LangChain outlines five core components for building internal company agents
·2026.10.25 08:49
Key point
The framework requires agents to live in Slack, handle identity via on-behalf-of or fixed credentials, use scoped memory, execute code in sandboxes, and run scheduled background tasks.
Details
- LangChain built three internal agents (GTM Agent, OpenSWE, and Content Studio) that changed how marketing, sales, and engineering teams work by solving multiplayer harness problems.
- The five core primitives are: living in Slack, handling identity (on-behalf-of vs. fixed credentials), maintaining global and user-level memory, executing code in sandboxes with secret proxies, and running scheduled or event-triggered background tasks.
- Identity handling varies by agent: OpenSWE acts on behalf of the user for PRs but uses its own identity for background code review; GTM Agent uses fixed credentials for Salesforce but on-behalf-of for email (DMs only); Content Studio uses on-behalf-of for Notion but fixed credentials for Twitter (whitelisted users only).
- All agents are built on Managed Deep Agents, which sits on top of the open source Deep Agents harness and provides infrastructure for channels, crons, sandboxes, memory, and identity.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.