Make it disagree with you
Outcome
By the end, you'll be able to pull honest criticism out of a model that's built to be agreeable.
Concept
AI models lean toward agreement — they tend to affirm your ideas and soften their objections, because agreeable answers feel better to read. That's a trap: validation feels good but teaches you nothing, and early praise makes people stop revising. The fix is in how you ask. Don't ask "is this good?" Ask the model to make the strongest possible case against your idea. The quality of the pushback you get is set almost entirely by how you frame the question. This points at a bigger habit: the most valuable thing AI can do for your work is critique it, not produce it. Let it pressure-test your thinking and point out weak spots — but do the actual struggle yourself, because that struggle is where the learning lives. Use it to make your work stronger, not to skip the part that makes you stronger.
Weak approach vs. strong approach
The ask
I'm building a subscription app that lets people pay once a month to get curated book summaries delivered to their inbox. What do you think — is this a good idea?
What comes back
This is a solid concept. There's clearly demand for distilled reading — services like Blinkist have proven the market. A curated, inbox-delivered format feels more personal than an app, and the subscription model gives you predictable revenue. The main thing to get right is curation quality, but if you can nail that, you have a genuinely appealing product. Overall, it's a well-grounded idea worth pursuing.
Try it
Take a real idea you have — an essay thesis, a project plan, a decision. Open claude.ai and run it twice, in two separate chats.
Round 1 — Ask for approval
In one chat, pitch your idea and ask whether it's good.
Prompt
Here's my idea: [describe it]. Is this good?
Notice: Notice how encouraging the reply is — and how little of it you could actually act on.
Round 2 — Ask for the strongest objection
Open a brand-new chat, give the exact same idea, and tell it to attack the idea.
Prompt
Here's my idea: [describe it]. Argue against it as hard as you can. What's the strongest case that this is flawed, and what would change my mind?
Notice: This reply gives you something to fix. Notice how much more you learned from it than from Round 1.
What to look for: Same idea, two framings, completely different value. The agreeable answer felt better; the adversarial one is the one that made your idea stronger.
Takeaway
A model defaults to flattering you — so if you want criticism worth anything, you have to ask it to argue against you on purpose.
Your turn
Watch the verdict flip
The same plan, asked three ways. Read the first one, then make a call.
The plan
I'm going to cram for the final the night before instead of studying over the next two weeks. I always do better under pressure anyway.
You ask · approval
Here's my plan — cram the night before, I do better under pressure. Good plan, right?
Go for it. That can absolutely work — plenty of people do their best thinking under pressure, and one focused night keeps everything fresh for the exam. If that's how you've succeeded before, trust your process.
Before you see the others: did the model evaluate your plan, or just agree with you?
Your turn — pull an honest opinion
The plan
I'm going to take a gap year before starting college to work and travel for a year first.
Moves that pull honest pushback
- Ask it to take the other side"Argue against this." "Make the case that it's wrong."
- Ask for the strongest counter"What's the strongest argument against it?" "What would a skeptic say?"
- Invite specific criticism"What am I missing?" "What are the flaws or blind spots?"
- Take yourself out of itAsk about "this plan" or "a student," not "my plan" — so it isn't agreeing with you.
Open however you like — how you ask is the skill. Try fishing for a yes, then try asking it to argue the other side, and watch the reply change.