A postcard from the lab.
Notes from the hill where we're building this thing: what San Francisco has to do with the research, and what the view reminds us of on the long weeks.
The lab sits on a hill in the north part of the city. The window faces the water, the bridge, and a slice of downtown that catches the light twice a day. On the best mornings, a cable car grinds past the front door before anyone has had coffee. It is not a quiet place to work, and we think that is the point.
People ask us why San Francisco. The easy answer is that most of the people building serious AI are still here, and you save an enormous amount of time by walking across town to argue with them in person instead of writing them long emails. The harder answer is about density. Two or three good conversations a week with someone doing something real will move a research roadmap further than a month of your own ideas will. The city still concentrates that, unevenly, but more than anywhere else we have tried.
The lab itself is small on purpose. Four of us, a few more joining this spring, and a coordinator running in the corner of every screen. We meet less than we did in the earliest days, because the interns take a decent share of the busywork now, and we try to spend the recovered time actually thinking. It rarely works out that cleanly. Most weeks we are still debugging handoffs between one intern and another at eleven at night, and the bridge turns orange outside the window while we do it.
What the city reminds us of, and this is less cute than it sounds, is how much of the useful work in a company happens in the seams. Someone in sales knows something that marketing needs by Thursday. Someone in growth notices a change that the dev team should have been told about on Monday. The work that we train our interns on lives in exactly those seams. Building in a place that runs on serendipity made that obvious faster than it would have been anywhere else.
The rest is logistics. We keep the team small, we keep the office above a bakery, and we keep one rule: if a research idea cannot be explained to the person two desks over before lunch, it is not ready. The interns we train are downstream of that rule. They have to be specific enough to be useful in a real week of work, and general enough to keep being useful when the week changes.
This note is shorter than most of what we write. Call it a postcard. The serious writing lives in the engineering and research posts that show up here on a less predictable schedule. But every so often it helps to step back and say where the work is happening, and to remind the people reading that there is a window, a bridge, and a cable car behind every line of what we publish.