Browserbase and Microsoft Research developed Universal Verifier to assess whether a browser AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → completed a task when several different sequences of clicks could succeed. Their approach uses explicit, non-overlapping criteria and separates the outcome from the process used to reach it. Browserbase reports near-zero false positives, meaning failed tasks incorrectly marked successful, in its evalA repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.Full definition →. This distinction also matters during training: rewarding an apparent success can reinforce the wrong behavior. The company's account describes judging evidence of completion instead of requiring one predetermined path.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
videoYour LLM Judge Is a Confident Liar: Building Better Verifiers — BrowserbaseAI Engineer
articleUnit Testing Ai Agents In The BrowserBrowserbase
articleThe reach of a verification tool decides its value: A controlled study of verification surface, artifact quality, and cost in AI coding agentsAchint Mehta