Claude Code Desktop is a full blown IDE now. Love the workflow. I am curious if there is any need to actually open Cursor anymore. Spoiler: there is.

tl;dr

Pro

Full dev environment with local machine, testing, git workflow and visual pointers for referencing specific areas in the design. (Is Lovable cooked?)

Contra

It’s a dev environment from Anthropic. So no independent agent handling anymore. No switching to Gemini or Codex for specific tasks where they would perform reasonably better.

Anyway

Currently testing it on a deliverable for a MVP. First I pointed the agent to a Figma design that I prepared. And by prepared I mean: I put some effort into designing small details and it turned out nicely. Now the layers had to be renamed for giving claude maximum context. (Not sure if that actually helps).

Design MVP

From this design context I built a semantic design token file that I can reference while building the MVP. It takes the design context and translates everything into semantic tokens. It’s not perfect but a start. There is some refinement happening as I am typing this.

Token and component level

I wanted to get more out of the design to build all the prerequisites. This will matter later when we are going to build the actual MVP. I want the design being really used. Im thinking to build a component page, where the agent pulls every detail on component level and put them on a HTML canvas. With this approach I can see how my design is being understood. With this file I can also make adjustments on the front-end level.

/limit hitting

It’s not even 9 am and I am already near hitting today’s limits. This sucks but helps me to refocus to other tasks. ¯_(ツ)_/¯ Now if I would be in Cursor I would just switch to Gemini or Codex. But I encapsulated myself using Claude Code Desktop. Here we have the main issue. As much as I love using it, I am trapped in a golden cage.