My toxic relationship with AI - Abeba NGWE
summary
Frontend at leboncoin moves from a 48-line monolithic cloud.md and ignored test commands to modular unit test skill files using the Anthropic skill standard, then consolidates into a broader creating feature skill. Claude initially picks the generic npx test command and wastes tokens; with the new skill it implements a list of classified-ads feature, uses the project’s npm run test check, and finishes in 5m32s using 23% context window versus ~12 minutes and 33% before, showing the larger-scope skill works.
AI-written summary · about
chapters (11)
- 0:00Introduction to the talk
- 1:47The honeymoon phase with AI
- 5:10Context engineering with Claude
- 7:44Red flags and hallucinations
- 10:13First failed skill demonstration
- 13:07Anger and micromanagement
- 15:45Couples therapy and documentation
- 19:04Reframing the approach
- 20:51Second successful demonstration
- 26:31Key lessons learned
- 30:22Conclusion and summary
related talks
18m"Software Fundamentals Matter More Than Ever"
01h 37mFull Walkthrough: Workflow for AI Coding
17mThe AI Skill I Rely On Daily
16mYou can’t write “AI” without an “I” Control and responsibility in the age of copilots
13mDeveloping with AI: Balancing Speed and Code Quality
23mHuman-Centered AI: Supporting, Not Replacing, Frontend Developers
26mAI and the Future of Testing
59mAI is Eating the Developer Career! Wait, what? Maybe it's the other way around?