Real AI Agents and Real Work
One Useful Thing 9 months ago
OpenAI released a test showing AI systems now perform real economic work, with AI nearly matching human experts on four to seven-hour professional tasks across finance, law, and retail sectors. Claude Sonnet 4.5 successfully replicated complex economics papers with replication data in hours instead of the manual effort that would take days, demonstrating AI agents can now accomplish multi-step tasks autonomously using tools. AI's improvement in agentic capabilities means organizations will need to deliberately choose whether to use these tools for job replacement, thoughtless task proliferation, or collaborative work acceleration rather than letting implementation decisions default to cost-cutting.