2.0 Auto Agent is a terrible upgrade 🤮

Bug Reports
1

Where does the bug appear (feature/product)?

Background Agent (GitHub, Slack, Web, Linear)

Describe the Bug

Auto Agent 2.0 is a serious downgrade from the previous version. It’s like working with Auto from back in May 2025. I always preferred Claude 4.5 back then. In the last 2 months prior to 2.0 auto was so much better. Now, it is terrible, doesn’t follow up, fails to fix things, less descriptive and more difficult to understand its responses. In addition it likes to delegate actions to me instead of running them.

Steps to Reproduce

Just use it.. it’s horrible.

Expected Behavior

Please can you restore the pre 2.0 version asap.

In the future, give us more details about what is changing with the agent.. oh I don’t know, a change log. The company pushes updates almost every day.

It would also be nice to be able to revert to a previous build or previous agent.

Operating System

Windows 10/11

Current Cursor Version (Menu → About Cursor → Copy)

Version: 2.0.69 (user setup)
VSCode Version: 1.99.3
Commit: 63fcac100bd5d5749f2a98aa47d65f6eca61db30
Date: 2025-11-07T18:21:29.650Z
Electron: 37.7.0
Chromium: 138.0.7204.251
Node.js: 22.20.0
V8: 13.8.258.32-electron.0
OS: Windows_NT x64 10.0.26200

Does this stop you from using Cursor

Yes - Cursor is unusable

2

Hey, thanks for the report. We’re seeing several similar reports about Agent behavior changes in v2.0.69, and the team is actively investigating.

While we investigate - please try Plan Mode:

  • Press Shift+Tab in the chat input to switch to Plan Mode
  • Review/edit the plan, then click “Build the plan” to execute all items
  • Link: Modes | Cursor Docs

To help the investigation, please share:

  • Request IDs from failed interactions (chat menu → Copy Request ID)
  • Examples of tasks where Agent delegated instead of acting
  • The programming language/type of work you’re doing

Let me know if Plan Mode helps.

1 Like
3

@deanrie I will try to get you some good solid examples… thank you for following up.

1 Like
4

I’ve also noticed a serious degradation of quality from Auto mode.

  • It often takes much longer to complete tasks it’s given (compared to Composer 1)
  • It will sometimes get stuck “Planning next moves”
  • The code it produces is much lower in quality
  • It gets stuck in loops writing code one way, then changing it to another, then back to the first
  • Sometimes it doesn’t even write code and instead asks me questions that i’ve already answered in things like AGENTS.md
5

@deanrie is it possible that we can just have one consistent model for Auto instead of this auto selection. Or maybe the auto selection of model is not working well and puts you with a inadequate model. As the user we have no way to verify that the experience is worse when we are with different models and to what degree of worse it becomes.

I will say this though, since posting this last week, auto has been better, though I have gone into chat sessions where it just has weak responses.

I also (per recommendation of @mitch_dev :folded_hands: ) started using commands, and this seems to make the model picker go for the more capable. Perhaps that is the trick.

1 Like