LangChain Lets Managed AI Agents Schedule Follow-Ups From Chat

A request in chat can now give a managed AI agent work to do later, using the requester’s permissions and returning results to the same channel. LangChain released Managed Deep Agents v0.9 in public beta on October 7, 2026, adding conversation-created schedules, per-run model and tool choices, and configurable Slack acknowledgments.

A conversation becomes a schedule

The new Schedules SDK lets agents create reminders, follow-ups and recurring tasks during a conversation. Developers can expose it through a tool so people can create, list, update or delete schedules by chatting, rather than handling those changes outside the conversation.

For recurring work, the agent supplies a prompt and a cron expression—a timing rule for when the task should repeat. Each time that rule fires, LangSmith starts a new run with the prompt. LangChain’s example creates a daily digest scheduled for 9 a.m. on weekdays in the America/Los_Angeles timezone.

Scheduled work runs as the person who requested it, with that person’s permissions and connections. Recurring results post back to the conversation where the schedule originated. One-time schedules instead use a specified execution time and reply in the original thread, fitting a request such as checking a deployment in an hour.

One deployment, different team toolkits

Per-run configuration lets developers define an agent as a function that receives runtime information and returns its setup. That function can choose the model, instructions, skills, connected tool servers and sandbox for each run. A single deployment can therefore serve different teams or repositories without maintaining a near-copy of the agent for each one.

LangChain illustrates this with a shared Slack agent that loads billing skills for a finance channel or incident-response tools for a platform team. Its coding example chooses different models and instructions for different repositories. The example also includes a default configuration for cases where no repository context is supplied.

Slack gets an acknowledgment before an answer

Slack reactions tell the sender that the agent picked up a message while it reasons or calls tools. The eyes emoji is enabled by default. Developers can disable reactions, choose another emoji or use a function to select one for each message.

The release post demonstrates a simple rule: show a bug emoji when a message contains “broken,” and eyes otherwise. It also describes using a model to select the reaction. For developers starting a new agent, LangChain supplies these initialization and deployment commands:

uvx --from managed-deepagents mda init my-agent
cd my-agent
uv run mda deploy