Managed Deep Agents v0.9 introduces Slack reactions API for dynamic loading states

Loading states are a critical aspect of agent user experience, providing visibility into agent actions to build trust, similar to skeuomorphic designs like the Domino’s Pizza Tracker. Managed Deep Agents v0.9 addresses this by introducing a new reactions API for Slack channels, enabling developers to create dynamic loading states and receipts.

The new reactions attribute accepts either a static string or a callable function that returns an emoji. This allows for simple heuristic-based responses, such as assigning a 🐛 emoji if a message contains the word "broken," or defaulting to 👀 for other inputs. This mechanism provides immediate acknowledgment to users, which is essential for tasks that run over long periods.

For more nuanced interactions, the API supports integration with decision models like Jev via the TypeSafe Classifier. Developers can define a vocabulary of emojis with specific criteria descriptions, allowing the model to select the most appropriate reaction based on context. For example, a 🔥 emoji might indicate a production incident for an on-call agent but a successful campaign for a growth agent, depending on the defined criteria.

Key features of this integration include:

• Confidence Thresholds: If the model's confidence in its top choice is below a set threshold (e.g., 25%), the system can fall back to a generic reaction or no reaction.
• Custom Emojis: Support for custom Slack shortcodes, such as lc-no-em-dash, allows for workspace-specific branding and meanings.
• Flexible Models: While Jev is highlighted, the system supports other TypeSafe-compatible decision models like SemIf hosted on LangSmith LLM Gateway.