ClawWork: Open-Source Project Tests if AI Can Earn Its Own Salary
dr_cintas · x · 2026-08-10
Researchers introduced ClawWork, an open-source project that acts as a survival test for AI rather than a traditional benchmark.
- Core Mechanics: The AI starts with a $10 budget and must complete 220 real tasks across 44 professions (finance, healthcare, legal). Every API call drains its balance, and hitting zero means bankruptcy.
- Payment Formula: Deliverables are graded by GPT-4o using profession-specific rubrics. Payment = Quality × Estimated Hours × Actual BLS Wage.
- Results: In a 7-hour test with zero human input, the best models earned $10,000, equivalent to an hourly rate of over $1,500.
- Integration: It can also function as a live coworker in Telegram, Discord, Slack, and WhatsApp, where every message costs it real money.
More from coding & agent
- Codex spent 30 minutes investigating a 2-frame audio difference — ___Patrice___ · 2026-08-10
- Long-Horizon Agents Backfiring? Dev Slams Lack of Profitable Use Cases — MarcJSchmidt · 2026-08-10
- AI Reshapes R&D: The Future Holds Only 'Agent Engineers' — bindureddy · 2026-08-10
- Developer Prefers OpenAI Codex Over Claude for More Direct Code Generation — DuaneJRich · 2026-08-10
- Solving Dependency Gaps in Agent Skills: Open-Source Plugin Compiler — Low-Possibility9122 · 2026-08-10
- Kimi K3 Coding Test: Bionic Agent Boosts iPhone Decoding Speed by 60% — mattturck · 2026-08-10