Ty Geri has a habit when he opens ChatGPT: he asks it to impress him.
"Please impress me," is how he puts it — and then, by his own admission, he thinks that shouldn't be necessary at all. "You shouldn't have to," said Geri, a product manager at OpenAI, according to a LinkedIn item surfaced by Google News. "The models should try to impress you regardless."
The same item notes that Geri joined OpenAI last year, after working at Google on AI search products.
That one-line career detail is the part worth pausing on. Search and chatbots started as different products, and they are steadily converging: a search box that writes you an answer and an assistant that goes and looks things up are, increasingly, the same thing. People who have built on one side of that line are valuable on the other, and Geri's move is a small, concrete instance of talent flowing from Google's AI search efforts to OpenAI.
His complaint is also a product philosophy in miniature. Today, getting a strong answer out of a chatbot often depends on how you ask — the prompt-engineering folklore of telling a model to try harder, think step by step, or pretend it's an expert. Geri's framing treats that as a design failure rather than a user skill. The burden of effort, in his view, belongs to the system, not the person typing.
It is a modest data point — a quote and a job change, reported secondhand through a social post — and it should be read as such rather than as a statement of company strategy.
Why it matters: the fight between Google and OpenAI is not only about who has the better model, but about who makes the good answer arrive without the user having to beg for it.