Vercel's fastest-growing user
isn't human.
Agents now account for ~30% of deploys, most of them Claude Code. As a human-adoption framework goes blind to that user, the operating rhythm and the landing measurement have to be rebuilt for the agent-as-user era. This is what that looks like.
Who's actually deploying
Peak vs. last period
Did it land? It depends on the lens.
The same launch, scored two ways. Drag the weights to argue for any metric you like. Under the human-adoption lens, an agent-driven launch stays invisible no matter how you weight it, because the user who adopted it isn't human. The newest signal, composite acceptance, came from pushback by Kira M. Allen in Lenny's community: a chain of agents can pass every per-step check and still miss the mandate, so the accountable human's keep-or-return verdict on the final result carries the heaviest default weight.
Landed. Agents integrated it in 42% of sessions within an hour, reuse is growing, no retry spike. Owner verdict: keep.