Daybook, 15 July 2026

Two conversations made the day. The first was with an old man driving for Uber — he'd retired from a manufacturing job, and now he's driving around eight hours a day, and the moment he hits his daily goal, he stops and takes his time back. And then a funny thing happened: instead of going to my founder's place to work from there, I went to the old office by mistake. I realized, re-booked a cab — and the same driver showed up again. So we ended up with two trips and about an hour of conversation.

It went everywhere, the way the good ones do — Thanksgiving, and how he'd make a good ham and steak (he'd take a cold barbecue over the kitchen any day); where we're each from; the gold rush, and the old townships, and he was quite concerned about the government taking over some of the houses to build new infrastructure. And his family's story — his great-grandfathers worked in the Philippine military, then moved to New York, or Philadelphia, then Kansas, and from Kansas to San Francisco. And we got onto a sheep-detective story, of all things — I didn't even know there was such a thing as a winter lamb; I'd always thought lambs were only born in spring. Food, films, the city and how it's transforming, family. He mentioned he loves talking to people because sharing the journey and the experience is what relaxes him. One of the memorable Uber drivers.

The second conversation was a planned catch-up with my founder — we went for about a two-hour walk. And it reminded me that over the last seven years, I've somehow kept ending up around really good, kind, formidable people. And the world's so small: I often follow Professor Subbarao Kambhampati's work on LLM planning and reasoning — and it turns out our co-founder was his first, or one of his very early, students. He shared his journey — his thesis getting published at AAAI, and his path into entrepreneurship — a lot of good references on what to do and what not to do. Both conversations kind of landed on the same point for me: there's a purpose, and an importance, in how we actually interact with people, and in letting the natural conversation emerge.

And then a sharp exchange that clarified where AI actually belongs. I'd written about this on one of my LinkedIn posts — where you'd genuinely need specialized models on tabular data. For a network, which is very well designed by humans, I don't think you really need a model to do anomaly detection on a single element, because the component itself is quite capable these days. Where things get interesting is where things change — user behavior, loads shifting. And this came up when I was discussing with my co-founder. He put it bluntly: when I can control anything with a couple of lines of low-level code, why do we need AI or data science there? It's all deterministic. Where it's actually needed is where complexity and uncertainty emerge — different environments, locations, loads. When you put everything together as a topology or a network — n elements interacting, the network topology, the clients and people using it, the different loads — that's where the uncertainty comes in. It's the holism, right — the system is greater than the sum of its parts. One element, one switch, one AP can be perfectly determined; the whole thing can be well-designed; but when it's all interacting, that's where the real problem spaces appear, and that's where AI or data science is actually needed to govern it. And the flip side is a discipline I want to keep: if the design is deterministic, don't put an agent there — just ship clean, deterministic automation. Reserve the AI for the emergent part.

On the personal front: I'm still deep in tabular foundation models — it's kind of kicking in now. And I'm keeping the habit of recording myself, and later visualizing it, just to keep the mind map consolidating over and over again.

Settled · 15 Jul 2026