Transcription
In the past few weeks, I have been suffering from agentic fever. It is a sort of burnout. It's what the experts in burnout call it when you're feeling crispy. And it's that feeling that if I'm not coding agents, if I'm not creating agents, and if the agents are not working for me at a certain moment, I'm wasting my time.
Now, I want to today define this agentic fever a little bit more because a lot of Silicon Valley, for instance, right now is suffering from this agentic fever, and a lot of people around me are as well, kind of really enthralled with this AI evolution. I want to first define this AI evolution, and at the end, I want to give you a little bit of my agent philosophy, both for the humans that need to take care of the agents and then also how to instruct agents.
But how the hell did we get here? So, it was pretty fast. Right at the beginning, what we thought we had was a chatbot. And actually, [snorts] the story of ChatGPT is OpenAI releasing the API, realizing that nobody is putting a chatbot out on the internet, and then doing it themselves. And what we had here is the artifact was a conversation, and the innovation was context. When we ask Siri a follow-up question, even today, Siri doesn't know what we're talking about. Instead, these AI agents had context. And so, the beautiful thing is that we finally kind of had a friend, right? And a lot of people kind of befriended this, this new technology.
Then something else happened with the release of Codex, with the release of uh, Code Interpreter, and so on. We'll windsurf, you name it, all of these technologies. The, the, the coding tool was born, and the artifact that it produced was code. Like, "Lovable, make me a new website," right? And here, the interesting thing, the innovation was that finally, we went from conversation to craft, and I finally had a junior engineer or coder at my service.
Then something else happened. I was using these coding tools three weeks ago, right? Like this. I was telling the system, "Create code for me. Create a system for me. Create a CRM for me. Create a Kanban board for me." Then I realized that the same tool, we're going from from chatbot, the coding tool, to coding tool. But the coding tool can be used as an agent. And you can instruct the coding tool to be the artifact, not to create something, but to *be* that something. And once you start getting into these agentic loops, you realize that here, the innovation is not just craft, it's craft plus self-improvement. If I ask the agent to create code, that code stays as it is until I tell it, "Please improve the code." Once the agent is running the whole process, once the agent *is* the code, I can inject this self-improvement. And here, we finally have a COO, a Chief Operating Officer, in the AI.
But as I'm doing this, what is happening on the internet? And I'm going to not going to tell you what is happening on the internet either, you know, or you don't, is that we are getting to agent orchestration. So, the next coding tool will be, and already kind of is, an agent orchestrator. And here, what is the product of an agent orchestrator? It's a team of agents. And so, you have to start imagining that within these two last categories, we have one great revolution happening, which is for right now, we have compounding improvement. And the beauty of the agent orchestrator is that we might give it skin in the game. And this is what people are doing right now in a few online projects that maybe we can discuss more in the comments. And we are seeing the shift of AI from a friend to a CEO. In what am I going to say? Two and a half years, AI is going from a friendly conversational system into a CEO. This is, this is a giant evolution, and it comes with problems, and it comes with human problems. Agentic fever.
So, agentic fever is this sense that, well, some people in Silicon Valley are already thinking it's kind of over. This is the last few years to make money, maybe the last few months to make money. So, run. People are constantly feeling that they are missing out. I was constantly feeling that I was missing out because I was not on the agent actively, um, getting obsessed with understanding all the tools, setting the tools. So, this whole thing with OpenClaw, which I'm not going to explain, you can find other videos. I, yes, I have installed my OpenClaw. I wasted a lot of time on it. But it was, it was teaching me this agentic philosophy as I was using it. So, it's not in itself a technology. It doesn't do anything really, and it doesn't do anything that Code Interpreter cannot do, but it's a really interesting learning experience for the human. And, you, you have this feeling of constant fear because things are changing so fast. And so, I think a lot of us have been thinking and feeling this, this AI burnout, this agentic fever. And I believe that this is just a step. It's, it's, it's a learning curve. The things are moving too fast for us not to have a human reaction.
Tell you, the way I'm dealing with it is understanding more and using these tools more. Obviously, I'm sleeping much better than two weeks ago. I've already gone back to, to my healthy routines. I've never kind of lost my healthy routines, but I'm, I'm trying, and I'm starting to understand what is the right posture towards this technology because I think we can go to a better, better way, an agentic philosophy. And this agentic philosophy has two parts: the human part and the machine part. The human part is more important than the machine part, but the machine part is more detailed.
So, how do we get out of and how do we deal with agentic fever? So, let's take care of the human. So, if you're using Code Interpreter today and you're watching this video, if you are using OpenClaw today, if you're using Code Interpreter today, if you're using agent mode on ChatGPT, you are not missing out. You are on the bleeding edge of what is happening. So, forget about the fear of missing out. And humans are great at being human. And it's quite paradoxical. I'm giving you a tool that is so great at execution, and it can do, it can work as a COO. And what do we do? We get the burnout. So, we, it's, it's paradoxical, but this is, this is how our, our psyche works. And I have never been as excited with a technology. The last time I was as excited with a technology, it was 1993, okay? And it was the web, and it was creating HTML pages by hand. Okay? There were no HTML generators, and the, the bleeding edge web technology was CGI BIN. And most of you don't even know what it is, but you know, Google CGI BIN if you want to suffer. Okay? Server side. Okay. [laughter] That's another, that's another video. That's another rant.
But humans need to continue being sharp, sleep, stay calm because they are the trainers of the models. They are, and we are good at being human. And even if we fail sometimes, and we get overexcited, and we lose sleep, then what we need to do is go back to what we do best. And we'll discuss this in a second. AI needs us sharp, okay? Because it's, it's very tricky to have it work right the way we want it to work. And what we need is a, just to go back to systems thinking. And if you don't know about systems thinking, go look next. The next video on the playlist, hopefully, is a video on on systems thinking. And in case I will do one, is that good inputs bring good outputs. And so, if you are feverishly burning out in front of the machine and trying to make it do more and more and more because the world is going to end tomorrow, you're going to create a frantic, anxiety-driven agent that is going to fail. If you distill, calm down, and then instruct the agent, I think what you will achieve is something much better.
So, as I told you, three weeks ago, I was still in this mindset: the coding agent creates code. And then I shifted into the coding agent is an agent, is my agent, and it does the work, and every once in a while, will call some code, but it will call it from a conversation. So, this is a whole agentic philosophy that I am understanding. I am learning from other people. This is how I would summarize the machine side of agentic philosophy.
So, the first thing is that code is new, and language is now code. So, simple markdown SOPs, standard operating procedures, are equal to code. And the human, not only is the human in the loop right now, but the human is the instructor. And the instruction needs to be very clear: that we are not here to build, we are here to instruct. And your data is the training. And so, for instance, I created an outreach machine. And I thought at the beginning, the way I thought through the outreach machine was in terms of of systems and and silos and mechanisms. And I was thinking, "Oh, I need a CRM. I need this, and they need that." And what I did in the end, and I was thinking, oh, sequentially. And then I was thinking, you know, in the end, uh, I will need to have a, a web interface where where I can click on things. And what I did instead in the agentic world, and I said, you know, the agent is the center. And then there's some API calls to Mailchimp, there's some API calls to LinkedIn. Well, LinkedIn doesn't have APIs, but there's tools that have the APIs. There's, um, there is a Google Sheet, so where some data is stored. There is the file system. But the agent is the surface, the agent is the software. So, your data is connected to, once you connect your data to these agents, and be mindful, be careful about that, that is the training. And then you start instructing the, the agent to log everything, review and improve, and be proactive. And if you are successful in this, you are going to enter this kind of compounding phase for, uh, for the agent, where the agent, every day, gets a prompt from you saying, "Go into the improvement routine. Look at the logs. What can you do better? What can we do more efficiently?" etc. And then we have these new paradigms where we're creating skills and getting skills from the internet, and these are the job descriptions. And then we have agent orchestrators. But in general, already, any agent tool can create sub-agents, and you will have these sub-agents be the workforce and use different models. Like, you use a, a model, a more expensive model to orchestrate, and then you use another less expensive model to do some of the work. And then for the brute force, for the web research, etc., you use an even cheaper model, and you have this tiered architecture.
So, I've tried to do a ton in this video. Um, this is, this is what we explored. I believe there's, there's so much more to explore than, than what we just said here, and probably there's a lot of shorthand in this video that wasn't completely understandable. I would love to, to hear from you and hear about your experience with, uh, this excitement, this agentic fever, uh, how you're dealing with it. Give me, give me some feedback here in the comments. And I'm right now, one of the reasons why I've been posting less is I'm obsessed with this as well. And I'm coming down. I'm slowly coming down from, from, from the, the high and the risk of getting crispy and then burning out. And actually, being here with you, sharing what I'm doing, being more mindful, staying with my, my good routines, uh, going back to having an excellent sleep score on my, my little Aura, and, and all is going really well. So, don't worry. But we are in such exciting and fast-paced times that I do understand that I was going into burnout. I do understand that I was towards burnout. I do understand that we have this fear of missing out. I do understand that we are very much in a pickle. And I do understand that everything is changing really, really, really fast. In my talks, I talk about this idea of going from malaise to renaissance. And this malaise to renaissance, you see what I did there? Yeah. And this is possibly the malaise phase, and it's going to be a little bit tricky until it gets better. So, hang in there. Let me know in the comments. Thanks a lot.
There might be another video here appearing, which is a really good watch after this one. I really suggest. I think it's going to be here. It's going to be an amazing video here. You shouldn't miss the video that is appearing here. It's a really great video. I hope it's appearing.