This OpenAI product manager asks ChatGPT to impress him, and his real advice is about how the PM role is changing

Ty Geri, a product manager at OpenAI who helped build ChatGPT Work, ends every prompt by asking ChatGPT to impress him. He admits it might not change the output at all. “It could definitely just be in my head at this point,” he told Business Insider.

His other strategy is cruder: long, unstructured rants. Geri books one of OpenAI’s meeting rooms, turns on his phone’s dictation, and rambles at ChatGPT until something useful surfaces. He describes the model as his thinking partner, one that can sort through a stream of consciousness and find what is worth keeping.

He shares the approach with Andrej Karpathy, a founding member of OpenAI who later led AI at Tesla and joined Anthropic in May. Karpathy wrote in July that he liked ten-minute voice sessions that were, in his words, a total mess of stream-of-consciousness rambling. The technique works, he said, because the model reconstructs incoherent thoughts more clearly than the speaker delivered them.

Geri joined OpenAI last year after working at Google on AI search products, including AI overviews. The 28-year-old is the product manager behind ChatGPT Work, the agent OpenAI launched in July to handle multi-step tasks across a user’s apps and files.

The bigger theme in his interview is how AI is reshaping the product manager’s role. PMs at OpenAI are now shipping code, he said, and roles no longer start and end in discrete places. The primary job has not changed, to care the most about the product, but a PM’s ability to act on that concern has expanded dramatically.

That expansion cuts both ways. AI tools have given PMs the ability to understand parts of the engineering process they once left to specialists, but they have also removed the excuses PMs used to have for not grasping those processes. “You really can do anything,” Geri said, framing it as both an opportunity and an obligation.

Still, Geri said he hoped users would eventually not need prompting strategies at all. There were areas OpenAI needed to improve, with writing quality at the top of his list, but the goal was for things to just work without users having to learn tricks.