The Next Billion Form Fills Won't Be Human

One week in. The ledger has real history in it, real notices have gone out, real certificates verify at public URLs, and the test suite clicks through the whole platform on every release so I find the breakage before you do. That’s where we are. Here’s where this goes.

The quiet shift underneath everything

AI agents are beginning to fill forms, book appointments, and hand over contact details on people’s behalf. Not a demo — a browsing pattern that’s already measurable, and growing.

Now look at how incumbent consent tooling works: it records a session. Mouse movement, keystroke timing, a replay a human can watch. All of it assumes a person at a browser. An agent produces none of those signals — so its session either looks like fraud or proves nothing. The evidence model and the traffic are heading in opposite directions.

A protocol with published test vectors is exactly what an agent can implement. A screenshot of a human hesitating over a checkbox is exactly what it can’t. Evidence built to be recomputed survives the shift; evidence built to be watched doesn’t. That’s the bet, and everything we ship leans into it.

We’ve built our own side of that line carefully. PRYVC has an agent surface — a real one, with tools an agent can call to read, update, and revoke on a person’s behalf, with tokens the person mints and can kill. But there is deliberately no tool that grants consent. Not for agents, not for the extension, not for anyone. Agreeing still requires a person at a consent screen. If that costs us a checkbox on someone’s agent-readiness scorecard, fine. Some checkboxes deserve to fail.

Where the next quarters go

Deeper into regulated lead flows, where certified consent stops being nice-to-have and becomes the thing your carrier asks for by name. More reach for individuals — the extension already fills any form on the web and leaves a receipt; the removal machinery gets more thorough every week. And more of the protocol in more hands, because every independent implementation makes the standard harder to ignore and the evidence harder to dismiss.

The category consolidated into a single custodian this year. I think the answer to consolidation isn’t a second custodian — it’s a protocol nobody has to trust. One week in, that’s still the whole plan.

Building in this space, or holding a lead flow you can’t prove? I’ve been thinking about this problem for a long time. Let’s talk — george@pryvc.com.