Yes, this post is translated by AI, you use it for code, why not for english?
Okay, Cursor changed their policy, and we’re cooked, we can’t do much. Yes, most of us are unhappy. But then, what can we do?
I’ve been studying prompt engineering, but the auto model usually doesn’t allow for thinking models. So again, what can be done? What are you all doing to stay productive even with a weaker model?
Everyone complains, but some people will push past these issues by crafting better prompts.
Considering that it’s practically impossible to do Tree of Thoughts or Chain of Thoughts now, what I’ve personally been doing is blowing up the context window with most of my files since that actually helps.
Another trick I’ve been using is the magic phrase:
“Ask 20 surgical questions about the implementation, analyzing my codebase to help you infer the step-by-step process.”
This helps me understand what the AI (which isn’t really “thinking”) is at least inferring, and gives me space to correct some logic mistakes.
Where did I get this from? Experience. GPT-4.1 doesn’t even come close to Claude Sonnet (even without thinking). It doesn’t matter what the benchmarks say experience tells the real story. Before thinking models existed, I used to split my prompts into larger blocks. (with claude 3.5/3.7)
Sure, this turns one prompt into 2 or 3 interactions but it actually makes the tool usable!
Nowadays, we’re already seeing competitors releasing alternatives. But for those who still want to stick with the chaos that Cursor IDE has become, I believe this is the way.
Let me be clear: it wasn’t impossible to code before thinking models existed. They’re extremely helpful but we need to think too.
What I want is for you all to share your experiences. What have you been doing to overcome the complete inefficiency of Auto Mode Models?
P.S.: Since it’s probably against the rules, please don’t talk about model activation here we all know it happens. What I really want are techniques you’ve been using. Let’s share!
And for those who still have access to good thinking models, especially beginners, I’d recommend learning the basics of prompt engineering. You can start on YouTube to understand how personas work, or check out https://www.promptingguide.ai/ or relevant subreddits.
That’s it!