LangChain has introduced RubricMiddleware for Deep Agents, a feature that lets developers define a rubric, configure a grader, and run agent outputs through a self-evaluation loop. The goal is to make agents check their own work against task-specific criteria.
That is useful for workflows where a plausible answer is not enough. Agents handling research, support, compliance, or coding tasks often need explicit standards for completeness, correctness, and format.
The release reflects a broader shift in agent development: teams are adding evaluation and correction loops directly into runtime systems instead of treating quality checks as a separate offline process.