According to monitoring by Beating, LangChain has released a new component for Deep Agents called RubricMiddleware, which allows AI Agents to check and modify their outputs based on preset standards. Developers can clearly define the 'completion criteria' for tasks, such as code needing to pass tests, reports covering specified sections, and responses avoiding prohibited content. Each time an Agent is ready to deliver results, the system invokes a review model to check each item; if standards are not met, feedback is returned to the original Agent for further modifications until it passes the check or reaches the iteration limit. This mechanism addresses the common issue of Agents falling short on long tasks, where many Agents are not completely incapable but often overlook rigid requirements such as formatting, testing, citations, and sections. RubricMiddleware acts as an automatic quality inspector in the task workflow, helping Agents understand what constitutes true completion rather than just generating a seemingly adequate answer. LangChain's documentation also specifies that this approach is best suited for tasks with clear acceptance criteria, such as whether haiku syllables are correct, whether tests pass after code refactoring, and whether reports include all necessary parts. For ordinary users, its value lies not in making Agents better conversationalists, but in making them more like executors who can deliver work according to a checklist.