Sabi is building a baseball cap that turns thoughts into prompts for AI agents. An early wired helmet version hit up to 77% accuracy in the lab, the company says. That error rate would sink a keyboard. An AI agent might live with it. Brain-computer interfaces were built for a harder job. Neuralink, valued at $42 billion, implants chips so people with paralysis can control a computer by thinking. Precision like that has mostly meant hardware inside the skull. Khosla, who led the round, put the scale problem bluntly to Forbes. “You’re not going to do a billion brain surgeries.” A cap trades surgery for a weaker signal. A University of Chicago neurology professor told Forbes that signals filtered through skull and scalp come out lower quality. An agent asks for much less. Sabi’s pitch is that you think of booking a flight and the model fills in the details you left implied. The company also says accuracy climbs as the model learns each wearer. That 77% came from people with more than two weeks of use. I suspect the harder problem starts after the thought is read. A typo stays on screen until you catch it. An agent acting on a misread thought may have already booked the wrong flight. The pitch even has agents moving before you finish a thought, which skips the moment you’d normally notice. If I were designing it, the cap’s planned voice assistant would read the plan back first. A rough signal can start the work, and the cap that asks before it buys is the one people will keep wearing.