← All IntelClip / EducationA descriptive agent model, and the tool that breaks it
From Designing AI-Intensive Applications - swyx · ≈8:05
Sets up the talk's sharpest point — a genuinely valuable, 70,000-reader AI product that uses none of the intent/control-flow/memory/planning machinery the agent definition demands.
What’s in it
- Sets up the talk's sharpest point — a genuinely valuable, 70,000-reader AI product that uses none of the intent/control-flow/memory/planning machinery the agent definition demands.
Clip transcript
OpenAI is pushing. Um um the way that I approach sort of the agent standard model has been very different. So you can refer to my talk from the previous conference on that. Um basically trying to do a descriptive u top down u model of what people use the words people use to describe agents like intent um you know control flow um memory planning and tool use. So there's all these there's all these like really really interesting things. But I think that the thing that really got me um is like I don't actually use all of that to build AI news. Um by the way who here reads AI news? I don't know if there's like a Yeah. Oh my god, like that's half of you. Thanks. Uh uh it's it's a really good tool I built for myself and you know hopefully uh now over 70,000 people are reading along as well. Um and the thing that really got me was Sum at the last conference. Uh you know he's the lead of PyTorch and he says he reads AI news he loves it but it is not an agent. And I was like what do you mean it's not an agent? I call it an agent.
Comments
Sign in to comment.
Loading comments…