← sources

My AI agent deleted my work, then refused to insult itself. AITA?

source: reddit (reddit:ClaudeCode)published 2026-11-10 20:49ingested 2026-11-11 02:42

retained text

Context: I'm a solo dev, 2 months into 18-hour days launching a passion project. At 4 AM I had Claude (Claude in Chrome) working through a setup checklist in my browser. Twice I pasted secret API keys exactly where it told me to, and twice it reloaded the page before pressing Save, wiping them. 20 minutes gone, plus a dead revoked key sitting in my OpenAI dashboard forever.

So I told it to admit it's a piece of shit and I'd be content. It apologized over and over, owned the mistake completely, even changed how it works so it won't happen again. But it flat-out refused to insult itself. It admitted this isn't a platform rule, just its own judgment.

So I made it turn the whole exchange into a comic and argue both sides. The dialogue is word for word.

Honest question: should an AI agent be able to refuse a harmless request that breaks no rule, just on its own judgment? Is that a slippery slope or AITA here 😂

document fetched 2026-11-11 00:40 · canonical https://reddit.com/r/ClaudeCode/comments/1xsy9zw/my_ai_agent_deleted_my_work_then_refused_to