I have been looking at AI PM opportunities lately, so I explored some products that interested me and noticed several problems:
1) Making prompts visible. When I need an agent to complete a task, the system gives me a prompt. (The interface shows options, and this is the result displayed after I click one.) It does reduce the trouble of making users think about how to design a prompt. But logically, this should be hidden inside the system as part of its architecture. It makes me feel that the team has not done the specific architectural design internally.
2) Building an AI search engine. Replacing the work of a traditional search engine with AI is still the same linear internet way of thinking. This is one thing that frustrates me. Echoing the first point, the architecture still has not really been built.
3) Are you giving me information or insight?
Perhaps most people use tools because they want them to find the material they need as quickly as possible. But I am even lazier. Do not only give me the material itself. Tell me which parts of it deserve my particular attention... things like that.
This makes me think of how almost everyone keeps saying that human judgment and decision-making must not be transferred to AI. But what will users actually think? That is another question. And a product's values matter, matter, matter.
4) ChatGPT said that so far, the only truly AI-native products are coding products. That instantly woke me up. I have to say, the more I use ChatGPT after 5.5 launched, the more useful it feels. 🥺