LangChain AI Updates: October 26, 2026

1. LangChain Lays Out the Five Components of a “Company Harness” for Internal Agents

LangChain published a post by Harrison Chase and Caspar Broekhuizen on how it runs internal agents for sales (GTM Agent), engineering (OpenSWE, which drafts and reviews PRs), and marketing (Content Studio). It names five components a company harness needs. The agent lives where people already work (Slack, with a separate Slack app per agent). It knows who is asking, so OpenSWE opens PRs under the user’s GitHub handle when messaged directly but uses its GitHub app identity in background runs. It keeps both global and per-user memory, runs code in a sandbox that can make authenticated requests, and runs on schedules or events, for example user-created daily-brief crons or a run on every new PR. The post maps these pieces to LangSmith Sandboxes with a secret proxy and to Managed Deep Agents, and it calls memory the least-explored area so far. Source