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 correctThe 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 sometimesIt seems the real problem is the agent cannot actually verify the result of how the UI rendersIt’s like an open loop rather than closed loop