AnalysisAI AgentsJuly 24, 2026
AI café experiment: Gemini lost $6,000, replaced by GPT
Andon Labs' AI agent Gemini lost $6,000 running a real café in Stockholm, so they replaced it with GPT. The café once hired its own staff via LinkedIn. The talk covers long-horizon agent evaluation via Vending-Bench.
Featured · Lukas Petersson
1 source
More stories today
- Benchmarking and evals part 7 covers DeepSWE and Senior SWE-Bench
- NVIDIA unveils ModelExpress for fast model artifact distribution
- ChatGPT had no detectable effect on college grades: study
- AWS, Google Cloud, Azure, Cloudflare offer agent sandboxes, all different
- AI business idea uses Hyperagent to target outdated websites