LangChain Shares Lessons on Proactive Agents

LangChain Shares Lessons on Proactive Agents
Wednesday, November 18, 2026
  • •LangChain says scheduled runs account for more than 70% of Managed Deep Agents activity
  • •GTM Agent and Robin use event, developer-set, and agent-set schedules to carry out work
  • •LangChain highlights memory and Slack-based review as ways to shape proactive agents’ check-ins

LangChain says its internal sales and engineering agents now handle work triggered by events and schedules, with more than 70% of runs on Managed Deep Agents coming from schedules rather than one-off human prompts. In a post dated November 17, 2026, LangChain describes memory and user experience as central to making these agents useful while limiting interruptions.

The agents respond to three trigger types: real-world events such as a new email or pull request; developer-set schedules such as nightly memory updates; and schedules created by the agent, such as a reminder or a requested daily brief. LangChain’s GTM Agent monitors leads and replies, sends daily briefs, and drafts sales emails. Robin, formerly Open SWE, drafts pull requests and reviews each new one. Both are built on Managed Deep Agents and use all three trigger types.

Memory gives an agent information it has learned over time, including preferences that are not supplied in each prompt. LangChain groups the sources of memory into seed information from the past, such as sent emails; present feedback, such as a user rewriting or rejecting a draft; and future outcomes, such as whether a drafted email receives a reply. The company says useful memory must also be maintained: agents need to judge what is worth saving, remove stale details such as passed launch dates, and merge new information instead of keeping contradictory duplicates.

LangChain says many agents use a scheduled background process called “dreaming” to review recent activity and update memory. To assess whether memory helps, developers can track outcomes such as how long drafts take to be approved, how often sent emails receive replies, and what share of an agent’s check-ins users act on. The company argues that remembered feedback can reduce repeated mistakes and support greater autonomy, such as allowing an assistant to book meetings after users stop changing its proposed times.

LangChain identifies three user experience priorities for proactive agents: high-signal notifications, quick review, and working where users already are. GTM Agent sorts conversations into those awaiting a user, needing a draft reply, or requiring no reply. Robin uses Jev as a decision model to determine when to join a Slack conversation. LangChain says review gives users a way to return feedback to memory; it also recommends aligning agent goals with work that people can review easily.

The company’s agents operate in Slack, where users can message them directly or tag them in channels. Agents can schedule follow-ups and post reports in the same thread; GTM Agent places daily briefs alongside existing conversations. Robin presents a one- or two-sentence headline before details in a thread reply, and its pull request review request includes a one-line summary, change size, and a claim button. LangChain says agent builders have spent the last three years on context engineering for models and now also need to consider how to give people useful context in a suitable format and at the right time. Managed Deep Agents supports schedules, memory, and asynchronous subagents; LangChain also points developers to memory and asynchronous subagents in the open-source Deep Agents harness, including its JavaScript implementation.

Subscribe to the AIB Newsletter

The fastest, most reliable AI news, straight to your inbox