AI Loops: How the World's Best Engineers Use AI

**Ejaaz** (0:00)
90% of people are using AI models the same way that they use Google. But recently, a new way of prompting your AI has emerged that doesn't just replace the way that you work, it promotes you to the CEO of your very own AI company. It's called Loops, and it's part of a growing development in agent autonomy, where AI agents basically spin up and autonomously complete tasks or goals that you set for it, often working throughout the night. In 2019, the longest that an AI agent could work autonomously for was for two seconds. Fast forward to today, and they can work autonomously for 12 hours, and that's doubling every couple of months. Andrej Karpathy calls this phenomenon the autonomy slider, where you can take a dial that slides from humans that approve everything to humans that periodically check in, and it's part of this growing trend of agents consuming and taking up more of human capital and labor. The question that remains going forwards is, what will humans do and will they be entirely replaced by AI, or will they be the ultimate orchestrator of their destiny?

**Josh** (0:58)
Yeah, I think the goal for this episode is really just to inform people on what's possible current day with these agents, with these LLMs, with writing these loops, as well as where you can possibly find yourself within that stack, because it gets pretty complicated when we're getting into loops. Not everyone needs to use loops, but everyone should be using LLMs probably slightly different than how you're using them today. So maybe we could start with a little history lesson in terms of the four levels in which we have been engaging with LLMs starting with the first level, which is just prompting generally like most people are probably still doing. This started in 2022, 2023 around the release of ChatGPT. The way that you would engage with these LLMs is you would just submit a question or you submit a prompt and you get some language back. Now, if you are still doing this, that's okay because I find a lot of people are. But understand that this is how you engage with the model three years ago, four years ago. It has since advanced pretty meaningfully since then. The second step of this is agents, and we're going to spend some time on agents.
Everyone's heard of an agent, maybe not everyone knows what an agent is. An agent is something that could think for a little bit longer. It can run a bit longer than just a standard prompt. It can go off and do things. It could call tools for you.
It's much more capable version of the text box. Then like we talked about all the time on the show recently in the last few weeks, there's the harness feature in which you put an LLM into a container, and that gives it a memory feature that gives it complete tool use. That's something like an open cloth that we've talked about a lot, but some people do use, and that's level 3 Now level 4, which is the new thing that has come this week, that's really been highlighted by some of the top leaders at these AI labs, is loops. A loop is essentially a version of an agent that has an orchestration layer and builds upon itself.
It allows you to continue to scope yourself out. If you can imagine you're dealing directly with an employee at level 1, and then you're directing that person to go off and do their own in level 2
At level 3, with the harness, you're kind of directing a series of people to help you. Then level 4, you're just the top level CEO who's directing your C-suite to go and manage all the employees below you. There's an entire stack to this. It's very cool. Ejaaz, how do you use your AI currently? Where would you say that you fit in this stack?

**Ejaaz** (3:12)
Looking at this diagram that we have on the screen here, I'm somewhere between number 2 and number 3 I'm somewhere between using agents and trying to figure out the whole harness thing. Now, what am I doing when it comes to spitting up agents? If you look at either my Cloud or my ChatGPT desktop apps right now, I've renamed a bunch of my conversations to a particular focus or subject, and then agent after it, and so I can go to it, and this agent basically has all the context of what I wanted to do, whether it's research a particular topic, create some kind of an outline for something, research a particular investment angle. It already knows and has the embedded context for what it needs to do, and there's usually like one to maybe three tasks that it needs to autonomously execute on its own, and so it runs in kind of like a sequence, but if any of that sequence kind of breaks, let's say it kind of tries to retrieve data from some particular website and it is unable to do so, it breaks and it comes to me and it says, hey, Ejaaz, is there some other thing that you want to look at or retrieve from blah, blah, blah? It's not fully autonomous. Now, number three, the harness side of things is what I'm trying to like kind of like mold my understanding around. What I've noticed is, when you type in a prompt and you get a response, you can kind of tell that it's AIE, like usually when we kind of create artifacts, it comes in a particular font or it speaks in a particular type of language. The harness helps kind of like take your prompt and kind of mold it into something that is more human-like, but also more nuanced with what you are trying to do. Like it effectively gets closer towards that ultimate goal. Like we were talking before recording this episode about human taste and how AI doesn't really get human taste. The harness helps you get towards that ultimate kind of taste profile for the particular output that you're trying to generate.