LangChain has added a reactions API to Managed Deep Agents 0.9, allowing Slack-based agents to acknowledge requests before a longer task finishes. A developer can assign one emoji, choose through ordinary code or call a model to select from a controlled vocabulary.
The feature treats reactions as lightweight task receipts. A support agent might add eyes when it begins reading a message, a bug symbol for a reported defect or an hourglass when work is blocked. That immediate state helps users distinguish a running task from a lost request, especially when the final answer takes minutes rather than seconds.
Applications can supply a callable that receives message context and returns an allowed Slack reaction. LangChain also demonstrates using a decision model to map messages into predefined categories instead of asking a general language model to invent a response. More detailed status still requires a message explaining how the agent interpreted the work.
An emoji does not prove that a task is progressing or that the final result will be correct. It is an interface signal, and misleading automated status can erode trust. The API is most useful when reactions correspond to real lifecycle events and are paired with explicit updates for long-running, blocked or failed work.