Slow AI models make you nervous? Claude goes down for a few hours and your working day is done? GitHub losing 9s of uptime faster than you spend your daily AI quota makes you lose your hair? These are rookie problems.

Our AI agents struggle in a world of fiber internet, managed clouds, and a tool for everything. So what happens when we want to build an AI agent that will (eventually) work on Mars?

From the creator of “How to build a website that will (eventually) work on Mars,” a talk that aged about as well as you’d expect, comes a sequel with worse latency and higher stakes.

I gave a fun conference talk about building a website that would eventually work on Mars 10 years ago, and it went better than I expected. It's time for a sequel!