What do AI systems of the early 2020s like Agents/Bots/ChatGPT have in common?
All are text based. Just like the internet apps Telegram, XMTP, Farcaster, …
Chats and human communication are text based.
All verbal communication can be modelled as written transcripts and written communication is text.
A simple such model for communication between two parties is:
which portraits a seriesof message, response pairs.
Now this model can be extended to n parties trivially.
As AI these days sends responses to so called prompts, we get message, response pairs.
Now extend the in and output destinations and you get AI everywhere. The connections can be made by relaying. In the end the whole web can be modelled as message, response pairs. HTTP is message, response pairs. So any API interaction is message, response pairs which is why Model Context Protocol (MCP) works well to connect the two.
This mental model allows us to think in a systemized way and use these traces of message, response pairs to reason about system correctness, availability as well as speed and capabilities.
You can develop the same core and use it for traces (message, response pairs) in all in/ooutputs like Farcaster, XMTP, Telegram, Chats, APIs and so much more.