What's this about?
The former Twitter CEO, now running Parallel Web Systems, argues that agents will query the web a thousand times more than humans ever have, and that infrastructure built around human clicks is wrong for them. Parallel deliberately treats human click data as a bug rather than signal.
What exactly is wrong with click data for agents?
Search ranking is built on what humans clicked — which encodes human impatience, headline appeal and the fact that people read the first result. An agent does not skim; it can read fifty pages and synthesise. Optimising its retrieval for what a distracted human would have clicked actively degrades the result.
What's the community saying?
This is an interview without a public comment thread. It connects to the WebMCP item below: both are bets that the web needs a machine-facing interface rather than a scraped human-facing one, approached from opposite ends — one rebuilding retrieval, the other standardising how sites expose tools.