Fair question! Honestly, if that loop works for you, keep it. Onlook doesn’t replace your harness; it runs on your own Claude or Codex.
What it would add to your workflow is the part you called “a few cosmetic fixes”, where we’ve found it can be slow to describe everything in words.
On our canvas you can see every screen in every state side by side, and point at the things you want to change.
It’s the same agent, with a view of your whole app.
My usual workflow here is to ideate using variants and then committing to which ones to use or fold in before I create the PR, so I can visually verify the result of the merge
Design work with AI is definitely a game of balancing creativity and flexibility with the constraints defined in the design system, and it's been something we've thought a lot about.
Onlook lets you add browsers and images to your projects to be used as context, so it's already a strong tool for working with design references
Another way we're addressing this is by keeping the projects you're working on separate from the self contained design artifacts you may generate in the process. Since these can be easily used across projects we find that it's very low friction to take a cool but discarded thing you sketched once and repurpose it for another project where it may fit in better.
What it would add to your workflow is the part you called “a few cosmetic fixes”, where we’ve found it can be slow to describe everything in words.
On our canvas you can see every screen in every state side by side, and point at the things you want to change. It’s the same agent, with a view of your whole app.
reply