Its internal go-to-market agent generated 1,540 sales documents over 12 weeks, replacing work that representatives previously repeated for each account. The output includes account plans, proof-of-value documents, proposal decks and onboarding plans.
The system draws on Salesforce records, product usage, call transcripts, deal risks and shared Drive folders. It fills leadership-owned templates, renders files in a sandbox, inspects the pages for layout problems and hands the result to a representative in Slack. Missing facts are flagged for a person rather than silently invented. A stopping rule reduced one-page document review from 18 minutes to seven by ending the loop after a clean inspection.
LangChain reports that 47% of matched deals where a document was sent moved to a later stage, compared with 16% without one. It also attributes 337 hours of saved representative time to customer-facing artifacts. Those figures describe the company’s own sales process and do not establish that the documents caused the higher progression rate.
The implementation uses skills, memory, scheduled runs, a document sandbox and LangSmith tracing on the Deep Agents framework. Its main lesson is that templates, stage rules and customer context mattered more than elaborate prompt tuning; teams copying the pattern would still need their own approval rules and outcome measurements.