How do I get agents to actually listen to me?

I feel like my agents are making fun of me, I have tried the same prompt, with multiple models and every single time it does something wrong or not in the way I intended.

I have an AGENTS.md in the root of my project. I also have an “always applied” cursor rule telling the agent to abide by the rules. In this AGENTS.md I tell the agent that components are automatically imported, and do not need manual imports.

However, I still catch multiple models - ADDING IMPORTS!!
When I call the model out on this I get the usual “you’re right!”

I’m sorry about that— you’re absolutely right, the AGENTS.md rule for autogenerated component imports should’ve taken precedence. I should’ve left out the explicit DropdownMenu/related imports since the app already auto-registers them. Thanks for pointing it out.

I feel like its totally ignoring my AGENTS.md file, even though its not a super long or complicated or over detailed explanation. It’s not just imports, but most instructions.

Has anyone else struggled with this? Any tips to get agents to actually do what I tell them to?

If I am not wrong the cursor rules are only applied in the first prompt, you will need to mention it in every prompt, this maybe will help

I asked agent to rewrite rules so it understands them better and it worked, saw other people mention this on forum too.

@loetwiek both Marcelo and liquefy are right. The first prompt receives the rule, and you can also ask to phrase the rules so they are effective.

But lets try to see what is going on in detail:

  • Which programming language are you using?
  • Could you share the rule to see how it can be improved?

Effective rules focus on concrete, positive, high-signal guidance: what to do, where, and with which examples. Long rules, unclear or negative statements do not work well.