Building tracing and eval tooling for LLMs means chasing a moving target. With a dozen-plus providers, including OpenAI, Anthropic, Google, and others, shipping constant API changes, evolving response formats, and deprecations, it is nearly impossible for any team to keep up manually.

This talk will explore how we leverage AI to monitor, audit, and maintain the Braintrust AI observability SDKs. The result are libraries that are more robust, break less often, and easier to maintain for engineers.