🙌UIs and internal metrics are two things used to skip because they are a hassle, but language models drop the cost of that to almost zero (and multiply the upside)

Oh totally, “find all the places i should have metrics and add them” works really well, and even if its overzealous its so much simpler to remove extra ones than find the motivation to go add them

You can also tell Claude to work it over and polish it up.

One thing that helps is re-use a single UI and just add tabs for new projects. It’ll keep the style and components somewhat consistent across them so you’re not starting from zero; there’s a few default “claude UI conventions” I really dislike and have to undo when starting greenfield.

I put the UI in the main directory then use a subfolder for each project. To start a new project, start in main directory and prompt: “Add a new tab in the UI and folder for a project called XYZ”. Claude will tend to stay within it’s local directory but can go read adjacent project code if needed.

Shared CWD across projects is nice because you can for example say “match the schema returned by the service in <other project path>” and it can just go read that and get it done

Now just use it to up the janky CLI to a great CLI or TUI! 🙌

Every little DX improvement I can imagine I now have time to make in the ever evolving search for the perfect monorepo. I love it! Even the cli scripts become beautiful

Its because of AI all my projects in R get a neat shiny web app nowadays.

This is something I started doing, too. I hope it becomes more reliable with ex: an electron app at least. Asking it for a desktop app is a lost cause. I also want more htmx and less react.

Try having it make you a TUI instead of a GUI—it's not half-bad. (Most of my CLI hackery is in golang, but I'm sure there's fancier stuff out there.)

I also do my stuff all in go, what tui frameworks do you have it use? I havent done one of those in ages

github.com/charmbracele... I confess I haven't done much work on this... I just asked Claude Code to evaluated the top 3 search results through subagents and pick one for the project.