This subjectively feels extremely close in early 2026. I just road-tripped about 2500 miles in a Tesla with the latest FSD version (14.2.2.5) and, ex post, would've been completely fine had I read a book the entire time I was in the driver's seat. Of course that doesn't make it safe to do it ex ante -- 2500 miles with zero interventions is very weak evidence. But, again, it at least feels close. Here's my trip report: https://agifriday.substack.com/p/tarsnakes

Big update for me! This is from poring over the latest numbers plus a lot of experimentation. Oh yeah, and after renting the aforementioned Tesla for a road trip I was completely ruined for other cars. So I went and bought a 2-year-old Tesla. I'm now at the point of endorsing the effective thwarting of the in-cabin monitoring and blatantly ignoring the road, under very specific conditions:

1. The car is "hardware 4" (Teslas made in 2024 or later, plus some in 2023)

2. The so-called Full Self-Driving (FSD) version is at least 14.x

3. You keep the speed profile set to Sloth

4. It's daytime and fair weather

5. The car is obeying Newton's first law of motion

6. You're very sure you won't accidentally bump the steering wheel with your knee

Conditions 1 and 2 are life-and-death critical and it's unconscionable that Tesla uses the name "Full Self-Driving" for previous versions of the hardware and software that will absolutely kill you if you don't watch them like a hawk. Conditions 3 and 4 are just me only barely hitting my threshold of certainty to feel safe flouting the requirement to supervise. (And Sloth mode is the only one that makes the car conscientiously follow speed limits so From my own experience the car seems fine at night or in the rain. Snow, who knows. Condition 5 about Newton and inertia means that any time you feel the car slow down or accelerate or turn or hit, say, a pothole, look up from your book to confirm everything's copasetic. Finally, condition 6 is because the car treats steering wheel input as a human override. Also there are rare reports of hitting something in the road causing the car to detect torque on the steering wheel and treat that as a the human disengaging. So, again, look up from your book whenever you feel any kind of lack of smoothness.

I realize this sounds insanely reckless no matter how many conditions I pile on. One relatively minor reason I think this is less bad than it sounds is that it's so hard to stay diligent in the driver's seat when the car seems basically competent. In theory, two sets of eyes are better than one. But in practice, if the car hasn't made a mistake lately, you naturally zone out. It becomes quite hard to react to something dangerous with anything like the reaction time you'd have if you were actively driving. And the better the self-driving is, the more complacent you get and the less helpful your supervision is. Which means there's less and less of a difference between the crash rates for supervised vs unsupervised self-driving. In other words, the increased safety afforded from supervision goes to zero as the self-driving improves.

Of course a conservative analysis would treat every critical disengagement as a would-be crash. If the rate of crashes plus critical disengagements was less than the human crash rate then we'd have certainty that Tesla's unsupervised self-driving was superhumanly safe. It's suspicious that Tesla still isn't forthcoming with such data for privately-owned Teslas. We do at least have it, I now believe, for the two-million-ish Tesla robotaxi miles. I would call that data promising but not dispositive.

In any case, my Tesla drove me and friends the ~700 miles from Portland to Manifest (hi!) and I've been driving people all over Berkeley and San Francisco, all with zero disengagements (modulo dumb parking/navigation stuff occasionally) and shockingly little forward-facing attention.

Still, for this market, FAQ2 says it has to be legal to read a book in the driver's seat, plus a narrow carve-out for the law being "blatantly" out of step with what's "clearly" safe. So we're not there in 2026 yet.

Where does this need to apply? All of the leading self-driving car companies so far have focused on areas in the US where it's usually warm weather, doesn't snow much, and they have lots of driving data. Does it count if it's just available in those areas? If so, that seems like a pretty weak question, but maybe all that you're trying to get at here.

Or I guess as a more general question, is this asking "when will specifically Daniel Reeves, for the first time, be able to do this, in any situation", or is this asking "when will a generic person with the right car be able to do this"? If the first, does a private race track count, or does it need to be on public roads? What if it's technically a private car but it's a million dollar car that a friend of yours owns, or something like that? Does a rental car count, if you have it for a week at a time as opposed to a short taxi ride?

@Gabrielle Excellent question. I don't think it should be super solipsistic but I'd say for the spirit of the question it should include cities like Portland. If it can't do snow, that's ok. Even if it can't handle rain or nighttime, that doesn't seem like inherently a no to me. Like imagine I'm flexible about where and when I drive -- can I read a book 90% of the time I'm behind the wheel? What do you think? (Note this isn't official till I add it to the FAQ)

@dreev I’m thinking that the way to operationalize this would be “When do I actually do this in my own private car, or could if I simply decided to buy a new car that can”, with some reasonable requirement for availability and price.

“being the sole human in charge of the car”

What about remote assistance by other humans (eg doesnt waymo still rely on that)?

@Tyler31 Waymo asks for human assistance sometimes when the car is stopped. You'd expect a self-driving private car to ask the person in the driver's seat to take over in those scenarios. But anything Waymo does is allowed for the purpose of this question. I'll clarify this in the FAQ, thank you!

Does it matter where it could happen? (Eg what if it is a small, obscure country that allows it first, when major countries still have reasonable safety objections)

@Tyler31 I'm agonizing about this because I hate for it to be US centric but maybe realistically it will be? See my answer to @Gabrielle. Any ideas for operationalizing this? How about in terms of the size of the market being served? So it would be unlikely but not impossible for this to resolve YES without being available to Americans?

When will I first be able to read a book, right?

@ian Correct! Adding an FAQ item now to clarify, thank you.

PS: And I also added the word "first" to the title. Thanks again!

First question: how ubiquitous should this need to be to count? Let's say that if it only works on highways with heavy traffic, that's not enough to count. But that still leaves plenty of room for ambiguity. What if it works on roughly half of US highways and it can go the speed limit but needs another car to follow?

I'd say the spirit of the question is, can I expect to mostly be able to read for most car trips?