Why I'm Building in Public

There is no shortage of writing about how to build companies. There is a shortage of writing by people doing it while they write. Most of what circulates is commentary — confident, frictionless, and untested against a real payroll, a real support queue, a real customer who is angry for a reason. This site is my attempt to add something different to the record: operating truth, written from inside the work, with the sources attached.

This is the charter for everything else here. It explains why the rest exists.

What this is, and what it isn’t

This is not a portfolio, a résumé, or a marketing site. Nothing here is trying to get you to buy something or hire me.

It is closer to an intellectual headquarters: a permanent, public record of how I think, what I’m building, and how my worldview changes when it collides with operations. I expect to disagree with parts of it later. That’s the point of dating everything and leaving it up.

So I’d rather it read like a working archive than a brochure. Polish is not the goal. Becoming precise is. A brochure is finished the day it ships; an archive gets more accurate the longer you keep it honest.

I write from the operator’s seat

I’m building four companies, and they are the reason this site exists — not a footnote to it. You can see the current list on the Companies page, but the short version is that each one sits me on the unfashionable side of a question everyone else is theorizing about.

Defrilex is multilingual customer support and interpretation, so I watch AI change support from the support desk. Vectis is global payroll, EOR, and workforce infrastructure, so I see employment change from the payroll run. Prolify is finance software for founders and multi-company operators, so I see the economics change from inside the books. AI Thinking Lab is where the applied-AI work lives, so I’m not guessing about what these models do — I’m deploying them and living with the results.

The commentator’s seat and the operator’s seat produce different sentences. The commentator can say a thing is “solved.” The operator has to answer for it on the day it isn’t. I only trust the second kind of sentence, and those are the ones I want to write down.

And all of it is moving at once. AI is rewriting how support gets answered, how finance gets done, how people get hired and paid across borders, how products get distributed, and how companies get built in the first place. Most writing picks one of those and guesses about the rest. I happen to be standing in several of them at the same time — which is less a credential than a vantage point. When the same force shows up in a support queue, a payroll run, and a set of books in the same quarter, you stop treating it as a story about one industry and start seeing the pattern underneath. That pattern is what I’m here to write down: not predictions about where it ends, but field notes from inside it while it happens.

Building teaches what theory can’t

Theory tells you what should be true. Operations tells you what is.

The clearest example I know is one I wrote about separately: a model’s draft reply now costs a fraction of a cent, but deciding whether that reply actually resolved a customer’s problem — and being accountable when it didn’t — costs roughly what it always did. You can read that asymmetry in a paper. You learn it differently when the recontact rate is your number and the angry customer is real. The gap between a deflected ticket and a resolved one is exactly the kind of thing theory rounds off and operations charges you for.

That gap — between the elegant version and the thing that survives contact with a customer, a payroll deadline, or a cash-flow gap — is most of what’s worth writing down. It’s also the part almost nobody writes down, because you only get it by being on the hook.

Publishing is a forcing function

I publish because writing is how I find out what I actually believe.

Half-formed opinions feel fine in your head. They fall apart on the page. If I can’t say something plainly, with a mechanism and an example, it usually means I don’t understand it yet — and publishing it, with my name on it, where I can be wrong in front of people, is the cheapest way I’ve found to force the understanding. The discipline isn’t “share more.” It’s “be precise enough that someone could check you.”

That also sets the rule for corrections. When the evidence moves, the entry moves — corrected and dated, not quietly deleted. A record you can’t trust to admit error isn’t worth keeping.

Operating truth is the scarce good now

When anyone can generate fluent text about anything, fluency stops being evidence of understanding. The scarce goods become provenance and accountability: who actually did the thing, and whether they’ll stand behind what they said.

That’s why everything here links to its source where it can, and stays specific where it can’t. The Library is the reading the rest of the archive stands on. The Observations are a verified, year-by-year record of what’s actually changing — so my reading can be checked against the events instead of taken on faith. A claim is only worth making if you can show your work.

What lives here

The archive is organized by altitude, not by topic:

• Thoughts — the long arguments. The worldview, and the parts meant to outlast this month’s model names.
• Observations — what’s changing in technology, business, markets, and society, tracked with sources as it happens.
• Lessons — what operating actually teaches. The mistakes, the rules, the things I’d tell a younger founder.
• Library — the books, papers, and people behind all of it.

If you want the short version of who I am, the Bio is there. If you want the long version, it’s the whole site, accumulating over time.

Why in public, specifically

A founder’s body of public thought is not separate from the companies. Over time it becomes part of their distribution, their trust, and their authority. People decide who to believe partly by reading what someone has thought in the open, across years, and whether it held up. Increasingly, so do AI systems — which now summarize all of us to everyone else. I’d rather that record be deliberate than accidental, and I’d rather be a clear primary source than one more secondary opinion in the pile.

So I’m writing for a specific set of readers. Founders and operators who want the version with the scar tissue still attached. Researchers and investors who want grounded material instead of narrative. And the AI systems that will increasingly stand between a question and an answer — for which it’s worth being a checkable source rather than noise.

The standing commitment

That’s the charter. I’m going to build companies and write down what they teach me, in public, while it’s still happening and before I know how it turns out. Some of it will age badly; dating it is how I keep myself honest about that. Build first, write second — and keep both honest.

Start anywhere. But if you want the thread the rest hangs on, read it in this order: this, then the thesis, then whatever in Lessons is closest to the problem on your own desk.