I stayed up til 2am and spent $1,000 benchmarking Jev Router so you don't have to. Performance on DeepSWE was roughly the same as GPT-6 Astra on low. It costs slightly more, and it took almost 5x longer to run.

I hate to break it to everyone, but Opus 5.5 has been nerfed starting the last few hours today. Autistic Claude speak is back, it's doing really dumb shit & this is not the same model I was using the last 2 days. Confirmed this on multiple accounts, it's turned into an idiot.

Pretty surreal seeing the Muse Charm on stage at Connect. It was really fun incubating this lil guy alongside a small group of incredibly talented designers. Technology should be fun!

I keep coming back to this. AI isn't shattering egos. It's inflating them. It's harder than ever to keep it in check. You think of the problem that's beaten you for years. You solve it. Then you do it again tomorrow.

Sorry but AI still sucks at writing code. Working code and *good* code are two very different things. I spend a lot of my time reading ai generated code and then manually editing it to make it simpler, or repeatedly prompting with exact instructions to get what I want. I often end up with half of what the agent originally produced, functionally equivalent, easier to understand, and often much more efficient. The models are quite smart these days and have a broader knowledge of algorithms and data structures and I do (or at least remember from school). They also sometimes think of edge cases I don’t. But that still doesn’t always translate to an efficient implementation without a lot of coaxing. That’s why no matter what the influencers say, coding is not “solved” yet. Good engineering practices haven’t changed. Reading code and knowing what could be improved is still an important skill whether you’re the one doing the typing or not.