Agents trained with EnvHarness consistently outperform baselines. Under identical budgets, while traditional environments quickly plateau, EnvHarness keeps improving as environments scale!

sooo.. training an agent on real environments made it worse. 😳 Google just dropped EnvHarness, and the number that stopped me is on SpreadsheetBench. base agent scores 46.4. train it on real…

Google releases EnvHarness A programmable layer that turns static agent environments into adaptive ones, adding plug-in components (Setup, Rule, Link) without modifying the original code. Up to +9…