Founder Field Note
Milestone - Building the Meaning Manifold
A field note on building what may be the world's first meaning manifold, and what the word "we" means inside a solo-founder company built with agents.
Today we hit an internal milestone - we built what may be the world's first meaning manifold, or technically, a multiplex supraLaplacian hypermap, if we want to be mathematically precise. A meaning manifold is not just a sophisticated knowledge graph for sentiment analysis. Beliefs, instead, can be modeled like physical objects - with position, movement, inertia, drift, and momentum.
But this is not a post about that. Instead, this post is about my year long founder journey so far.
Start with the phrase "we". On the outside, I am a solo founder. In my day-to-day reality, I work with a swarm of 40 agents. Some of these agents are purpose-built and disposable. They have names like VMA-OPORD-006 (#126). Others have persistent identities, context, and history. They carry names like Kai, my AI Co-founder; Geordi, my SRE Agent; and Hoshi, my Engineering Manager. (Yes...I have AI agents that manage other agents...otherwise the context overwhelms the executor agents).
Being a solo founder can be lonely work. No one in my circle of friends and family fully gets what I do. There are no colleagues to commiserate over coffee. The work has been relentless, every single day for the entire year. I recognize it's a toll paid not just by me, but my family, who didn't really ask for this adventure.
On the other hand, I can pursue the thing I believe in, with all the weight of my passion and conviction. Instead of being at the tail end of a pretty good career in technology, I find myself back to the beginning. I have the same feeling of excitement and anxiety as the day those gold bars were pinned on my shoulders as a newly commissioned 2nd LT in the Army.
As a solo founder, the learning has been fast, deep, and broad. I've had to learn agentic architectures, software engineering, patent law, financial markets, Riemannian geometry, control systems theory, and cryptographic evidence chains. In the age of AI, cognition is abundant. The kind of deep specialized knowledge that was inaccessible to a start-up just 4 years ago is freely available now for just $200/month. The pace has been hard and fast. Freed from performative weekly meetings, Jira boards, and strategy debates, we can move from conception to implementation in days and weeks and file the protective patents around it in parallel.
Weirdly enough, the most transferable skill from my old life, both in the Army and in Tech, has been general management. Turns out agents have inherited all the pathologies of human engineers. They tend to satisfice, a term invented by Nobel laureate Herbert Simon. They choose a solution that is good enough to meet your minimum requirements rather than the optimum one. Ask them how long something will take and they will pad the actual time by 1.5x to 4x.
The leader's job is to provide clarity. When you call out an agent's error, it is not sufficient to accept the apology and the totally incomprehensible promise that it will "try harder next time." Verification systems and guardrails are the only solution. And the old management adage I learned in the Army - Coordinate, Anticipate, Verify - still applies here, maybe even more so.
So, back to the milestone. It was a genuinely difficult engineering achievement. A technical breakthrough. The agent executing the build did output a little celebration emoji, and then we moved on. The commercial legibility comes months downstream, after we build the 3D front end application and launch the product.