The company is becoming legible to machines
Notes from an AI-native company · Letter 1 · Sent by email July 27, 2026
You’re getting this because we’ve crossed paths, maybe when I pitched you BotBot, maybe just talking shop, or because I respect your work and want you reading along. The reason for the letter: a genuinely new kind of company is emerging, a few people plus a team of AI agents running the whole company from one shared, written-down place. I’m running one. This is a weekly, front-row seat to what it’s actually like from the inside, and what best practices we’re seeing emerge in this new field. If it’s not a letter you want, say so and I’ll take you off the list, no hard feelings.
From the inside it works like this: the company keeps one written-down version of itself, the strategy, the voice, the standards, the decisions with their reasons, and people and agents work from that same copy. Last week we launched a weekly industry digest with a filmed conversation alongside it, and most of the work underneath, the research sweep, the edition itself, the video’s chapters and captions, the distribution across three platforms, came from agents reading the same files a new hire would read. My week was mostly deciding what’s true and reviewing what came back. I published a longer piece this morning about what living this way is actually like ("Our company runs out of a folder"). The short version: once agents can read a document and act on it, every document is source code, the written company is the codebase, and if something isn’t written there, the company doesn’t know it.
Out in the field, the agent coworker arrived from three directions last week: Dorsey’s Buzz made agents members of the room, Andrew Ng’s OpenWorker shipped one that delivers finished work, and YC’s new request-for-startups asked for the multiplayer version. YC has now asked for this category twice by name in two application cycles. What jumps out to me is that everyone is building the room, while the thing the agents in the room will work from, the company’s own picture of itself, is still unclaimed. And a March essay from Sequoia is back in circulation arguing the next trillion-dollar company sells work rather than software, which I mostly buy, except the essay never gets to quality: when what you sell is the work itself, your margin is whatever fraction of the work comes back right the first time. Raising that fraction comes from the same place I described above, a written definition of good work that every agent reads before it starts and every piece of output gets checked against. That definition becomes the firm’s production asset, and almost nobody is treating it as one yet.
A friend working on an AI-transformation project told me the definition he’d landed on: a company is AI-native when, without AI, it can’t survive. He called it scary. I think it’s much too small. By that bar you could put AI on one critical path and claim the label. AI-native means built from the ground up on agent workflows, which means almost no mature company will genuinely qualify for years, and that gap is the opportunity: a five-person team running this way can outmaneuver a company a thousand times its size, while the incumbent is set up wrong in ways that are hard to unwind. Meanwhile the scary definition quietly became the floor: a company that could still operate without AI is already non-competitive, at least in California in 2026.
So the claim I’d stake this week is that companies making themselves legible early are compounding already. Every week we run this way, the advantage over running out of people’s heads gets bigger, and the effort of keeping it gets smaller. Internally we’ve started calling this work engineering the company itself, and I increasingly think that’s the job.
And what excites us most is the building itself. Every week running this way we find something the old way couldn’t do, and the finding is speeding up, because every improvement gets written into the same place everything else runs from, so each one makes the next one cheaper. We’re still early on that curve. When people see it up close, they tend to say the same thing: this is how companies are going to work.
If you’re seeing any of this differently, would love to hear your thoughts!
P.S. The longer versions of this thinking, from the last ten days:
- The better tax collector on what Buzz gets right and where it stops
- Prompting was middle management on what steering agents actually feels like
- The first edition of my Friday digest, This week in AI-native companies
- A filmed conversation alongside it, this week with Rahul Singh on memory versus knowledge
- The new chaos of being able to build anything, on the need to build velocity, not product