← sources

My AI assistant keeps refusing to do work it has done before

source: reddit (reddit:ChatGPT)published 2026-11-07 06:36ingested 2026-11-08 02:36

retained text

​

I asked my dot to help add a proper dark mode to my website. I gave it permission to work on the development site and then on the live site once the changes were working.

When it said it couldn’t write to the dev site, I had the access fixed and asked it to try again. Instead of trying, it kept talking about restrictions and asking for permission I had already given. I had to ask repeatedly whether it had actually attempted to write anything.

I specifically asked it to test the access itself, without another agent. It told me it would, then said it couldn’t, and then used an assistant anyway. The write test succeeded, but by that point I had spent far too long arguing over a simple request. The dark mode still wasn’t implemented.

This has happened multiple times before. I’ve asked it to complete work I had already authorized, only for it to introduce extra barriers, stop short, or leave me pushing it to continue. It keeps giving me conflicting explanations and saying it will do things without following through.

What makes this even stranger is that it has done this kind of work for me before without these problems. This time it gave conflicting explanations for why it supposedly couldn’t, even though it had worked on the site previously.

If there’s a real limitation, explain it clearly. If there isn’t, follow the instruction. I shouldn’t have to spend this much time arguing with an assistant to get it to try.

Has anyone else run into this kind of behaviour?

document fetched 2026-11-08 00:19 · canonical https://reddit.com/r/ChatGPT/comments/1xpzokz/my_ai_assistant_keeps_refusing_to_do_work_it_has