I have been using GPT for development for months. I landed on GPT after testing various models. For me, it had the best performance for the price.
For these past couple of days GPT 5.5 has been terrible. It quickly latches onto components and then develops a response. The easiest way to summarize the behavior is that it is very unpredictable and unproductive. Things were so bad I had to “discard changes”, something I never use. Development is currently a huge waste of time and money.
For example, I say “don’t change anything” as part of my request. Included in the response the Agent says, “with your permission I will examine the documents”. I have a ton of examples but this type of misinterpretation is common.
So I asked the Agent to look at the last several chats and see if there is a reason for these problems. The Agent could not see the past chats. I have done this many times before. The files exist, I just verified it with another agent chat. Curiously, the second Agent could also not find the files, told me where to look, I find them, show a screenshot and then they say they did not look very thoroughly.
Conversations 92633c29-2513-438e-84d2-be79002010df and d07fe456-ba02-4f72-8455-0809f5019daa, 6913706d-bb10-44ed-af14-47dfc610db6b are examples.
Help.
edit: for clarity, GPT 5.5 Medium reasoning, 272k context not Fast.
Hey @Mark_Klimp, thanks for the report and for including the IDs.
I looked at the three chats. Each one ran on GPT-5.5 with Medium reasoning, as you selected, and completed normally. Nothing has changed on our side with GPT-5.5 Medium recently, and its behavior across all users this week looks the same as before.
Medium is tuned to start answering fast, so it can lock onto the first thing it finds. If you had been using a higher reasoning level or another model recently, Medium will feel noticeably hastier. A few things that should help:
For analysis or planning work, pick a higher reasoning level for GPT-5.5 (High) in the model picker.
When you don’t want any edits, switch the chat to Ask mode. Ask mode is read-only, so the agent can investigate without changing files or asking for permission to edit.
Start a new chat for each task. Quality drops as a long chat fills up its context.
For past chats, the most reliable way is to attach them directly instead of asking the agent to search the disk: type @ in the chat input, choose Past Chats, and pick the chats you want it to review. The agent then gets the full conversation without having to find the files.
Thanks for taking a look. I have been using GPT 5.5 medium for months. I never had problems looking at past chats and now the Agent in two separate conversations could not find the past chats. This recent performance is extremely different from prior development efforts. If performance was consistently the way is has been this week, I never would have selected this model. It is flat out counter productive.
Is there a way to reset the Agent environment for a folder/workspace?
I get the impression every new chat is a new instance. I’m just trying to figure out what I can do to fix this problematic and money wasting experience.
Every new chat is a fresh instance. The agent doesn’t keep any memory or state between chats, so there’s no hidden agent environment for a folder that could get into a bad state and need a reset. The only things that carry over from one chat to the next are:
Your User Rules, plus any Project Rules, Skills, or AGENTS.md file in the folder (Cursor Settings, under Rules)
Your past chats, saved on disk, which the agent only sees if it reads them or you attach them with @ Past Chats
If every chat in this folder behaves oddly, it’s worth checking that settings page and the folder for a rule or AGENTS.md that was added or changed recently. A rule telling the agent to ask before acting, for example, would explain the “with your permission” replies.