I worked for a founder who did this all the time and it's incredibly stressful as an employee. It would be fine in the limited case where the features are fast and easy to implement: say two days' work at most. I enjoy my paychecks clearing every other week, and losing a sale over small amounts of effort is dumb, particularly since sales where relatively large (low hundreds of thousands of dollars). But this founder didn't know and frankly didn't care about that; he talked about features we'd just begun to do basic research on as if they existed.
I'm being intentionally somewhat vague here, so it's hard to really relate the details, but in one case he discussed a feature (having to do with distributed markov chains) that not only didn't exist but, afaik, nobody in the world even understands how to do and is the subject of active research. Needless to say, not something that could be implemented in a couple days. Worse, we where talking to some phds who fucking understood the above, and they immediately started asking me how the fuck we did it!
And yes, it often made us look like complete assholes in front of customers.