Bayes now traces agent issues to the root cause in your agent's config

Bayes, Monte Carlo's AI teammate, makes your agents self-improving by connecting what goes wrong in production to the fix. It does this by orchestrating two capabilities: Conversation clustering, which groups production conversations by the problem users hit, and the Reinforcement Loop, which investigates those conversations, diagnoses the cause and proposes the fix.

This release makes that step more targeted and more specific.

You can control what the loop investigates. The Reinforcement Loop picks conversations from the agent's Issue clusters, including clusters built on your custom evaluation monitors, whether or not those monitors alert. Point it at SQL errors, slow answers, or a custom evaluation like "did the user leave with their request resolved?". All are enabled by default.

Issues name the thing to change. Each conversation is diagnosed against the agent's configuration:

• Platform agents (Cortex, Genie): agent configuration, verified queries, semantic views, search services, skills
• Code-based agents: the agent's code, skills and KBA's included in the agents code

Learn more here: Reinforcement Loop | Conversation clusters