How we built LangChain’s GTM Agent

LangChain published an account of how it built a go-to-market agent that researches leads, writes personalized email drafts, and aggregates account-level signals. The company says the agent, built on Deep Agents, produced a 250% increase in lead-to-qualified-opportunity conversion from December 2025 to March 2026, drove 3x more pipeline dollars in the same period, and saved each sales rep 40 hours per month, totaling 1,320 hours across the team.

The agent triggers on new Salesforce leads, performs checks for reasons not to send, gathers context from Salesforce, Gong, LinkedIn and web research via Exa, then sends a Slack draft with reasoning and sources for a rep to send, edit or cancel. It follows a defined outbound skill for warm and cold contacts. LangChain added a 48-hour SLA for silver leads, after which drafts send automatically if not approved or declined.

The company reports 50% daily and 86% weekly active usage among sales team members, and says reps increased follow-up with lower-intent leads by 97% and higher-intent leads by 18% since December.

Account intelligence runs weekly on Mondays, pulling from Salesforce and BigQuery and checking external signals such as funding rounds, product launches and AI initiatives. Reports are tailored for sales and deployed engineering teams.

Rep edits are compared against original drafts, with substantive changes analyzed for structured style observations stored in PostgreSQL per rep and read before future drafts. A weekly cron compacts these memories. Account intelligence uses compiled subagents with constrained toolsets and structured output schemas, spawned per account and run in parallel, with LangSmith Deployment handling scaling.

Evaluation uses rule-based assertions and an LLM judge in CI, with a test harness that mocks external APIs. Every rep Slack action is logged to LangSmith and attached to the trace. Human-in-the-loop review is mandatory before sending.

LangChain says a conversational Slack interface, built as a side experiment, spread to engineers, customer success and account executives who used the agent's existing access to Salesforce, Gong, BigQuery and Gmail for other workflows.