Debugging
The bugs that don't throw errors. Silent failures, wrong assumptions, and the detective work behind each fix.
An Empty AI Response Corrupted Chat History
Gemini returned HTTP 200 with zero content. I saved it. The conversation never recovered.
Three Bugs That Were Actually My Prompts
Three debugging sessions where I spent hours chasing AI misbehavior — and each time discovered the model was executing my instructions exactly as written. The bug was me.
The Production Bugs That Never Threw an Error
Six bugs across OAuth, Next.js, launchd, n8n, browser APIs, and OpenAI. Every log said success. Every result was wrong.
When "Polish Over Security" Costs Real Money
Client wanted to "polish features first, security later." Found exposed OpenAI API key in frontend code. Anyone could steal it and rack up unlimited charges.