Pinned
It just got much easier to evaluate agents in production—against real business outcomes.
Agent Evals are now in public beta on v4.8.0 of the TS SDK, and include:
💯 𝗦𝗰𝗼𝗿𝗶𝗻𝗴: Attach a metric to any run or step. Include outcomes that don't arrive until after the run
