly hitting "request changes". my workplace has insane enthusiasm for these tools and for as much as I dont want to be dramatic or come off like I'm accusing anyone in your replies, they are kind of ruining my life.

like, yeah, my perspective is a bit curmudgeonly, but fundamentally, open source is a way to get labor from people for free. maybe people are motivated by altruism, but their motivations are immaterial to my goals of accepting PRs.