Developers With End-to-End AI Agent Website Builds - Evidence

Halfway through a research project last week, my AI research stack handed me a detailed, confident account of a Webflow MCP release that doesn't exist. Version number, ship date, specific error behavior. All fabricated. A second verification pass with a different tool caught it, plus three more errors just like it. That was supposed to be a side note. It became half the story. The question I was actually chasing: can an AI agent build a complete small website yet, end to end, from a structured brief? I evaluated ten platforms (Webflow, Payload, Framer, Sanity, and the rest) across three different integration pathways. The short version: the surfaces are real and expanding fast, two platforms stand apart for opposite reasons, and there is no public evidence anywhere of a fully agent-executed site build. Nobody has published proof. As of mid-July, the demos outrun the record. I published the full analysis as an interactive research artifact instead of a blog post: coverage matrix, platform profiles, a Webflow vs. Payload head-to-head, and test protocols anyone can run. Plus the four fabrications my research tools produced along the way, and how verification caught them. https://lnkd.in/e57B2QKD If you've run one of these builds end to end, I want to hear about it. The evidence gap is the finding.