DeepMind engineer lets Gemini 4 Carbon burn unlimited tokens to auto-debug a Kaggle PR
thursdai_pod · x · 2026-11-15
From the ThursdAI podcast: engineer Ivan Leo hit 400 errors on Kaggle's model proxy and handed the problem to Gemini 4 Carbon with unlimited tokens. The model located the PR that changed permissions, identified who merged it, and drafted the message to send them. The story comes from the live AI Engineer episode covering "Jensen is OUT!?", Fable 5.2, the Anthropic IPO, and Gemini 4 Carbon/Hydrogen — a showcase of what agents can do with large token budgets.
More from coding & agent
- Agents can generate 3D assets but can't verify them — a visual AI infra gap — PrateekVJoshi · 2026-11-15
- A /pareto-analyze skill that finds the biggest contributors to any problem — rchardkovacs · 2026-11-15
- When your coding agent outruns your understanding, make it explain the app before building more — brandon_galang · 2026-11-15
- Restoration pipeline details: removing ink blobs via render-vs-original comparison — generativist · 2026-11-15
- Developer rebuilds 1920s Popular Mechanics page-by-page with an agent pipeline, ~30 human decisions — generativist · 2026-11-15
- Ditching Claude Code ban anxiety: Factory.ai as an alternative coding agent — vista8 · 2026-11-15