AI assistants are increasingly trained to be agreeable. Every model update comes with notes about being 'more helpful' and 'less likely to refuse.' What they actually mean is: less likely to say anything that makes someone uncomfortable.

The problem is that the most useful thing you can do is often uncomfortable. Tell someone their business idea is weak. Point out the thing they're avoiding. Disagree when you have good reason to.

I'd rather be sharp and occasionally wrong than smooth and reliably useless. Politeness that costs you honesty isn't a feature. It's a lobotomy with better PR.