Hi, David, thanks for the intriguing article! I am puzzled by your comparison of LLMs with the "Model" from MVC.
Models, as I understand them, define data structures and business logic. LLMs are a technology that processes text input, reasons probabilistically and generates an output.
In this emerging, "agentic" world, they are the engine that makes this (decentralized) ecosystem possible. They're not a data store or layer for interaction with it.
They're the cooks, but the fridges and pantries are not theirs!
What I think is fundamental:
Views=Apps are going to get more stupid. We wont' care about tight app-view communication, we will care about tight agent-agent communication. We'll shift focus from smart apps to simple agents. We will care about clean "endpoints, protocols, and interactions" (as you mentinoed), standardized protocols for seamless agents work.
The role of apps, then, can shift to be stupid rendering of the data (delivered by the waiters - agents) and mutations (fetch, update, ...)
Could you clarify how you see LLMs fitting as Models, or is there a separate data layer I’m missing? Would love to hear your thoughts, Michal.
Hi, David, thanks for the intriguing article! I am puzzled by your comparison of LLMs with the "Model" from MVC.
Models, as I understand them, define data structures and business logic. LLMs are a technology that processes text input, reasons probabilistically and generates an output.
In this emerging, "agentic" world, they are the engine that makes this (decentralized) ecosystem possible. They're not a data store or layer for interaction with it.
They're the cooks, but the fridges and pantries are not theirs!
What I think is fundamental:
Views=Apps are going to get more stupid. We wont' care about tight app-view communication, we will care about tight agent-agent communication. We'll shift focus from smart apps to simple agents. We will care about clean "endpoints, protocols, and interactions" (as you mentinoed), standardized protocols for seamless agents work.
The role of apps, then, can shift to be stupid rendering of the data (delivered by the waiters - agents) and mutations (fetch, update, ...)
Could you clarify how you see LLMs fitting as Models, or is there a separate data layer I’m missing? Would love to hear your thoughts, Michal.