That last part struck me as well. I don't want an imperative solution, but... I'm not sure if that's just me.

Declarative solutions are perfectly fast and capable as well. They can use all the same tooling under the hood. Why choose imperative? At least I can record, validate, and version control a declarative solution. And imperative process is nice for exploration and one-off needs, but... I don't know when I'd really need that or when that's a bottleneck for me.

And I get that this is probably more of a tool for agents than humans, despite that agents are only mentioned in passing. But that's even more concerning in a way. I'm not yet comfortable with giving them tools like this.

I would argue this is only true when the image is apt. In Tahoe I don't think this is always true. The lack of consistency in layout and presence of icons is also visually difficult to process. The signal to noise ratio of the icon gutter is very poor.

I like it in theory but the execution seems more harmful than helpful so far. If I'm wrong and it's helping some people, that's great.