Users Report Claude Injecting Unsolicited Moral Judgments Into Responses

I was about to post a very similar thing. I stopped using Claude and went back to Codex and it’s going so much smoother and faster.

First of all Claude is just slow. It’s easily 3x the time as GPT. Second, it’s hella expensive.

But third, it will do stuff like this:

“Here’s a robust plan of 10 tickets to implement this feature”

“I did 1, 3, and 5. I deferred 6-10. And I decided to not do 2 and 4.”

Does my feature work? No!

Me: “it’s not working, we need to complete that plan.”

“Actually that’s not the right plan. Here’s a plan of 10 things we need to do before we can reevaluate the remaining items on the old plan.”

And I can run in circles - very slowly and expensively - like that.

I completely defeats the point of agentic development if the plans are shit and the coding agent refuses to do the work and gets distracted and all I get is something endlessly broken.

If I wanted that, I’d hire a real dev! 😂