Milo: fast like Rust, simple like TypeScript. A memory-safe systems language on a new point of the tradeoff curve: no GC, no lifetimes, one owner per value. Also: SMT-proved contracts, reverse debugging, packages, 86 std modules, checked FFI, LSP. github.com/milo-language/mil…

When we achieve super intelligence, RSI, all that, the TI-89 calculator will still cost $90.

I've been building an AOT TypeScript compiler, playing around with Rust and Clang forks, and creating new languages since December 2025. It's cool to see more people doing it, and of course it won't stop there. Writing software and apps is now so easy that in many ways it's not even interesting. I'm finding what was fun was the challenge of building or discovering something new. I'm growing bored of compiler rewrites and all the other games and apps I've speedrun. Projects that took me years before AI I've rewritten in my own language with better outcomes, and nobody even bats an eye. With AI, the scope of projects can now go to the absurd: trawling through all recorded data and analyzing it at a scale that was never possible before. Government records, scientific data, grand mathematical challenges, archaeology, weather patterns, biological research: all of it can be analyzed with endless intelligence. All usable data can and will be mined and combed over for nuggets of utility. Software won't go away, of course, but the more interesting things are much further up the abstraction chain. The best part is that we no longer need to argue about semicolons or naming conventions. We can focus on more impactful things like drug research, clean energy, and materials science. We can see it happening with companies like Discovery Loop, Open AI and Anthropic's research, Warp Science, and surely more to come. Even looking back a year or two, I feel like a caveman tediously writing line after line of code. The side projects I had in mind were embarrassingly unambitious compared to now. It's hard to express to a non-swe how different AI has made my job. At some point in 2025, I wrote my last React component by hand. I don't know what lies ahead, but I could never go back.

The perfect programming language doesn't exist because of the varied use cases and environments in which the code runs. Maybe we should be building languages that are easily modifiable/extendable for the specific runtime attributes we want to optimize for. Or is that that just asm.

Remember when we used to make our agents talk like a literal caveman because tokens were so expensive?

Goose (xitter.cf/wvo/status/21079796605…) beats Milo on all five of its benchmarks on the largest inputs (1% to 34% faster) and uses less memory on three, up to 3.9x less. Impressive work @wvo.