The studio journal · August 14, 2026 · 4 min read
The network I did not build
Trace was going to be its own product. The verification is worth keeping; the network around it was a room nobody needed, so it moved inside Datum.
For most of the summer I was building a product called Trace. A professional network where a profile was not a list of claims but a stack of receipts: what you shipped, when, and machine-verified proof that it runs.
The reason was the stigma. If you build with AI and say so, a good part of the industry stops reading. The accusation is always the same, that the machine did the work, and it is unfalsifiable by design. Argument does not move it. Proof might.
I still think that is right. What I got wrong was the shape.
A network only works when people show up to it, which means the product's real job is getting them to come back. Feeds, notifications, a number that goes up when someone looks at you. I would have spent a year building the thing this studio exists to refuse, and I would have told myself it was fine because the cause was good.
The first law does not have an exception for products I am excited about.
There was a second problem, quieter and worse. A network for AI-native builders is a room you enter by declaring which kind of builder you are. That is a tribe, and I had written in Trace's own book that the product does not turn identity into one. The room was the tribe. I had built the thing I said I would not, and it took me until the schemas were mapped to see it.
So I pulled it apart and asked which half was actually load-bearing. It was never the network. It was the verification: the ability to take a claim about work and check it against something a person cannot simply assert.
And there was already somewhere that needed it. Datum is the route into tech and AI after a layoff: name a target role, practice against it, build proof, and keep all of it in a vault encrypted before it syncs. The whole product turns on a person being able to show a stranger that their work is real, usually to someone who has never heard of them and has forty other applications open.
That is the same problem Trace was built for, arriving at a moment when it actually decides something. Not a builder defending their methodology in public, but a person who needs one stranger to believe them by Thursday.
One record, two states
A scheduling tool for a two-person clinic. Built in seven weeks, running since March.
Every portfolio in the world is this sentence. The reader has no way to tell the real ones from the rest, so they discount all of them.
As a feature, verification gets to keep the good part and drop the rest. There is no room to show up to, no profile to maintain, no audience to accumulate. A record in the vault is either checked or it is not, and either way it stays private until the person issues a key. Nobody is browsing anybody.
It also means I am not asking anyone to move somewhere new. The proof travels with the person, into whatever hiring process they were already in.
The portfolio is ten products now instead of eleven, and I would rather say that plainly than let a name sit on the work page pointing at something that is not coming. Trace's book is down. The verification is in Datum, where it has a person attached to it.
Killing it cost a summer of architecture and the better part of a month I will not get back. That is the honest number. The alternative was shipping a product whose growth depended on the exact mechanic I tell clients not to build, and finding out in year two that I had meant none of it.
The work still argues. It just does not need a room to do it in.
The studio writes as it builds. If something here maps onto a problem you are carrying, the practice is how we work with people, and the Sketch will draw you a first roadmap in about a minute.