Daybook, 18 July 2026

A rest day, mostly logistics. Shifting from Mountain View to San Jose for a week or so, until I find something long-term — Bay Area house-hunting is quite tiring, honestly. And when you're searching, people ask you a lot of questions — are you this, are you that, do you eat this or that. Fine, they want housemates who align, that's fair. I'm just not really liking it, because in New York it wasn't this way. And you know what — I keep getting startled by how much people sublet here. People renting out the living room, the hall. That was quite new to me.

One chapter left before I finish The Meaning of Your Life. It's been a really good read — nearly every chapter validates some practice I'm already doing, or gives me the confidence to try an idea. I always used to ask myself, why am I like this? Why can't I just conform, why don't I settle the way others in the same segment seem to? And the book kind of says — no, that's fine, you just keep exploring. Because meaning is coherence, purpose, and significance — and holding some of those unanswerable questions is a good thing. It also gave me a term for something I already have: Aristotelian, virtuous friendship. At this point I've got three or four people — juniors, seniors, doesn't matter the age — with whom I can have really open, trustworthy conversations about any part of life: relationships, personal care, finance, family, the loneliness, the harder stretches. And with my mom too — she's able to hold that kind of conversation now, where I can openly share what I'm doing, what I'm thinking, my life plans. She used to always ask me about the marriage thing; now she kind of understands, and I'm really happy about that.

The synthesis I've stopped resisting. Coming off the neuro-symbolic and antifragility deep-dive, I went looking — is there an avenue that actually supports this kind of view? And I found the FAST — Foundations of Agentic Systems Theory workshop, whose whole premise is that the community has built agentic systems way faster than it understands them. I've got around 35 days, so maybe I'll try to turn this into a good workshop paper. And while I was exploring, I sort of bumped into this book — Control Systems: A Historical and Philosophical Perspective — along with some philosophy-of-data-science reading. These philosophical layers aren't just decoration; they give you a sense of what you should be ensuring when you research, or design, or read, and how problems actually evolve. And that's how a shape fell out for me — a dual, infinity loop: when we get new knowledge, we infuse it into existing problems; and when knowledge enters reality through problem-solving, that creates a new set of problems, which we try to solve again. A loop — kind of an infinity loop. And brainstorming it aloud, it related to David Deutsch's The Beginning of Infinity — like, hey, this is what he kind of means. So: research, engineering, philosophy — three points of view on the same AI / data-science work. My co-founder actually put words to it before I did, appreciating the systems thinking. And I realized that somehow, over the years, I've groomed myself into a person who sits at that intersection — I'm only now choosing to adopt that view.

Why I'm not a “method-corpus” person — and why it feels like freedom. If you go and talk to a lot of people in AI and data science, there's often no conversation to be had — not at the business level, and not even at the science or methodological level. They'll just throw out method names — oh, just use XGBoost, just use this tool or SDK — or they'll chase the fad of the week: graph engineering, loop engineering, this-engineering, that-engineering. But the way I see it, in real problem-solving AI is just another element — the way a PID controller or a control valve is an element in an engineering system, now playing a similar but more advanced role. The real edge isn't naming what already exists; it's identifying the unknown unknowns. Self-organization was already there in complexity science and in the world long before anyone rebranded it; cognitive science already gave us the architectures we now call agentic; control theory already gives us a way to model subsystem-level inference. Even finite state machines — nothing new, but genuinely useful. In my previous work, the playbook I was building was really just that: I mined the customer-engineer interactions and represented them as a finite-state graph, so that an agent could look at its current state and know how it should proceed. A playbook, but as a finite-state graph. The concept already existed; the real work was in applying it well. So reuse the concept — don't rename it just for content. New vocabulary should be reserved for genuinely new knowledge, not another “X is all you need.” And building agentic systems, when you actually think from first principles, is a craft — not a matter of picking the right buzzword. Honestly, seeing it this way feels liberating — I don't feel that FOMO about everything being written under a hundred different names, because I let myself arrive at it by deeply questioning and conversing.

Settled · 18 Jul 2026