So, like a lot of people (well, more like a lot of people who are perpetually online), I started using Claude Code again in December after trying it back in late May of 2025.
Now, I wasn’t that impressed, though very excited, with the initial iteration of Claude Code. I think it was inevitable that this sort of agent orchestration type thing, with just full integration of a frontier model, the lab’s services, engineering prowess, etc., was going to happen. And yeah, I used it quite a bit. Got frustrated, and I think it was born of the limitations of the models at the time. And that’s okay. I think that will always be the case.
Fast forward to December (early December, late November really) and I see murmurs of Claude just being immensely improved. The latest model, Opus 4.5, just really, really impressed me, and they’ve done something unique and special here, because I can fire up a CLI session of Claude Code and have a genuine back and forth brainstorming session, providing inspiration to the model about what I want. And it will, at times, aggressively try to implement it. And it’s damn good.
I think (and this is not my original thought) we’ve reached a tipping point in model capabilities, and even more so, the agent harnesses.
And yeah, I don’t want to spend too much time on this blog post because I just want to get back to it.