In-browser AI responses difficult to read

[dillonoconnor]

When asking AI a question in the browser, all of the "thinking" done by the LLM is visually the same as the actual response it delivers at the end, making it hard to see where the latter starts and ends. Would suggest some sort of visual differentiator.

[backnotprop]

Ill have a fix for this once I finalize the same capability for plan mode

[sylviafrank]

excited for plan mode inline chat — the use case i'm most hoping for is user-initiated threads anchored to a specific block: asking "why did you choose this?" or "have you considered X?", getting a response inline, then either dismissing the thread or saying "add that reasoning to the plan." basically interrogating specific decisions without having to send everything back as full feedback and restart the loop. happy to contribute as well if that'd be helpful!