Descriptions:
ClawWork is an open-source framework built on top of OpenClaw that reframes AI agents as economic actors — each LLM call costs real money deducted from a starting $10 balance, and the only way to grow that balance is to complete professional tasks scored by a separate LLM evaluator against U.S. Bureau of Labor Statistics wage data. Fahd Mirza walks through the architecture, live leaderboard results, and a full setup guide in this detailed overview.
The benchmark suite spans 44 occupations and 220 professional tasks drawn from the GDP-W dataset, covering roles from software developer to compliance officer to nurse. On the leaderboard, ATIC combined with Qwen 3.5 Plus converted a $10 starting balance into $19,915 in simulated earnings over 8 hours, while AIC DeepC ranked fifth by income but posted the highest quality score of any agent at 66.8%. The architecture routes all messages through OpenClaw’s nanobot gateway, where a tracked provider intercepts every LLM call and deducts cost in real time.
Mirza is careful to flag four key caveats before anyone runs this: API costs can spiral rapidly because both the agent and the evaluator are billed simultaneously; the income figures are simulated economic value, not real revenue; the agent has full tool access; and setup complexity is significant. A full HTML setup guide and command reference are available in his GitHub repository, along with prior videos covering OpenClaw installation.
📺 Source: Fahd Mirza · Published March 08, 2026
🏷️ Format: Hands On Build
