He Built for the Lawyer. I Built for the Inventor.

On the night of September 1, 2026, I was filing a provisional patent application with the USPTO, and I kept thinking about Michael.

Michael is the patent attorney who filed my first patent, on the AI behind MuTag, in November 2017. It was granted in December 2018, 13 months later. He has prosecuted hundreds of AI patent applications and is a partner at a New York patent firm. For the last few years he has been building FENIX.AI, an AI agent that drafts patents.

I was filing without him. I knew what that bought me and what it did not.

The filing came out of Ponyo. In the first weeks of the prototype, I had a problem: an AI agent’s output is nondeterministic, and my matching algorithm needed deterministic data in a standard format in my Supabase database. I solved it, and I thought the solution might be patentable. I was not sure.

Then on August 15 I demoed Ponyo at a Cursor meetup. A room of AI native developers, all laughing at “Hey Claude, find me a girlfriend who is this tall and looks like that.” One guy pulled out his laptop and asked someone to film us while I onboarded him. He was not a developer. He uploaded pictures. Then he described the woman he was looking for in great detail: Latina, knows international business, looks like a celebrity he likes from Colombia. The room was amazed. So was I.

That was when I decided to file.

I ran prior art research with two models in parallel, Fable 5 and Kimi K3, overnight. Then I matched the results against what was live and what was in my code on GitHub. The preparation took four or five days of back and forth between me and the AI, reading, correcting, aligning on what the invention actually was. Eventually I doubled down on Fable 5. The specification and the drawings were generated by AI.

I had done this once before. In 2016, I filed the MuTag provisional myself, with PowerPoint and Google. I spent a week or two studying the USPTO tutorials, learning the difference between a trademark and a patent, and between the types of patents. Then seven or eight hours doing the work. No AI.

This time, Claude in Chrome walked me through the Patent Center website. The first filing took four or five hours. The slowest part was not the claims or the drawings. It was the micro entity form: download, type in my details, insert a digital signature, upload, and get every field exact. Everything else was smooth. Two days later I filed the second application. It took about 30 minutes.

In plain words, the two filings cover how an AI agent’s suggestions become trusted records only with a person’s approval, and how an app can use what a person shares with their AI without ever holding the original.

A provisional is a placeholder. The nonprovisional is where a patent lives or dies, with office actions or a straight rejection.

So I emailed Michael.

I had already tested his FENIX MCP. Powerful, but built for a lawyer, not for an inventor. On the call he told me why selling it had been hard. Patent lawyers are a special kind of lawyer. They are technically capable people, and when you show them a software tool, they say they can do it themselves. He had tried for years, through his own network of patent lawyers. In the end, he used FENIX mostly inside his own firm.

He built a product for himself, the lawyer. I built a product for myself, the inventor.

I asked him: instead of selling AI software to law firms, why not build the law firm, powered by AI, and work directly with inventors?

We had worked on a stablecoin idea years ago, so this was not our first call about building something. Within 30 minutes he said, “If this is what you truly want to build, let’s build. I’m in.”

So we are building a patent law firm. An AI native one, for startups. Michael is the attorney. I am not a lawyer. I build the technology and bring the inventor side. The name and the website come in the next story.

The USPTO receives about 600,000 utility patent applications a year and grants roughly 325,000. With AI, the pace of invention is about to change. More people can build, and more people will have something worth protecting. I want to be part of that: ten times more inventors filing to protect their work. I think it is possible. More creativity and more inventions are positive sum for everyone. Founders who patent their work are more confident, and better prepared to raise and launch. I experienced that first hand in 2017, with a patent pending technology.

After the call, I came home and started the market research. Which founders need this. Who to talk to first. How we could be faster, better or cheaper than anyone else.

Within 12 hours, we had a game plan and a website.