Agent Optimization is now available in public beta, giving teams a systematic way to improve their AI agents by running iterative optimization passes across models, configurations, and prompts - and deploying the best-performing result directly from LaunchDarkly.

Agent Optimization lets teams run structured optimization experiments against their AI agents to find the best-performing configuration. Rather than manually tweaking prompts and models and guessing at what works, teams can configure an optimization run that systematically tests variations across models and prompt strategies, evaluates each result using a judge model, and surfaces the winner so it can be deployed immediately.

Agent Optimization works with the LaunchDarkly Python AI SDK and integrates with popular agent frameworks. Configurations can be managed entirely in the AgentControl UI and pulled at runtime, or defined inline in code without any UI setup required.

To get started, read the Agent Optimization documentation.

Our team is happy to walk you through using Agent Optimization. To get help or share feedback, contact your LaunchDarkly account team or email product@launchdarkly.com.