AgentLens: A new benchmark for coding agent evaluation

Core summary

One sentence to understand this update

AgentLens is introduced as a new production-assessed benchmark that provides detailed trajectory reviews for evaluating interactive coding agents, moving beyond simple pass/fail metrics.

Impact & opportunity

What this could mean

Builders of coding agents can use AgentLens for more granular and realistic evaluation, improving the development and fine-tuning of agents by understanding their step-by-step performance.