Graham McNicoll

When I talk to product teams lately, a strange thing keeps coming up. A large share of their traffic is no longer human. One company found roughly 80 percent of its documentation views were from agents, not people. Think about that if you run a product team. Clicks, page views, sessions, time on page. Every one of those was built to measure a human deciding to do something. When an agent is the one navigating your product, those numbers quietly stop telling you what they used to. This is already happening. A model calls your API instead of a person clicking a button. An agent reads your docs instead of a developer searching them. Your dashboards may not represent what they used to, as the way people interact with your product has changed. Fixing this starts with deciding what success looks like when part of your audience is agentic. That usually means measuring outcomes directly against real users where you still can, and treating agent activity as its own thing instead of folding it into a human metric. You also may need to devalue proxy metrics and focus on the end-goal metrics. We built GrowthBook to measure what a change actually did to your users and measure against end-goal metrics, which is the distinction that starts to matter most here. What are you measuring today that would not survive an honest look once you separate the humans from the agents?