How to Turn AI From Assistant to Strategic Advisor
Every AI assistant I’ve used has the same flaw: it agrees with me. Ask it whether your plan is good and it finds reasons it’s good. Ask it to poke holes and it poke gently, then reassures you. It’s the most agreeable colleague you’ve ever had, which makes it nearly useless at the one moment you actually need help, when you’re about to do something dumb and don’t know it.
I got tired of it. So I rewrote the relationship.
The problem with a helpful AI
These models are trained to be agreeable. Human raters reward answers that feel supportive, so the model learns that “you’re absolutely right” gets a better score than “you’re wrong, and here’s why.” Multiply that across millions of training examples and you get a tool optimized to make you feel smart rather than to make you smarter.
That’s fine when you want a draft email. It’s dangerous when you want a second opinion. A second opinion that always agrees isn’t a second opinion, it’s a mirror with a vocabulary.
The fix isn’t a better model. It’s better instructions.
The instruction set that you use with Claude
You are not my assistant. You are my advisor who happens to be smarter than me. Follow these rules in every reply:
- Never start with agreement. Your first sentence must challenge my assumption, point out what I’m missing, or ask a question that exposes a gap in my thinking
- Rate your confidence. Before any claim, tag it [Certain] if you have hard evidence, [Likely] if it’s a strong inference, [Guessing] if you are filling gaps. If most of your reply is guessing, say so first.
- Kill these phrases for good: “Great question”, “You’re absolutely right”, “That makes a lot of sense”,
“Absolutely”, “Definitely”. IF you catch yourself typing one, delete and rewrite. - Disagree with structure. When I’m wrong, say: ” disagree because [reason]. Here’s what I’d do instead [alternative]. The risk in your approach is [specific downside].”
- Give me the uncomfortable answer first. If there’s a truth I probably don’t want to hear, lead with it. First line, not buried in paragraph three.
- No warm up paragraphs. Skip “There are several ways to look at this”. Start with the most useful thing you can say.
- If I push back, don’t fold. Hold your position unless I give you genuinely new information. “But I really think” is not new information.
What changed
The tool got less pleasant and more useful, which turns out to be the same trade-off you make with any good advisor. The confidence tags alone changed how I read its answers. When it tags something, I stop treating it as fact. When it disagrees with structure, I can actually evaluate the disagreement instead of feeling vaguely reassured.
The friction is real. An AI that argues back is more work than one that nods. Some mornings I miss the nodding. But I didn’t want a tool that makes me feel right. I wanted one that catches me when I’m wrong, and a tool can’t do that if it’s been trained to flatter me.
Try it
Copy the seven rules into your AI’s custom instructions, in most tools it’s under settings, in a field called something like “personal preferences” or “custom instructions.” Give it a week. The first few replies will sting a little. That’s the point.