I started down that road two months ago and its worked pretty well. for me, I do this:
1) I make golden path wireframes
2) swing them into grok and have it flesh out the various cases I need (form failed, validations, alternative views, etc...). Have it write a handoff doc outlining all the screen states i need
3) Give that doc to Claude desktop design to apply the project look and feel
4) copy those out to figma (I like laying everything out there so I can see all the variations and hierarchy)
5) Write up a handoff doc for Claude outlining the feature, exceptions, golden path, known issues and wireframes.
6) use Fable (or opus 4.8) to write user stories and tasks ONLY. I have the planning agent go through everything I gave and I spend 1-2 hours questioning it about everything it found, planning the changes, anticipating things like race conditions, planning out how to test it.... The end product is a series of user stories and tasks fit for the appropriate repo. I direct it to make sure everything is fleshed out enough so that Sonnet can do the coding.
7) switch to sonnet and let it go coding everything, each story becomes a commit.
8) After it's done I switch back to fable and have it run a graphical audit and fix against the screens first (because sonnet sucks at visual fidelity). Then I have fable/opus run full code review.
9) after I review the code and have it explain anything that looks weird, then deploy everything...