What's New in Managed Deep Agents: schedules, per-run configuration, and Slack reactions

LangChain released Managed Deep Agents v0.9 today, available in Public Beta. The update adds three capabilities: agent-created schedules, per-run configuration, and Slack reactions.

The new Schedules SDK lets agents create reminders, follow-ups, and recurring tasks mid-conversation. A schedule runs as the person who requested it, using that person's permissions and connections, and results return to the same channel. Schedules inherit their creation channel, so a weekday reminder requested in Slack posts a new message to that conversation on every run. One-time schedules using "at" instead of "cron" reply in the original thread. The SDK is designed to sit behind a tool so users can create, list, update, and delete schedules by chatting; the agent passes a prompt and a cron expression, and LangSmith starts a new run each time the cron fires.

Per-run configuration lets one deployment choose its model, instructions, skills, MCP servers, and sandbox at the start of each run. The agent is defined as a callable function that receives the runtime and returns a define_deep_agent definition, reading context such as channel or repository. LangChain notes the configuration is set before the model runs, so the agent never sees skills or MCP tools outside its configuration, keeping context small and doubling as access control. It also lets the deployment pick the model, which the model itself cannot do, and avoids non-deterministic outputs when the correct toolkit is known in advance. An example shows one coding agent serving the payments-service repo with gpt-5 and Python skills, and storefront with gpt-5-mini and TypeScript web skills, selected via run context.

Slack reactions are on by default with a eyes emoji, telling senders the agent picked up their message while it reasons and calls tools. A reactions option on the Slack channel turns them off, picks another emoji, or selects one per message through a function; LangChain suggests a decision model like Jev to pick emojis quickly and cheaply. The release notes that v0.8 previously added per-user memory, custom HTTP channels, and built-in web search powered by Parallel.

Why it matters: teams building internal agents in Slack or their own channels get scheduling, model and tool selection, and message acknowledgment without maintaining near-copies of the same agent for each team or repository.