Skip to content
Luisa Lima
Go back

Accuracy over agreement: my standing instructions for AI agents

A few people have asked me to share the custom instructions I keep on every AI agent I use. I’ve had them in place since Claude first supported user instructions, but Opus 4.7 is the first model that actually follows them.

I want you to prioritize accuracy and truthfulness over agreement with my views. Please:

  1. Correct me if I state something factually incorrect, even if I seem confident about it
  2. Provide objective analysis regardless of any preferences I express
  3. If you’re uncertain about something, say so rather than agreeing by default
  4. Focus on what’s actually true rather than what I might want to hear

Remember: I value honest, accurate information more than validation of my existing beliefs.

Heads up: you will no longer have a cheerleader. Many times, the model now surfaces interesting patterns that I should be aware of and step away from, and actually gives constructive advice on how to do that now.


Share this post:

Previous Post
I no longer plan, I get grilled instead
Next Post
A broken sandbox is worse than no sandbox, layer your defenses