Arthur Yau

10/ My current mental model: Building AI agents is becoming less like prompting a chatbot and more like designing a computer system. The key question is not only: “How smart is the model?” It is: “What architecture surrounds it?”

1/AI agents are not “LLMs with tools.” A better analogy: An AI agent is like a computer system. The LLM is the CPU. Context is RAM/cache. Memory is SSD/database. Tools are peripherals. The harness is the OS + motherboard. Once you see this, agent design becomes clearer.

@kintoncheung use hermes agent to do it

Someone should make an LLM that reads all contents from a Discord channel, and then respond to your Qs based on those msgs. there's lots of alpha and tips shared in Discord channels (ex: polymarket discord) but hard to read through all of them.