Home

How I build

For most of my career I built big things — a game studio Electronic Arts acquired, the payments company Spreedly, and design and product organizations at Observable, Tray, and CloudBees. That work taught me how much a capable team can put behind a product, and how far a well-run product can reach.

The tools I make now start from the other end of that scale. One job, done well. A surface small enough to hold in your head. Nothing that needs a server, an account, or my permission to keep working.

The format is the interface

Field Bureau's markdown suite reads and writes plain Markdown you own — no database, no proprietary format, no import step. SR-7 records a conversation, transcribes it on device, and saves it as Markdown with the audio sitting right next to it. TR-2 keeps your tasks as Markdown checkboxes in a folder. ML-42 reads and edits the whole library. The connective tissue between them is the file, not an integration.

The same files you edit are the files your AI reads and writes. There's no copy-paste, no export, no second source of truth — and because the agent works the files directly, you can see exactly what it changed in the git diff. If any of these apps disappeared tomorrow, your work would still be there, in formats that outlive the tool.

The intelligence runs on your machine

What changed in the last two years is that the intelligence got small enough to run locally. SR-7 transcribes and summarizes using the models already on your phone — Apple's Speech framework and on-device foundation models. Nothing uploads. No accounts, no telemetry, no subscription, no server to rent. When the computation never leaves the device, privacy stops being a promise and becomes a property of how the thing is built.

That refusal is deliberate. A subscription is rent on software you don't control; charging once is the only model that's honest about "nothing leaves your machine." Owning your data and owning the app turn out to be the same promise.

Small teams can build more than they used to

A small pod ships work now that would have needed a whole department a few years ago. Working alongside AI, a handful of people can hold the entire stack — native macOS and iOS, on-device ML, sync, design — and move at a pace that used to take a much larger team. The shift is economic. When execution gets this cheap, the scarce thing isn't the ability to make software; it's the judgment to decide what's worth making.

So that's where the attention goes now. The bet is that the tools worth keeping are the ones you understand, own, and can run without asking anyone's permission. I spent twenty years building things that had to scale. I'm spending the next stretch building things that last.

You can see what I'm building at fieldbw.com, the longer thinking behind it at notsocommonthoughts.com, and how the same judgment shapes the way I run teams in how I lead.