Now what I didn’t realize at the time, because I’m not an expert in all like design patterns, is I didn’t in fact want a cache I wanted a catalot. I wanted to store all this data in a database long term.

But it was too late, Claude had gone out and used local storage (which is technically persistent. Like it has fucked all the things.

Coding problem? Skill issue? Did I try the latest models?

I don’t think these things are entirely separable. The code compiled but didn’t do what I asked, I don’t know what we mean by “is coding solved”. But if in order to be relatively efficient we require the user to know which model to use depending on the complexity of the coding task, then I’d say it’s not “solved”.

To answer your other question, I write code in the drivers seat. I’m looking to optimize my workflow, but I try to work atomically and do a combination of understanding the general code layout and architecture of the application before prompting Claude to make relatively specific changes. More autocomplete less agent (though I use agents or other things just not adding all the code blind.

And I remember saying to a bull on this idea: “I’m sure wearables will unlock some cool capabilities. But I think humans just innately like holding a screen in their hands where they can watch YouTube videos; like they _really_ like this. This is a killer experience. YouTube, games, whatever this is peak.

And while not on par with the portable television. Like - custome AI-generated software is going to be great! But a visible representation of your model that someone is responsible for curating and looks the same no matter who is looking at it.

Like that’s pretty good. And I think a thing most people can agree on is that even if code is free. Software is still hard.

He could mean a few other thingsc one would be, he has like 20 version tags and he merges 15 features into each version (does he explicitly say merge to main?)

The other thing he could mean is he has like a software engineering agent and that software engineering agent like loops through GitHub issues and his personal notepad and maybe uses a few different branches to test things out and build I dunno adversarially, running all sorts of experiments.

Which would be genuinely cool! But using Mr’s to say tweak bunch of variables back and forth isn’t what 300 MR’s implies.

But then the ultimate question is - it may be cool to fully automate a software engineering agent, and certainly is the type of research Anthropic and someone of Boris’ stature (and pay level) should be working on. But is it efficient?

I guess yes hems talked about this:

https://karozieminski.substack.com/p/boris-cherny-claude-cod...

I am in no way surprised a sufficient waterfall method passed to Claude code could result in a completely accurate application. But most applications aren’t built via waterfall for all the reasons.

Also agents are just loops. So if you use Claude Code you are doing. Everything with a loop. So I do believe him but Im not entirely grokking the flex.

It gets said ad nauseam but a lot of software development is remixing. Think about how much gaming innovation happened in the Warcraft and StarCraft map editors. The Birth of tower defense, moba, and probably many more.