It’s good to see a new fresh push for protocols and interop. However, I don’t think it will do the thing we hope for, for non-technical reasons.<p>During Web 2.0, we saw similar enthusiasm. Instead of AI agents or blockchain, every modern company had an API exposed. For instance, Gmail- and Facebook chat was usable with 3p client apps.<p>What killed this was not tech, but business. The product wasn’t say social media, it was ad delivery. And using APIs was considered a bypass of funnels that they want to control. Today, if you go to a consumer service website, you will generally be met with a login/app wall. Even companies that charge money directly (say 23&me ad an egregious example) are also data hoarders. Apple is probably a better example. There’s no escape.<p>The point is, protocols is the easy part. If the economics and incentives are the same as yesterday, we will see similar outcomes. Today, the consumer web is adversarial between provider ”platforms”, ad delivery, content creators, and the products themselves (ie the people who use them).
I scrolled straight to section 2.1 "Definition and Characteristics of LLM agents" to find out which of the many definitions of "agent" they are using here.<p>They went for LLM + short-term and long-term memory + planning + tool using + action execution.<p>Presumably "planning" here is covered by any LLM that can do "think step by step" reasonably well?<p>It wasn't clear to me what the difference between "tool using" and "action execution" was.<p>I haven't seen a definition that specifically encompasses both short- and long-term memory before. They say:<p>> This dual memory system allows agents to maintain conversation continuity while building knowledge over time.<p>So presumably, this is the standard LLM chat conversation log plus a tool that can decide to stash extra information in a permanent store - similar to how ChatGPT's memory feature worked up until about four weeks ago.
This is a very weird paper. They go through all these protocols yet haven’t don’t mention the most promising one (smolagents): <a href="https://github.com/huggingface/smolagents">https://github.com/huggingface/smolagents</a>