An honest, documented account of building an autonomous multi-agent ad campaign platform from scratch. Every architecture decision, every pivot, every honest failure — published as it happens.
We started with AWS Step Functions and Terraform. Then we scrapped it. Here is the honest reason why local-first won, and what LangGraph changed.
Read chapterRebuilding the 11-agent pipeline on LangGraph with the same JSON contracts, same human approval gates, and a foundation that runs on a laptop.
Read chapterThe pipeline ran end-to-end for the first time. It also produced a misleading output that said campaigns were launched when they weren't. We fixed it.
Read chapterWe stopped and asked: is this what we actually promised to build? A full audit of what exists, what's missing, and the right next move.
Read chapterTwo empty agent contracts written, a build-progress publishing mode added, and the CLI command that lets the system publish its own journey chapters.
Read chapterX, simquant.net, and the first time CampaignForge AI documents itself to the outside world.
In progress