AI Agents are still underweighted

People are not thinking enough about agents, whether that be coding agents or general AI agents, because they already seem super hyped and they’re almost feeling overhyped right now, so it’s kind of cringe to say that you believe agents are even better than they are. But I think if you’re not preparing for a world in which you actually can use 100 coding and general AI agents in your daily work, you’re going to be leaving impact on the table.

Thinking about everything that you’re working on in terms of what data do I need, how do I process that data, what outputs do I need, for what purpose? Where does intelligence fit into that picture? Do I need intelligence to help me find data, clean data, link data, parse data, translate data, extract? Do I need to create trends or find similarities or structure the data? What kind of artifacts do I need to output? I think you should be thinking about all of those questions in the frame of how can I do all of that but in a way that leverages AI agents? Because even if you have to do a suboptimal process or project to make the agents fit that shape, you’re going to get so so so much leverage from the agents that it doesn’t really matter.

And maybe there’s some weird implication here that what you should be optimizing for now is putting the thing out in the world to create the Schelling point, iterating on the idea space, finding the people that you can vibe with and iterate with, and just prepare and build the foundation. It’s going to be iterative still. You can already use agents right now to code and to do general work. So you should already be doing that a lot, even more than most people are already doing. And then every month you should just be thinking how can I use more agents and how can I use them in more impressive larger tasks or in new areas that they didn’t work before.

That has already been happening for two years now for coding. You could actually just use the first ChatGPT or GPT-4 for basic coding tasks. It wasn’t great, but you were getting some leverage and then every six months there’s been a step jump. And now it’s really strong and it’s going to get stronger every few months. You should be thinking along the same lines for generalized work.