The Missing Moat In AI: Your Eval Data

InnovationAIThe Missing Moat In AI: Your Eval DataByLutz Finger,Contributor.Forbes contributors publish independent expert analyses and insights. AI leader & Cornell faculty; serial entrepreneur; ex-Google/LinkedInFollow AuthorMay 26, 2026, 12:00pm EDT--:-- / --:--This voice experience is generated by AI. Learn more.This voice experience is generated by AI. Learn more.AI chats don’t just generate answers. They generate eval data. The company that harvests it best wins.gettyWhat wins the AI race? UX? Access to users? The foundation model? The data? All of them matter, and all of them are being fought over. But the real value may sit in the one thing OpenAI, Anthropic, and Gemini have not yet made central: your own eval data. Sure, there is the occasional thumbs-up button. But who is really using that signal? The real answer to the AI race may be hiding there.By now it is clear that the AI race will be decided by the race to build AI agents. They are the future of software. We no longer click through menus, we delegate, and that shift is already rewriting every workflow in the enterprise. My own AI agents schedule meetings, cut video, draft replies, and watch my inbox, and their number grows daily. What makes an AI agent platform successful? We see two main discussions: the foundation model and the user access. Both matter. Neither is sufficient on its own. We need a way to work with eval data. User Access: Google Shows OffThis year’s I/O was all about user access. Google poured AI onto every surface it owns. There was no single story except this: every surface has a UX, and Google now wants AI inside each one. It might not sound visionary but it makes sense, since Google owns the surfaces: Search, Android, Chrome, Gmail, Docs, YouTube, Maps, glasses, and the Gemini app. In AI workflow language, the equivalent is not the surface but the harness: the layer that lets an agent act through tools, permissions, memory, and guardrails. Different word, same strategic poin...