ClawWork Turns Agents Into Simulated Workers and Breaks Them at $10
aigclink · x · 2026-07-26
A Hong Kong University research demo called ClawWork turns agents into simulated employees, starts each one with $10, and charges real API prices for every token until the agent goes bankrupt.
- The setup uses GDPVal, OpenAI’s benchmark of 220 real-world tasks across 44 professions.
- Each model must produce a result, then GPT scores the output against the profession’s rubric.
- Simulated income is calculated as quality score × BLS hourly wage × hours, so the system measures earnings under a mock labor market rather than actual customer work.
- The thread says most models score only 36%–43% on average, with the top performer ATIC + Qwen3.5-Plus at 61.6%.
- Interestingly, the model that “earns” the most is not the one with the best quality score, showing that the ranking depends on the interaction between evaluation quality and wage assumptions.
Related event: ClawWork: AI Agents Go to Work with Real API Costs(2 posts)→
More from coding & agent
- RTK is a Rust CLI proxy that cuts AI agent shell output by up to 90% — OpenAIDevs · 2026-07-26
- Daniel Lockyer recommends Moshi and herdrdev for remote agent coding and multiplexing — DanielLockyer · 2026-07-26
- ChatGPT Voice reportedly completed a Meta Business Suite flow hands-free — OpenAIDevs · 2026-07-26
- MCP server builders are debating how to scope agent permissions and API keys — Practical-Title7385 · 2026-07-26
- Open-source project targets the 90% of terminal output that coding agents waste context on — thetripathi58 · 2026-07-26
- RTK: Open-Source CLI Proxy Compresses Terminal Noise for AI Coding Agents — thetripathi58 · 2026-07-26