LangChain's Engineering Breakdown: Building a Paid Media Deep Agent That Lives in Slack

LangChain · x · 2026-09-17

LangChain published an engineering breakdown of their in-house Paid Media Agent: a Deep Agent living in Slack that reads spend, clicks, and conversions across 6 ad platforms plus their data warehouse every Monday, explains what changed and why, proposes campaign changes, and applies them once approved.

Key design lessons:

Treat agents like knowledge workers: give them a well-designed workspace with a sandbox, software, business context, and clear operating instructions. The system prompt acts as a map helping the agent find what it needs instead of carrying everything in context.
Models for judgment, code for consistency: calculations, source-of-truth rules, and safeguards were handled in code—making the agent faster, cheaper, and more reliable—while the model focused on interpreting results and recommending next steps.
Design around the full workflow: the agent finds the right tools, works within clear permissions, and moves from analysis to action—proposing changes, routing them through human approval, and verifying they were applied.
Use abstractions: Managed Deep Agents handles hosting, sandboxes, Slack integration, and schedules so you focus on tools, context, and decision rules.

Context: after three years of organic growth via open source, content, and community, LangChain kicked off paid advertising in January, scaling to 5 paid channels in 6 months with a small marketing team.

Original post →

LangChain's Engineering Breakdown: Building a Paid Media Deep Agent That Lives in Slack

More from coding & agent