Persistent Thought Bubble Expansion Controls

Feature Requests
1

Current Behavior

Currently in Cursor, AI thought bubbles (the “Thought for Xs” sections) automatically:

  • Expand during AI response generation
  • Collapse immediately when the AI finishes responding
  • Require manual clicking to re-expand for reading the full thinking process

Problem Statement

This auto-collapse behavior creates several usability issues:

When trying to understand how the AI approached a problem or why it made certain decisions, users must remember to click the collapsed bubble before it disappears, or manually re-expand it after completion.

The AI’s reasoning process is valuable for learning coding patterns, debugging approaches, and understanding complex problem-solving steps. The current behavior makes this harder to access.

Users focused on watching the AI’s thinking process get interrupted when the bubble suddenly collapses, breaking their reading flow.

Inconsistent UX: Users can’t predict or control when they’ll be able to read the full reasoning, creating an inconsistent experience.

Proposed Solutions

1. Settings-Based Control

Add options in Cursor Settings > General:

   "cursor.chat.expandThoughts": true,
   "cursor.chat.autoCollapseThoughts": false

2. Per-Conversation Controls

Add UI controls directly on thought bubbles:

  • Pin icon: Click to “pin” a thought bubble open permanently
  • Auto-collapse toggle: Per-conversation setting to override global preferences
  • Expand all/Collapse all: Buttons to manage multiple thought bubbles at once

3. Keyboard Shortcuts

  • Cmd/Ctrl + Shift + T: Toggle thought bubble expansion for current message
  • Cmd/Ctrl + Shift + A: Toggle all thought bubbles in current conversation

Use Cases

Developers studying AI reasoning patterns for complex algorithms or architectural decisions need persistent access to thinking processes.

When the AI’s output isn’t quite right, understanding its reasoning helps users craft better follow-up prompts.

Teachers/students using Cursor for learning want to study the problem-solving approach, not just the final answer.

Users creating tutorials or sharing AI interactions want to include the reasoning process in screenshots or explanations.

Mobile or small screen users need predictable control over what content takes up screen real estate.

4 Likes
2

Hey, thanks for the suggestion. I partially agree with you, sometimes it’s helpful to trace the thought process. However, they can be too lengthy, and not everyone wants to read them. Maybe in the future, we’ll make this option toggleable.

2 Likes
3

Super helpful to keep it on if you are testing different models and there are a lot of models to choose from now. Just because some want it off doesn’t mean that others don’t make very good use of reading the thinking and keeping the AI from going off the rails. A toggle makes all users happy.

4

Hello, i would like to have a choice here. Who are you to choose for me that the tought output is to long?? i find it quite disruptive as it is now. I dont want to trace everything by hand. i want to read every line of reasoning. i need it to understand and more importantly to learn about the behavior of the model itself! this is a very important issue for me. i find it borderline unusable atm.

5

I’ve gotten adept at clicking the ‘Thought for…’ so I can read it.
The problem is the text is dimmed and I end up leaning into my monitor just to be able to read it. Why would you dim the screen output?
It sounds like relatively few people actually care about the thought process…

6

I would love to have this option. The entire point of the OPTION is so that users can customize their own developer experience. Just because some users don’t want to see thought details, doesn’t mean everyone wants them hidden. I am always digging into the thought details when the agents start doing things incorrectly or behave oddly. Without access to thought details on some models (i.e. Grok Code) there would be ZERO insight into what the model is doing at all.

I’m concerned that Cursor is taking the route of so many products in recent years: To decimate features, and force everyone into a tiny, insufficient mold. The recent removal of allowlist, and concerns about what’s appearing in v2.0 nightlies, are indicating Cursor is starting to travel a path that will without question drive me away from the product. Don’t make a hyperopinionated IDE that forces everyone into one way of doing things. Removal of the terminal allowlist alone, has had such a huge impact on my workflow, that I am already wondering if Cursor is still a viable product. The comment above from Dean about some people not wanting to read thought processes, seems to completely miss the point of offering CONFIGURABLE OPTIONS to the IDE. The ENTIRE POINT of an option to auto-expand these is so that people who DO want to read them can more easily, and people who do NOT don’t have to.

I have so many concerns about Cursor lately. The pace of releases, the complete disregard for highly impactful bugs many of which have been around for months, the removal of critical features, this constant, never-ending march forward with never-ending releases that, primarily seem to DEGRADE the user experience with Cursor, rather than improve it…

I’m very concerned what Cursor is going to be like a month from now. I cannot even comprehend what it might be like by years end. Half the features I once relied on heavily are either broken completely, being removed, or have serious issues that just make using Cursor a frustrating and annoying experience. You guys are going to start driving your users away, if you don’t address your bugs, and stop removing features people use and rely on!

1 Like