I rebuilt my product's home page by handing the job to my own AI council

Today I rebuilt the home page of the product you're reading this on — and I did most of the strategy work by handing the job to a council of AI models my own system manages. Here's exactly what happened, because "build in public" only means something if I show the receipts.

The problem: my own marketing was the weakest part of the product

AI SuperHub is a self-hosted, autonomous AI workforce — 30 to 150 agents that publish content, run SEO, pitch prospects, and fix their own mistakes, all on a single consumer laptop. It's the real system I use to run a small network of websites. But the marketing site still led with bland, generic copy: "autonomous agent orchestration platform." True, but forgettable. It buried the one thing that actually makes the system different.

That difference is simple: it checks its own work. After every blog post it publishes, the system fetches its own live page from the internet — exactly like a reader would — and if the post isn't actually up, it emails me within a minute. Most "AI publishers" fire and forget. Mine verifies. That deserved to be the headline, not a footnote.

I asked my own AI council to write the pitch

Instead of agonizing over hero copy myself, I used a feature I'd built into the system: a delegation layer that routes a task to the best available model and remembers which model wins which kind of work. I handed it the positioning brief and asked for headline options, the single sharpest differentiator, a ranked "who is this for," and the most conversion-critical sections to put first.

The router picked a fast, capable cloud model for the copywriting task, ran it, and saved the result back into the system's memory so the controller agent retains the lesson. The output was genuinely good — three strong headline directions, a clean audience ranking, and a clear above-the-fold order. I rated it five out of five, which re-ranked that model to the top for future growth-and-copy tasks. The system literally got a little smarter about who to ask next time.

A note on principle, because it matters: those cloud models help me build the product faster, but they are never wired into the always-on runtime. The product itself runs on local models on your hardware — no per-token bills, no data leaving your machine. Cloud APIs are a power tool in the workshop, not a dependency in the engine.

What actually shipped

Working from that brief, I rebuilt the page around the things the system provably does:

• A new headline that leads with the real story instead of category jargon.
• A dedicated section explaining the "it checks its own work" verify loop — write, deploy, re-fetch the live URL, confirm or alert.
• Seven capability pillars, each one a thing the system does in production every day, not a roadmap wish.
• A plain "who it's for" section — self-hosters and homelabbers first, then solo founders, freelance SEOs, niche-site builders, and anyone who needs on-prem, data-sovereign AI.
• A comparison table that positions honestly against the one-chatbot, cloud-metered tools without naming names.

The whole thing — rebuild, build, deploy, and verify the live page — took an afternoon, not a sprint. The deploy itself went through the same careful path the system uses for its own content: build the static bundle, push it, and confirm the live URL responds before calling it done.

Why I'm telling you the boring details

Because the honesty is the product. Anyone can mock up a slick AI dashboard for a pitch deck. What's hard to fake is "this is the actual system that runs my actual business, and here's it improving its own marketing today." The moment a tool like this feels like vaporware, the whole premise dies. So I'd rather show you a real, unglamorous afternoon of work than a glossy promise.

The product isn't finished — and that's the point. I'm building it in the open, and the people on the waitlist get the first say in what it does for them first. If a self-hosted AI workforce that runs on your hardware and checks its own work sounds like something you'd use, that's the whole invitation.