Strategist Evolves Its Own Agent Code

🔥 What happened Researchers from Meta and the University of Washington unveiled Strategist, a meta-agent that writes, tests, and refines inference strategies for AI agents on its own. Each strategy is stored as reusable code in a shared library, so later tasks build on earlier discoveries. 💡 Why it matters Across 30 benchmarks, Strategist beats the strongest baseline on 24 and matches average accuracy (74.7% vs. 74.0%) at 39% lower inference cost. The strategies also transfer to unseen tasks and other models without further search — the hand-written half of the agent stack now improves itself. ⚡ Our take This is the real lever: not the model, but the reasoning procedure around it is now being optimized. If you're still wiring agents by hand, you're about to optimize the wrong layer.