ChatGPT – The Super Assistant Era | BG2 Guest Interview

He says he’d put “all my points” on whether people come back months later, because that’s the best signal ChatGPT is solving real problems; he frames revenue as something that follows durable value rather than being optimized directly.

He attributes the “smile” to many incremental improvements plus users needing time to learn what they can delegate; he highlights search and personalization as major shifts that increased daily and personal relevance (including a move toward mobile-first usage).

He breaks historical growth into roughly three buckets: removing friction (e.g., removing an authentication wall), core product investments that combine research + product (e.g., search/personalization and crafted UI like writing blocks), and ongoing model improvements (both major steps and quiet iterations).

He says ChatGPT needs more “affordance” than a ‘computer terminal’ so busy, non-AI-native users can understand value quickly; he also emphasizes proactivity and productizing reasoning so the system can help without users needing to understand the underlying concept.

He distinguishes limited actions today (web search, images, etc.) from the broader action space of a human with a computer, which he says they aim to build; he notes earlier agent attempts were “slightly too early” because models weren’t good enough for trust and repeated attempts.