LangChain has open-sourced the paid-media agent its marketing team uses to analyze campaigns, prepare reports and propose advertising changes. The system runs weekly across five channels, posts summaries and PDFs to Slack, and lets authorized staff approve actions such as adding keywords or changing geographic targeting.
The company says paid media grew from zero to 20% of its marketing pipeline in six months. Cost per qualified lead fell 30% from June to August while monthly spending rose about 60%. These are LangChain’s internal results, not an independent evaluation, and they reflect both the agent and the team’s wider advertising program.
A key engineering change was moving arithmetic and fixed safeguards out of the language model and into conventional code. Python now aligns dates, calculates metrics and applies rules, leaving the model to interpret results. LangChain says this made an early reporting workflow 40 times cheaper and 13 times faster, reducing runtime from 18 minutes to 85 seconds.
The agent runs in an isolated sandbox, loads business knowledge from files only when needed and delegates each advertising platform to a separate subagent. It can recommend changes, but designated users must approve them, after which code applies the update and checks the advertising platform to confirm it succeeded.