AI development teams can now streamline the optimization of their agents by converting operational success into training data. LangChain released the smithtune CLI and LangSmith Fine-Tuning last week to let developers turn agent trajectories into fine-tuned models without building manual pipelines. This approach uses existing traces of an agent's successful performance as the primary signal for model improvement.

The new workflow encourages a cycle where every agent action is logged in Trajectories to be analyzed by both humans and other AI assistants. These logs then inform the creation of environments and evaluation tests for further review. By utilizing this process, LangChain aims to enable the deployment of open models that are faster and cheaper to run while maintaining high task performance.