LangChain's Paid Media Agent postmortem: treat agents like knowledge workers
LangChain · x · 2026-09-15
LangChain published a retrospective on how they built their Paid Media Agent:
- Treat agents like knowledge workers: The strongest results came from giving the agent a well-designed workspace — sandbox, software, business context, and clear operating instructions. The system prompt became 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 better 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 had to find the right tools, work within clear permissions, and move from analysis to action — proposing campaign changes, routing them through human approval, then verifying the changes.
Related event: LangChain Open-Sources Its Internal Paid Marketing Agent(2 posts)→
More from coding & agent
- Three days writing a 4-page decision-dense doc, then agents implement it all — lucasmeijer · 2026-09-15
- Codex app now installs natively on Arch Linux, no more workarounds — reach_vb · 2026-09-15
- Dev spends three days on a 4-page doc codifying every decision in his agent rewrite — lucasmeijer · 2026-09-15
- LangChain Labs lead teaches a 38-minute evals masterclass: tasks + verifiers — LangChain · 2026-09-15
- x402 hits $41M in volume as Alchemy ships a full guide to accepting agent payments — Thionne_WTZ · 2026-09-15
- OpenAI's Codex app officially supports Arch Linux via pacman installer — OpenAIDevs · 2026-09-15