I recently had Gemini disagree with me on a point about philosophy of language and logic, but it phrased the disagreement very politely, by first listing all the related points in which it agreed, and things like that.

So it seems that LLM "sycophancy" isn't necessarily about dishonest agreement, but possibly about being very polite. Which doesn't need to involve dishonesty. So LLM companies should, in principle, be able to make their models both subjectively "agreeable" and honest.