Agent response in chat is downgraded

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

The response feels very downgraded from past 2-3 days irrespective of models, agent and model is not doing extensive research and not following instructions. Yesterday it was completely off and today morning it was working but again from 12 noon it producing very downgraded response and lazy to no efforts in following instructions. It is just giving assumptive answers which was not happening a week back or even today morning

Steps to Reproduce

Not Sure

Expected Behavior

it should do deep research, check documentations and codebase , verify files, docker logs and provide detailed report

Operating System

Windows 10/11

Current Cursor Version (Menu → About Cursor → Copy)

Version: 1.7.46 (user setup)
VSCode Version: 1.99.3
Commit: b9e5948c1ad20443a5cecba6b84a3c9b99d62580
Date: 2025-10-14T01:21:46.830Z
Electron: 34.5.8
Chromium: 132.0.6834.210
Node.js: 20.19.1
V8: 13.2.152.41-electron.0
OS: Windows_NT x64 10.0.26100

For AI issues: which model did you use?

sonnet 4.5, gpt 5 pro, opus 4.1, gpt codex

For AI issues: add Request ID with privacy disabled

The respose feels very downgraded from past 2-3 days irrespective of models, agent and model is not doing extensive research and not following instructions,

Does this stop you from using Cursor

No - Cursor works, but with this issue

Hey, thanks for the report. To properly investigate the AI quality issue, I’ll need a bit more info:

  1. Request IDs (critical):
  • Open the Chat sidebar → menu in the top right → Copy Request ID
  • Share 2–3 Request IDs from recent “downgraded” responses
  • Make sure Privacy Mode is disabled when testing (Settings → Privacy)
  1. Specific examples:
  • Share one concrete prompt where the agent ignored instructions
  • What did it respond with vs. what did you expect?
  1. Project context:
  • Does this happen across all projects or only in a specific codebase?

I came to the forum looking for a post like this because I have also felt like response quality has been noticeably degraded in the past couple days (more hallucinations about what functions already exist when analyzing existing code, more rambling/repetitive responses when brainstorming new features).

When I asked in this request: 491fe32f-446f-4084-87f0-d0a36014a996 how to configure the vim extension to move by visual lines instead of actual lines when the text wrapped, and it responded with unrelated suggestions about the clipboard, I decided to come here to see if others have recently experienced degraded quality:

The solution is to add the vim.useSystemClipboard and most importantly, configure the Vim extension settings to use visual line navigation.

  1. Request IDs (critical):
  • Open the Chat sidebar → menu in the top right → Copy Request ID2: a587e037-8c1c-42ed-a58f-1b8b9e649e6d;
  • Share 2–3 Request IDs from recent “downgraded” responses: and few i am unable to copy as its showing duplicated chat option when i click
  • Make sure Privacy Mode is disabled when testing (Settings → Privacy): Yes its diabled
  1. Specific examples:
  • Share one concrete prompt where the agent ignored instructions: It is like it s not following the instruction complete and research and codebase scan is its doing tis very lazy, same prompt if I rerun after couple of days and when its working fine then I will do through research verifying all the required file in codebase, do required reasoning and then give concrete answer. But is scenario when its not working it gives presumptive answer, doing minimal research and reasoning of the implementation and reading of file in codebase
  • What did it respond with vs. what did you expect?: Same models same prompt, provides details and through researched and reasoned output and to will aware to what i am talking and what is going on.
  1. Project context:
  • Does this happen across all projects or only in a specific codebase?: All project I have tried, it experience this. And especially in weekend its is crazy, like Saturday end of till Monday, IST it always appears like models are running minimally and this have been there for months, no work can be done. But past 1 week, model performance have been unusable.

Thanks for the info, that’s very helpful.

This topic was automatically closed 22 days after the last reply. New replies are no longer allowed.