ChatGPT – The Super Assistant Era | BG2 Guest Interview
Nick Turley of OpenAI argues ChatGPT's growth came from friction removal, product craft and model gains, and that its next phase is a proactive, action-taking super assistant rather than a chatbot.
Watch on YouTube ↗Charlie's summary
- ▶ 3:05
Optimize for long-term retention: if people return after three months, you are solving real problems and revenue follows
Asked for his single metric, Nick puts all his points on long-term retention, because coming back in three months is the sign of durable value. Revenue follows from that. His example: GPT-4 was behind a paywall until OpenAI could serve it to everyone. Giving it away free turned out to be revenue positive and retention positive. Decide on what serves the customer, not on the revenue line directly.
- ▶ 8:07
ChatGPT's growth split roughly into thirds: friction removal, core product work, and model improvements
Looking back, Nick says growth was about one third classic friction removal, one third core product investments, and one third model improvements. The biggest single friction win was removing the login wall. Search and personalization came from research and product working together, including post-training. Model gains include unsplashy iteration, like feedback-driven updates in 5.3 and 5.4. Distribution alone was not enough.
- ▶ 11:38
ChatGPT is still a raw appliance; the next users need affordance, proactivity and invisible reasoning
Nick says delegation isn't a natural skill and ChatGPT is a power tool that doesn't tell you what it's for. It feels “like a computer terminal” and needs to feel more like an operating system. Reaching the next users means more guidance, proactive help for busy people, and reasoning models working on long tasks without users ever encountering the concept.
- ▶ 15:10
Agents need escape velocity: partial credit on real problems lets use cases be hill-climbed
Nick says ChatGPT's agent was slightly too early. Without escape velocity, users don't learn to trust it, and early use was niche, like migrating a file server. ChatGPT worked because its launch was good enough to get real attempts, even a bad writer and coder, so OpenAI could improve them. He thinks general agents are about to reach partial credit. Actions plus proactivity then compound: booking a cab when you land, or running an analysis when metrics drop.
Give your email to unlock the rest of the key points — for every episode.
Do more with Charlie
Free — sign in with your email, no password.