One of the big limiting factors there is infrastructure. You have to park, charge, clean and repair them somewhere. One city also probably can't support more than 20,000 in the short term. If you swarm the streets without people expecting it, people will protest.
I described the issues I was having, wanting the animation to loop, fixing arms intersecting with the body, and it built Python scripts to parse, analyze the animations and fix the issues. Took some iteration but it works quite well.
I'm using it to build a 3D engine and the way it's intermingling application logic and game logic in one big main module is really terrible imo. There is no planning ahead, even though I put together a design.md explaining where I intend for this design to go.
Also not sure businesses are gaining much by taking said risky bet. I'm seeing a lot of likely vibe coded bugs, not a lot of actual software improvement.
Sort of. I've had CC implement web software and found later that it had decided to not gate certain endpoints based on user login. It just decided that data ought to be public I guess, and never told me. You can easily fix the code, but leaked data can be more problematic.
I found that the code quality coming out of GPT 6.0 sol was so awful it was just... Objectionable... Then they released 6.1 a week after, like they knew something wasn't right. They immediately yanked that model 😆
Rewriting a single function is one thing, rewriting an entire large piece of software with nontrivial integration is much harder. Also, even at 10,000 tokens per second, you're still outputting what, 40 kilobytes of code per second? That's assuming *zero* reasoning tokens.