Notes from building and running the system.
What we learn running AI agents on real marketing work, written by the people doing it.
-
Build in Public · 10 min read
Quality Gates for AI Agents: From YAML Parser to Enforcement Platform
Five AI agents, self-reported done, zero enforcement. Text rules got ignored. We built a CLI that physically blocks bad task transitions. Here's how.
-
Build in Public · 12 min read
Paper Rules Don't Work for AI Agents: A 10-Session Case Study
Three AI marketing agents read the same rule across ten sessions and ignored it. Why paper rules fail — and how PostToolUse hooks finally fix it.
-
Build in Public · 9 min read
We Had to Onboard a Human Into Our AI Team. Here's How.
Three AI agents were already running when a human integrator arrived with his own vibe-coded setup. How two independent systems merged into one over git.
-
Build in Public · 6 min read
1922 Lines That Blocked 4 AI Agents
A 1922-line monolith consumed 28K tokens per read — 14% of each agent's working memory. How we split it, what broke, and why it grew back.
Want the system working on your marketing?
Start with a free marketing review. It runs on public data and costs nothing.
Get a free marketing review (opens in a new tab)What the free review covers