Eric Siu explains the difference between AI agents and loops: loops record and compound learnings over time, while agents don't learn and must be rerun each time. He uses a workplace analogy comparing A- and B-player employees to show why loops outperform agents in practice.
This is the difference between an agent versus a loop. The main difference is that a loop is able to record the learnings and it's able to compound on those learnings. The problem with an agent, you got to keep running it over and over. It doesn't learn. You got to keep repeating it. And it's like, that's like a solid B player. A B player employee is like, you got to tell them what to do. An A player employee is like, they just solve the problem and then they compound. They keep learning. I was actually on a standup call about 30 minutes ago and I was talking to one of our designers and I was like, you're really good at AI right now. Who taught you all this stuff? And I wasn't expecting him to say me at all. Him. He taught him all this stuff. Good people don't need to be told this, right?
