Erik on product management and such
The limits to language as an interface
In the Summer of 2023 I gave a lecture at New York University Shanghai where I examined the idea that all apps would soon be replaced by a single, universal chatbot. In January of the same year, Microsoft had bought 49% of OpenAI, partially because of their then goal to build Bing into a Super App inspired by WeChat that would replace Google Search and let you book hotels and shop for various products. Incidentally, pretty much what OpenAI just launched in October 2025, with Shopify integrations and their ChatKit.
That’s also why this post is relevant again. If your job is building products, your role isn’t to hype up new technology but to examine it carefully and understand its limitations so that you can understand what experiences are feasible and actually create value for users, and what experiences should never be built, or what simply works better as a traditional graphical interface.
Unlike other guest lectures on AI from the large tech giants like Alibaba that the students had listened to previously, mine was therefore a critical deep dive in text and natural language as an input method, and as a way of displaying the resulting information back to the user.
Here, I’ll recount my point of view on natural language as a user interface, and also add a section on OpenAI’s new ambition to turn ChatGPT into the Super App that Bing failed to become, and the business factors that will decide what becomes of ChatKit. In short, I believe it is more akin to something like a browser, in the same way as WeChat is a browser like I wrote in Why Super Apps took off in China.
Computing paradigms across time
Semantic search