Anecdotally, it's currently feeling nerfed. The model keeps restating my intent as if it had thought of it. Output accuracy is slightly down, but the increased friction in communication is having an impact on my worfklow. Very annoying. Also, it flat out ignored system claude.md several times over the weekend. These observations hold between my personal pro and works enterprise.
Speed, instant finetunes? Highly optimized iteration cycles? The applications are the same for any reason why you might want local llms. This is crazy exciting.
It's the railroading that's being decreased. Various tech orgs are pushing back on everyone in the company pushing PRs as they realized is profoundly wasteful and unfair for people to push 10 PRs and then just dump it into someone elses lap to validate. "Claude already made a PR for it" doesn't hold up when you can't make a meaningful judgement on the output. Not every org is there yet, but my read is its happening at places like Goldman.
Everyone has gotten a boost in effectiveness because they're using the same multiplier. No one really cared before how bad some developers were because they were likely cheap, onshore or offshore.
The label might come later, but cooperative or adverserial behavioral patterns matter deeply when deploying agents to act as weapons of defense or business builders.
Anecdotally, sveltekit 2 was super quick to use for me. I tried building a somewhat interactive app with htmx and found the user experience was often incorrect on the Claude 4 family models. I switched to sveltekit 2 and it produced what wanted almost immediately. Features were added instantly. Even today with React and Opus 5.5 it couldn't match the development performance of version 2.
That said, both are "fast enough" and if you're on an enterprise plan, the cost difference is neglible. I waste way more more tokens due to poor or imprecise requirements than I do on development.
I mean, how fast do you need it to be? If you're running lambdas with cold starts all day long yes, java is a problem. Otherwise, occasional restarts of java services are quick assuming they're probably stateless. A few seconds? Whats the problem that needs to be solved here?
I’ve experienced deployments under heavy load being quite painful with slow restarts, even when doing e.g. canary releases. Blue/green would work but was in an environment where we couldn’t deploy immutably, we had to mutate the existing deployment in place.
Where are the competitor to the big frontier labs? The Chinese providers are demonstrating everything can be done cheaper and practically identical. I'm surprised upstarts aren't showing up.
reply