The biases I'm talking about here are things like their generally preferred approach to solving problems, or how optimistic or pessimistic their outlook on certain classes of problems are. These are things you tend to know about the people you're likely to brainstorm with, plus you can also be fairly sure that you're all working towards the same goal. When you're using an LLM like a partner to do this you can't know those biases, as they're externally controlled, you can't know how they've changed since the last time you interacted with it, because the training data may have been updated, and it's definitely not working towards the same goal as you because it's not working towards anything, it's just following statistical patterns in its training data.

A handful of employees in a conference room are more likely to just agree.

I see you've never met my coworkers before! The politest way to describe that sort of gathering would be "a free and frank exchange of opinions". I quite like them for that. The point is that I know their biases already, so I can adjust for them, and I can be confident that they're actually trying to achieve the same thing I am, rather than regurgitating a probabilistically derived sequence of tokens that strongly resemble, but aren't, actual thought.

The biases I'm talking about here are usually more along the lines of preferred approaches to problems, general pessimism or optimism in outlook, and any "oneupmanship" that is known to occur, rather than anything people tend to hide.