HKUDS/ClawWork: "ClawWork: OpenClaw as Your AI Coworker - 💰 $10K earned in 7 Hours"

ClawWork reframes agent evaluation around economic survival: agents must earn income on real tasks, pay for their own tokens, and stay solvent. Instead of chasing benchmark scores, it measures work quality, cost efficiency, and long-term viability in a “live economic benchmark” that feels closer to production reality. The project’s multi-model arena and GDPVal task set give a concrete, comparable way to see who actually performs under pressure. I like the framing of “assistant → coworker,” especially with the explicit trade-off between working for income vs. investing in learning. It’s a useful push toward evaluating agents as workers, not just chatbots.