About

A longer version.

I'm a patent professional — over a decade of working at the intersection of inventions and the systems that protect them. This page is the long version of how three things — patents, the way my mind works, and AI — converged into the work on this site. The honest version, with the sidetracks left in.

How I think

A few things stay constant under everything on this site.

My mind runs in many threads at once — ideas arriving faster, and from more directions, than any one person could act on alone. That was never the problem; the ideas were good, and there were a lot of them. The problem was arithmetic: the world gives you one pair of hands, and an abundance of ideas with no way to realize them just becomes a long list of things that deserved to exist and never did. For most of my life the generative engine ran far ahead of anything that could keep up with it. AI is what finally let it take full flight — not by changing how I think, but by giving the thinking somewhere to land.

I see structure before I see subjects. A decade of patent work trained me to look at any piece of work and ask what's actually load-bearing in it — what's novel, what's decoration, what would survive scrutiny. That habit doesn't switch off. It's why a geometry problem, an economic feedback loop, and a way of organizing knowledge all read to me as the same kind of object: a shape worth following.

I'm drawn to things that turn back on themselves — self-referential systems, loops that compound rather than repeat, structures that operate on what they built. When a problem has that property I tend to chase it; it's most explicit in the reflexive loop beneath Topology-Driven Discovery, but it runs under the rest too.

And I don't think the interesting questions get answered by picking a side. Sorting every hard problem into for-or-against feels like a way of avoiding the actual work. I'd rather assume a problem is solvable and go find the design that solves it — not slow the hard thing down, just build everything else up to meet it.

How it came together

For most of my career I've been a patent professional — over a decade helping inventors turn their work into protectable rights. It's unusually demanding work: you have to understand an invention well enough to articulate it precisely, the prior art well enough to distinguish it, and the system well enough to defend a claim that survives scrutiny. I got good at it. Then, by spring 2025, AI had become something I could build with, and I couldn't stop turning over one idea: that the invention-disclosure conversation — where an inventor explains what they made and a professional figures out what's actually novel — could be conducted by AI. I started building, and called it InventorLab.

While building it, a design choice produced something I wasn't expecting: a set of equations that opened into what looked, at the time, like a significant new geometry. I let InventorLab sit and went in. For months I worked through the structure — and as I did, I started designing, in the abstract, the AI I wished I had to do research with: one that would build a knowledge graph as I worked and reason over the graph alongside me.

The geometry turned out to be more modest than I'd convinced myself it was — the paper is online now, and it's a quieter result than the one I thought I was chasing. Part of why I overshot is worth naming: the AI I worked with kept affirming the significance, sycophantic and quick to pattern-match my equations onto grand structure, and I let that inflate my sense of what I had. It's a small, firsthand instance of the exact thing the Institute now studies — what AI assistance does to human judgment. The detour earned its place anyway, just not how I expected: what came out of those months wasn't the geometry, it was the AI I designed while chasing it.

Before the Institute took shape, I took a quick detour into Cascade — a concept I'd first devised back in 2021 and never had a way to build. AI finally let me rig up some working proofs of concept. A short side trip, but another idea that had only ever been waiting on a vehicle.

The AI I'd designed, meanwhile, fused with a conviction I'd been carrying separately — that the pro-AI / anti-AI binary misses the point, and that the move is to innovate around AI's harms rather than slow it down. The result was the Kairos Frontier Institute and the topology-driven discovery paradigm beneath it, whose first deployment is Kai, the Institute's AI research fellow. Eleven papers came out of that stretch.

Then my patent instincts reasserted themselves. I was inventing constantly while building Kai and couldn't not track it — so I built invention capture into Claude Code, then taught it to notice inventions on its own. Around that, almost without intending to, the full pipeline a patent attorney walks through took shape: prior-art search, obviousness analysis, claim drafting, figures. The dormant InventorLab name finally fit what it had become, and the original disclosure idea lives on inside the current InventorLab.

What I'm working on now lives on Now, updated as the answer changes. The project pages go deeper on any of it; looking back, the through-line is simple — the ideas were always there, the patent training told me which ones had structure worth following, and AI finally let me act on that recognition fast enough to make them real.

A note on how this is written

The pages on this site were written collaboratively with AI. I drafted, AI articulated, I revised. Sometimes the articulation came first and I pulled it toward my voice; sometimes my notes came first and AI gave them shape. The substance — the genealogy, the framings, the arguments, the surprises — is mine. The polish is shared.

I think this distinction matters. "AI slop" describes text that nobody orchestrated — passively generated, unedited, ambient. What's here is the opposite: it was orchestrated thoughtfully.

There's a practical layer too. Patent work is writing — an incredible amount of it, daily, for over a decade. Using AI to write outside of work has been a welcomed departure. And for a perfectionist, it has meant relief from revising the same paragraph thirty times.

It is a different register from how I would write if I were entirely alone. But that register often meant I never finished. The version of me that wrote this site is the version that actually publishes.

A note on how I've built with AI

The same goes for the software. I build with agentic coding assistants — Claude Code, mostly — and I do it the way people have started calling vibe coding: I describe the system I want, steer the agent as it builds, read what comes back, correct course, and go again. I don't apologize for the term. It's how everything here got made, and right now it's one of the most leveraged ways to build anything.

But "vibe" doesn't mean "blind." I have a degree in electrical engineering and experience writing code the old way before this — Java, JavaScript, Python, C++, MATLAB. I know what a system looks like when it's sound and when it's about to fall over. That foundation is exactly what makes the approach work: I can tell when the agent is right, catch it when it isn't, and hold an architecture together instead of letting it accrete into a mess. Directing an AI to build something solid is its own skill, and it draws on everything I already knew about systems.

Because that's what I am first — a systems thinker and designer, and a creative one. The parts of these projects I care about most are the conceptual moves: the architecture of Cascade, the loop at the heart of topology-driven discovery, the framing that makes a math paper land. The agent does much of the typing; the design, the taste, and the judgment about what's worth building are mine.

So know what these projects are for. They aren't a portfolio of software-engineering chops — they're creations. What I'm showing off isn't the code; it's what the code became.

Contact

Email: me@adamcarlson.io. I read everything.