Whenever I use claude or copilot for a UI task I have to spent a lot of time getting the alignment and padding of elements correct

The initial pass of the agent usually works but everything is slightly off and things are out of alignment a lot.

Telling the agent to fix it takes like 5-10 tries sometimes as it tries different CSS changes

It seems the real problem is the agent cannot actually verify the result of how the UI renders

It’s like an open loop rather than closed loop