Patterns
Patterns automatically clusters production traces into recurring behaviors, so teams can see how users are actually using their agents at scale. Each cluster surfaces volume, cost, latency, and quality signals side by side, making it easy to spot emerging failure modes, expensive workflows, or unexpected use cases. Instead of sampling traces one by one, AI engineers get a living map of agent behavior in production, ready to drive the next round of evaluation and iteration.
Bits Eval
Bits Eval turns production signals into concrete next steps for improving your agent. It investigates traces, experiments, clusters of failures and regressions against your code, prompts, deploys, and downstream services, then proposes targeted fixes, new evaluators, and dataset additions to cover the gap. AI engineers can go from “something is off in production” to a fix in minutes, closing the loop between observation and iteration without leaving Datadog.