Currently, our interactions with AI seems to be very request-response. While for most purposes this is enough, I envision a much more realistic interaction with AI, close to how real people talk to people through chat apps.
friend.com seems to have nailed at least some part of this. You can see that the chatting flow is highly resemblant of a normal chat with a person through the internet. The lack of punctuation or small length of the messages, and the regular double/triple texting to put an idea forward, all make the experience very realistic. But there are still a lot of things missing here.
First off, it is still request response in the sense that I gotta open the site to talk to the “friend”. In a real setting, the person can send a message to me whenever they feel like it. Which means I should receive messages from AI unprompted, during random times in the day.
Second, I need the AI to use multiple texting and keep the length of the messages to a minimum.
Third, I want multiple AI agents to communicate between each other. Imagine something like an AI workflow: I get a business idea in the shower so I have a pre-prompted AI character named “Business Idea validator” who helps me validate this idea and help make improvements and refinements accordingly. Once the idea is refined, the validator AI agent automatically transfers a summarized version of the chat I just had to another AI agent called the “idea to execution” agent who then takes this idea and sketches out a very well refined plan from top to bottom. And this all happens behind the scenes, the only interaction I have had
And finally, fourth, I need the AI to be very customizable. I am thinking of a realistic Ali Abdaal chat clone who texts me every night to check up on my productivity throughout the day and every weekend morning to evaluate my progress on my goals this week and his insights on it. This requires a standardized framework which lets users do just this. A new app for AI interactions.