Skip to content
Hot eventDeveloping

LangChain details the shared harness behind its internal agents

2 reports1 reporting sourcesUpdated 7 h ago

Start here

AI digest

LangChain published the five shared building blocks behind its internal agents for marketing, sales and engineering: running inside Slack, handling identity, two-tier memory, sandboxed code execution and scheduled background runs. Its GTM Agent, OpenSWE and Content Studio all run on Managed Deep Agents, which sits on the open-source Deep Agents harness and supplies channels, crons, sandboxes, memory and identity. The post describes how each agent mixes on-behalf-of credentials with fixed credentials, and states that memory is the least explored of the five areas. In a follow-up, LangChain says more than 70% of runs for its two proactive agents, GTM Agent and Robin, now come from schedules rather than human prompts. It argues that memory built from seed context, user feedback and outcomes is what lets an agent act without a prompt, and that check-ins must be high-signal and quick to review, delivered in Slack where users already work. LangChain notes that keeping memory useful is hard because of salience, expiry and duplication, and that scheduled background jobs it calls "dreaming" are a common way to update it.

Generated from reports · updated 7 h ago

Developments

2 developments

  1. November 18 00:49 · 1 reports
    Sharing lessons from building proactive agents
    LangChain Blog:LangChain on building proactive agents: memory and UX lessons from GTM Agent and Robin
  2. November 17 14:39 · 1 reports
    Share five components of building a company harness
    LangChain Blog:LangChain outlines five components for building a company harness of internal agents

Report timeline

Follow the reports to see the event from different sides.

November 18
  1. LangChain Blog
    LangChain on building proactive agents: memory and UX lessons from GTM Agent and Robin

    LangChain outlines what it learned running two internal proactive agents, GTM Agent and Robin, on Managed Deep Agents, where more than 70% of runs now come from schedules rather than human prompts. The post argues that memory built from seed context, user feedback and outcomes is what lets an agent act without a prompt, and that check-ins must be high signal and fast to review, delivered where users already work in Slack. It also notes that keeping memory useful is hard because of salience, expiry and duplication, and that scheduled background jobs it calls dreaming are a common way to update it.

November 17
  1. LangChain Blog
    LangChain outlines five components for building a company harness of internal agents

    LangChain describes the five shared primitives behind its internal agents for marketing, sales and engineering: living in Slack, handling identity, two-tier memory, sandboxed code execution and scheduled background runs. Its GTM Agent, OpenSWE and Content Studio all run on Managed Deep Agents, which sits on the open source Deep Agents harness and provides channels, crons, sandboxes, memory and identity. The post details how each agent mixes on-behalf-of and fixed credentials, and notes that memory remains the least explored area.

Heat for this event

Current, comparable range
8
Peak, comparable range
10November 18 01:00
Change in 24 hours
–

The trend compares the same participants over hours observed continuously, so its range can be narrower than the current heat count. Hover or tap the chart for hourly heat; the arrow keys move through it.