Why this exists.

Most “built with AI” projects skip the strategic work. They demo a thing and never test whether the positioning is honest. We did the opposite — started with moat critique, found that three of our four supposed differentiators were already commodity at the top of the market (Jasper literally has a feature called Brand Voice & Memory), and rewrote everything from there.

What’s left after that critique is the wedge: vertical depth in things horizontal AI tools won’t bother to build. Per-executive ghostwriting profiles. External approvers without seats. Press releases and exec quotes, not blog posts and ad copy. That’s the whole product.

A FastAPI backend, a Next.js frontend, four prompts to Anthropic Claude, and a memory system that learns from your edits. Deployed to AWS for under $10/month total infrastructure cost. The full architecture is documented openly — including the parts that are deliberately unfinished.

The four agents are technically chained prompts, not autonomous agents. We were honest about this in the lessons-learned deck. Real multi-agent (with feedback loops, parallel evaluation, structured handoffs) is on the roadmap. What you can do today: ghostwrite for executives, route through approvers, ship to channels.

It’s not a startup. There’s no sales team, no SOC 2 report, no procurement specialist on staff. We’re a showcase project — a real one, with real users welcome to try it, but you should know what you’re signing up for. If you need enterprise controls, use Writer.com. If you need horizontal marketing AI, use Jasper. We’re neither of those.

We’re also not pretending to have solved the variant body parsing bug from issue #4 in our own critique deck. The fix is straightforward — switch the Editor and Writer to structured tool outputs instead of regex-parsing prose — but we haven’t shipped it yet. There’s a candor we’re trying to preserve here.

The build process is captured as a 6-phase playbook, documented in full on this site: strategy audit → architecture critique → foundation fixes → UX polish → workflow cohesion → methodology consolidation. Each phase makes the next more valuable. Doing them out of order wastes work.

If you’re a builder considering how to use AI on a real codebase, the playbook is the most useful artifact we created. Read the playbook →

One person, with Claude as a senior thinking partner. The architecture decisions, moat critique, code review, UX direction, and deployment were all collaborative work. The product itself is less interesting than the process; the process is the deliverable.