GPT-5 in Auto drives me mad (ᗒᗣᗕ) What do you do?

Discussions
1

When using Auto for models and you get GPT-5 models including Codex it’s obvious straightaway because the agents behaviour is so different … and problematic IMHO.

GPT-5:

  • doesn’t show as much thinking or activity in the composer
  • has long pauses of inactivity
  • it’s difficult to tell if it’s working or frozen.
  • doesn’t use Plan Mode in the same level of detail as other models
  • if you prompt it to add more detail it really struggles
  • doesn’t follow conventions like rules
  • over engineers features

Does anyone else have this experience with GPT-5 in Cursor?

I absolutely hate it when I see the interaction style of GPT-5 come up because I know it’s going to deliver something whack that I’ll have to undo with a better behaved agent.

What do you do to solve GPT-5’s behaviour?

2

One tactic I’m doing is that when I see GPT-5 has been assigned through Auto - I use it for Ask mode, documentation, and refactoring type work.

3

This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.