Judged From Outside, I Was Always Late

Three times this year I wrote up an idea for a web3 business and threw it away. One was a record of gifts that could not be edited later, one a trust layer between AI agents and the things they pay for, one an AI product for tokenised assets. The reason was the same each time: I was judging from outside, and by the time I had finished judging, a large company was already there. Last month I tried the other way. I spent a weekend at ETHGlobal Tokyo, my first web3 hackathon, built something that worked, and submitted it. The most useful thing that came back was not the product. It was the ability to read the field quickly, which three rounds of outside judgment had never given me.

Preparation went wrong in useful places

I write code every day for my studio, mostly web services on Cloudflare. I have tried writing smart contracts, but not enough to call it experience. So I spent the week before on the parts I could test in advance. I got a payment protocol for AI agents to complete a round trip, and I got a proof-of-human check, World ID, working on a real phone, after eighty minutes lost to a server that returned an unexplained error page whenever a request arrived without a browser signature.

The step I could not finish was registering a name on the Ethereum Name Service, the network's naming system, in its new second version. After more than an hour the signing request never reached my wallet, and I stopped with nothing recorded on the chain. Four days later, at the venue in Toranomon, the same registration went through, along with the delegation I needed on top of it. Most of what I had prepared turned out to matter less than the list of things that had failed and why.

The topic came from a conversation

I had brought a list of ideas. I did not use it. The topic came from a conversation with a fashion model about body scans: data captured on a shoot, and later at risk of being reused to generate advertising the model never agreed to.

What I built is a consent ledger that a generation pipeline has to query before it produces anything. The agency handles the everyday decisions. The person in the scans holds one power, the ability to take back what they delegated, and only that action asks them to prove they are a person. Every query gets one of four answers, each with a reason attached: allow, deny, revoked, or ask, which means nothing gets generated until a person decides. The permissions live on the chain, in the name I had failed to register four days earlier, and the service reads them before it answers.

Most of the decisions were cuts

The design decisions I am happiest with were things I took out. If the chain cannot be read, the ledger answers ask rather than allow. A system that is meant to protect someone should stop when it cannot see.

The payment protocol I had prepared was working, and I unwired it before submitting. It could not route each payment to a different person, and paying the model directly would have meant handing them a key to manage, which contradicted the reason the agency handled things in the first place. Forty-seven commits went into the submission; the most useful of them removed things. The submission is on the ETHGlobal showcase.

What came back

The week after, I kept the parts. A payment protocol, a naming system and a proof-of-human check are worth knowing while they are still thin, before anyone needs them.

The return I did not plan for showed up in my reading. Articles about blockchains that I used to read as narratives I now read with two questions: who is actually paying, and who holds the capital. I could not have learned those questions from outside, because from outside every announcement looks like demand. The other thing I could not have got from outside was the temperature of the people doing the work. Between builds I talked with people at the neighbouring tables who ship in this field every week, and the field stopped being a list of announcements and became people moving at a speed I could now feel. Three ideas judged from a distance had each been late. One weekend inside made me faster.