Here’s a simple test to see if you are who I described in the last blog post. If the money and users went away, would you leave technology? If you were building this stuff for free for you and your 5 friends, would you keep at it?

Math has this quality inherently. Perelman proved the Poincaré conjecture for him and his boys; no businessfuckers showed up and put ads in the proof, no politicsfuckers showed up and demanded the proof remind you that masks prevent the spread of covid. I wish this was true about tech.

At comma, our goal is to build one perfect box that drives a car better than a human. If we were the only one with that box, that would accomplish the goal of the company. Not making money or having users. Those things are fine, but the real goal is the building of the one magical box.


We need the cloud to go away. The cloud is a highway to serfdom. And this won’t happen with changes to politics or culture, technology itself is upstream of both of them.

Apps like Gmail can host 10,000 users on a single box, so there was never hope that everyone would run their own mail server. The economy of scale is too good. You’d be swimming super upstream to get Web 2.0 off of the cloud. Call it cloud-favoring.

There’s some good news in the form of the GPTs. ChatGPT can only host ~10 users on a box, though while ChatGPT has high requirements for compute, it has very low requirements for bandwidth. You could interact with it over a 56k modem. That tips the scales such that chatbots will remain in the cloud, even if we move to a model architechture that doesn’t benefit from batching. The providers want you in the cloud. Call it cloud-neutral.

However, robotics is different! Both openpilot and Tesla FSD don’t use the cloud and likely never will. The bandwidth and latency requirements are too high, and as with cars the same thing will be true for all shapes and sizes of robots. Gaming has always struggled to be in the cloud for similar reasons, note that when you buy a Switch or PS5 the compute actually is local. Call these cloud-averse.


Robotics is inherently cloud-averse. The companies will all try to shoehorn cloud into it, but like forcing you to create a cloud account for your microwave, people will see it as stupid and favor the products that actually just run locally.

As long as there’s decent open source alternatives, the robotics ecosystem will have checks and balances against enshittifcation. As the quality of macOS degrades and Windows forces more and more cloud account crap on you, is the era of the Linux Desktop finally here? I suspect the breakdown of robotics operating systems will look similar to computer and phone operating systems.

I also really like the way nanochat is thinking about the foundation model problem. I predict that robotics software will move towards on device learning, but there will still be a foundation model in the repo, and these things will be expensive to make. However, that cost can simply be a commodity – the way nanochat talks about a $100 ChatGPT.


More and more of the training code will make its way into openpilot as the robots learn online. There will still be a pretrained world model, but that will be a commodity. A fully reproducible build in 5000 UOps in tinygrad + some commodity data.

The way the waves of robotics will break looks bright for individual sovereignty, and robotics will be where the real value in AI comes from. Many “knowledge work” problems are adversarial; not true for physical things like farming.

I’ll end with the quote from ecromata:

we will abolish scarcity; there will be nothing for them to steal they can’t have for free