Something important is happening with FSD v15, and I haven’t seen it framed this way. For years, Tesla used the customer fleet as the debug layer. New builds went to owners first. Interventions, disengagements, and video from millions of supervised miles told Tesla what still broke. Robotaxi was not the proving ground. Owners were. v15 reverses that. Early v15 is already running unsupervised on Tesla’s own Robotaxi fleet. Customer cars are still on supervised v14. The newest stack is being validated internally and on a geofenced commercial fleet before it ever becomes a public FSD update. That is a strategy change. It does not mean v15 is a small, safety-neutral patch. Tesla has described it as a step-change: multiple parallel improvement tracks, a much larger model, and the software they want before they scale unsupervised operation. The point is different. Tesla now thinks it can ship that stack into cars with no safety driver without first farming bugs out to private owners. That matters for unsupervised FSD on privately owned cars. The day Tesla removes the requirement to supervise, it also loses the intervention dataset that built FSD. You cannot keep improving a system the same way if the customer is no longer supposed to grab the wheel. Unsupervised FSD only works long-term if Tesla can keep iterating with internal testing, simulation, and its own fleet. Putting v15 on Robotaxi first is an early sign they believe that loop is starting to work. I still expect the first unsupervised customer use to be limited to the Robotaxi geofences. Nationwide unsupervised FSD on private cars is a later step, not the next one. But the sequence has changed. The customer fleet is no longer the lab. Robotaxi is. Robotaxi getting v15 before owners is an early sign unsupervised FSD is being prepared for the public, not just the Robotaxi fleet.