Templeton and Merritt clash over "unsupervised" Cybercab
A public argument about Tesla's driverless taxi turned into a fight over one word, with no staffing figures on either side.
Brad Templeton split remote human help into two kinds and said only one of them makes a robotaxi supervised. Standby remote assist is event-driven, meaning a human steps in when something goes wrong. Remote monitoring means someone watches all the time during normal operation. In Templeton's framing, the second is supervision and the first is not.
Sawyer Merritt was not having it. He replied that having humans remotely on standby in case of an emergency does not mean the Cybercab is not operating unsupervised, and accused critics of moving the goalposts.
Having humans remotely on standby in case of an emergency or if something goes wrong doesn't mean the Cybercab isn't operating Unsupervised.
Where the line sits
Templeton then pushed the definition further, saying the location of the human is irrelevant and that what matters is whether anyone is supervising nearly full-time, in the car or over video.
Unsupervised means there is not a human supervising nearly full-time. Does not matter if the human is in the car or via video.
That is the crux. Merritt treats the empty driver's seat as the test. Templeton treats the ratio of humans to vehicles as the test. Both can describe the same deployment and reach opposite answers, which is why the thread never resolved.
Nobody produced a number
Ryan Wang jumped in on Merritt's side and made the sharpest version of the counterargument. He said there is currently no evidence Tesla is performing full-time remote monitoring, that anything close to one human per vehicle would produce staffing costs hard to hide, and that those costs would conflict with running a scalable robotaxi service. He added that a Cybercab without a steering wheel already meets the core meaning of unsupervised, since no human in the vehicle is responsible for driving, and that remote backup is industry standard.
Templeton's answer was an appeal to industry practice rather than to disclosure. He said every rational company following Musk's goals does this, listing Waymo, Zoox, Cruise, Chinese operators and Tesla, and said the burden of proof sits with people claiming Tesla is insane. He also claimed the staffing is similar to what it took to have a safety monitor in the car.
That is the whole exchange. No one in it produced a Tesla staffing figure, a human-to-vehicle ratio, or a company statement about how Cybercab remote support actually works. Templeton infers full-time monitoring from what he thinks any sane operator would do. Wang argues inference is not evidence. Merritt argues the word has a plain meaning and the critics keep changing it.
Tesla has been scaling its driverless fleet in Texas and published a safety page, as covered earlier. What it has not published, at least not in anything cited here, is the one number that would settle this argument.
Remote backup is industry standard, and the lack of a public statement does not mean full-time monitoring is occurring.
Sound like you were at the demo
One email a day. Under three minutes. The demos, the deployments and the part the press release left out.