I'd love to bring this on our radar again. I've be...
# proj-agentc
calm-artist-7206
02/22/2026, 4:30 PMI'd love to bring this on our radar again. I've been keeping an eye on this one for a while, but I don't know if Bend2 works or can work. I do think who can figure out formal verification in some usable way will be leading
cc: @wide-doctor-37830 would absolutely love your thoughts on any of this or related things
https://x.com/VictorTaelin/status/2025582923479466134?s=20
w
wooden-umbrella-84098
02/22/2026, 4:43 PMIMO you'd get 80% of the benefit of formal verification just by using a strongly typed compiled language like Go, Rust, or Java
wooden-umbrella-84098
02/22/2026, 4:44 PMThe big bottleneck is the amount of training data for the langauge. So I imagine e.g. Bend is not a great experience
wooden-umbrella-84098
02/22/2026, 4:44 PMI've been beating the drum for Go for a while, but tbh the agent still seems better at python
c
calm-artist-7206
02/22/2026, 5:57 PMI heard good things about Rust in particular, on this
calm-artist-7206
02/22/2026, 5:57 PMIn what ways does the agent seem worse in Go?
w
wooden-umbrella-84098
02/22/2026, 7:23 PMi'm mostly basing that on vibes 🙂
wooden-umbrella-84098
02/22/2026, 7:23 PMI haven't done a ton of Go w/ agents though, and the newer models might be better
wooden-umbrella-84098
02/22/2026, 7:24 PMI don't think the numbers in https://multi-swe-bench.github.io/#/ are super comparable across languages, but you can see a big difference there
3 Views