AI agents for lead generation, run to study them

The pitch for AI agents for lead generation is a pipeline that runs without anyone in it: find a business, qualify it, write to it, build it something, call it, close it. The fastest way to find out what that pitch leaves out was to run one. This one is an open-source project, not built here, adopted as a starting point and run end to end locally with real credentials wired up, specifically to study it.

It is roughly fifteen specialist agents behind a queue, with a dashboard showing their activity. A scout finds local businesses with no website, a verifier confirms the lead qualifies, copy is written and a site is generated and deployed, outreach reaches the owner by voice and message, and a closer books the call and takes payment. Every stage is a separate agent, which is what makes the architecture worth reading.

Studying someone else’s architecture properly is part of the work, and it is different from copying it. The useful questions are about the seams: what one agent hands the next, what happens when a verifier is wrong about a lead, and which stage is carrying the real judgement while the others carry out instructions. Those are far easier to see in a running system than in a diagram of one.

None of it is in production. It runs locally, which is exactly as far as it was meant to go, and whether any part of it becomes Rhomn’s own is still open. Knowing how a pipeline of roughly fifteen agents behaves while it is actually running is worth more than an opinion formed from reading its documentation.

It is labelled as adopted everywhere it appears, and on the case study the client field reads Adopted open-source template rather than a name. The pipeline is laid out there stage by stage.