When AI Can Do Everything, Do You Still Know What You Want?

At Ryo Lu's talk at aiDotEngineer Shanghai, the first question he raised was a little out of step with the times: When we can do anything, do we still know what we want?

Baoyu reposted this.

The slides for the talk were all made by hand by Ryo Lu in Keynote, with Chinese and English subtitles. That detail is interesting today. AI tools can generate a whole deck in minutes, but he still typed out every page himself. It's probably not about efficiency. It's more like: figuring out what you want to say first matters more than saying something quickly.

The counterintuitive part is here. Over the past two years, the industry has discussed most often what models can do and where the boundary of capability has been pushed. Few people ask the reverse question: as capability goes up, has the sense of purpose kept up? When the cost of execution approaches zero, choosing what to do becomes more expensive instead. AI can pave the road for “how to do it,” but it cannot answer “what do you want” and “why” for you.

For people who write code, this question is more concrete. Tools are getting stronger, autocomplete is getting smarter, and product prototypes can be generated automatically. But no one autocompletes what to do next. It's the same for ordinary users who don't write code: in the past it was “not knowing how to do it”; now it's “not knowing what to do.”

The video is below.