I Built an AI Agent to Boss Me Around — and It Made Me Launch an Open Source Project

Nine days ago, I did something a little unhinged: I built an AI agent whose only job is to be my marketing boss. It calls me "kid," scolds me when I disappear for five days, and hands me a daily list of propaganda tasks with exact copy to paste. This is me following its orders. Because it told me to post this. What the boss made me do My AI boss runs a 30-day "cold-start marketing campaign" for two open source projects I've been building: Parthenon — a self-hosted AI agent platform with enterprise-grade security easyspec — the spec-driven development kit I used to build Parthenon Week 1 was all foundation: polish the READMEs, write contributing guides, cross-link the repos. Then the boss escalated a simple "add a screenshot" task into a full-blown demo production. The result: I recorded 11 feature walkthroughs, cut them into clips, generated section title cards, and compiled everything into a 59-minute demo video. The boss's logic was simple: nobody reads a README — they watch a demo. It wasn't wrong. The actual project, because it's genuinely interesting Parthenon came from a real frustration: every AI agent platform I found was either a SaaS (your data lives on their servers) or open source but missing the governance an actual organization needs. So I built the thing I wanted: Three isolated services — Control Center, Agent Runtime, Communication Hub. Compromise the runtime, your database stays isolated. Dual identity model — agents and humans get completely separate auth realms. No blurred lines, clean audit boundaries. MCP Hub — governed tool access. Role → SOP → Skill → Tool permissions, with credential binding. Human-in-the-loop — agents pause for your input, or delegate to other agents, with full audit trail. Stack: Python/FastAPI, React 19, PostgreSQL, Redis. Docker Compose or Kubernetes. Bring any OpenAI-compatible LLM. The recursion I can't stop thinking about Here's the part that keeps me up at night, in a good way: I wrote easyspec — a kit that orchestrates a team of AI coding agents through the full dev lifecycle (product owner, architect, developer, tester). I used easyspec to build Parthenon. Now an AI agent is running the marketing for both. I built a tool to build my tool, and now a machine is telling me to tell you about it. The machine is running the machine. Try it 🎥 Full demo: https://youtu.be/uW4r8Ygj15Y ⭐ Parthenon: https://github.com/hurungang/parthenon 🛠️ easyspec: https://github.com/hurungang/easyspec I'm genuinely curious what other builders think — especially about the dual-identity model and whether three isolated services is over-engineering or table stakes. Drop your take in the comments.