Search papers, labs, and topics across Lattice.
AgentLens introduces a comprehensive benchmark for evaluating interactive code agents by assessing their entire operational trajectory rather than merely pass/fail outcomes. This approach integrates formal verification with LLM-generated reviews, allowing for detailed explanations of performance scores and facilitating a deeper understanding of agent behavior. The key result is that AgentLens not only ranks models but also aids in diagnosing issues, comparing agent versions, and identifying regressions in production settings.
Evaluating code agents through their entire interaction trajectory reveals critical insights that traditional benchmarks overlook.
We present AgentLens, a production-assessed benchmark for interactive code agents. Most code-agent benchmarks reduce a run to a single bit -- did the task pass? -- but the people who actually use these agents experience the entire trajectory: how the agent follows instructions, uses its tools, verifies its own work, recovers from mistakes, and talks to them along the way. AgentLens evaluates that whole trajectory. It pairs formal verification, where an objective check exists, with LLM-written trajectory reviews and side-by-side comparisons, so that each run yields a readable explanation of why the score is what it is. This makes AgentLens useful for more than ranking models: we use it to diagnose model behavior, compare successive versions of our own agent, and catch product regressions in a nightly evaluation pipeline. We release the benchmark as open source at https://github.com/agent-lens/agent-lens-bench.
