“Learn AI” Is Bad Advice. Learn These Instead

**Greg Isenberg** (0:00)
Imagine it's a few years from now. AI can build almost anything, write almost anything, and do most of the tasks people used to get paid for. In that world, I gotta ask, what skill is still valuable? I've been thinking about this nonstop, and I've narrowed it down to six core skills. None of them require a fancy degree or connections, and all of them could be started this weekend. And every single one of them gets more valuable as AI gets better, not less. By the end of this episode, you'll learn the six most valuable skill sets on the planet with some tips to get started. I'm gonna go through each of them super clearly so you will understand where the world is going, what you need to learn, and just concretely how to get started. Let's not waste any time, and let's go right through it.
So the first skill that I think is the most valuable skill right now is people who can set up agents properly, manage them and run local AI models. This is basically the grown up version of prompt engineering. So a lot of people learned how to type a good prompt into ChatGPT, which is useful, but the next layer is being able to design a little AI employee that has contacts, that has tools, that has permissions, that has memory, that has a goal, and a way to check its own work before it bothers you. That skill, that little skill is going to be valuable because most companies are about to have the exact same problem. They're going to have 10 AI tools, 50 workflows, a bunch of half working automations, and nobody understands how to turn that into an operating system, which is what they really want. The person who can walk in and say, hey, here's a customer support agent, here's a research agent, here's the sales follow-up agent, here are all the rules, here's what it's allowed to do, here's where it needs approval, here's how we know if it's working, that person becomes really, really hard to replace. And the local AI part, which I've talked about on this channel before, I've done a couple of videos on it, is important because there are certain workflows where privacy or cost, the prices of these models are going up and up, latency or control matter a lot. If you could run these models locally with something like Ollama, LM Studio, you start to understand what can happen on your own machine and what needs the cloud and what needs to touch private docs and what should stay behind the wall and how this all interacts. So I think this whole idea around local is going to be more and more important as time goes on. Even if, because a lot of people in the comments will say, well, local models are going to go smaller. Even if local models go smaller, you learn the architecture for the future. You learn which jobs need a giant brain and which jobs just need a reliable worker that never sleeps. So what would I do? What's a rep I would do to learn this skill set? The first rep I would do is simple. Build a daily briefing agent for yourself. Give it three sources. Give it your calendar. Give it a folder of notes. Give it a few saved links. Its job is to tell you what matters today, what decisions are waiting for you, and what follow-ups you owe people. Then you can add one rule. You can say, it has to show sources and ask for approval before sending anything. That one project teaches you context, it teaches you retrieval, it teaches you tool use, it teaches you permissions, and it teaches you evals. It might sound small and maybe a boring first agent to build, but I believe that that is basically the shape of every serious agent inside a company. The mistake people I think make is they try to build an all-knowing agent first, these really big agent projects. The better move is basically just to build a small agent to start, get it to be very valuable, schedule it, have a clear success metric.
And the success metric might be, did it save me 10 minutes? Did it catch something I would have missed? Did it produce something I would have actually used? If the answer is yes, then you're learning the skill.
So that's the first super, super valuable skill. And by the way, this is in no particular order. These are just the six that you know, the more you know, the better. The second skill is marketers who know how to build distribution. I think this one is underrated because people confuse distribution with posting. Distribution is way more deeper than just like posting on social media. It's knowing where attention already lives, what people are already anxious about, what language they use when they describe the problem, and how to turn that into trust before you ask them to buy anything. In an AI world, we all know building products is really easy. Building demand is just getting more and more important. So when anyone can ship a landing page or an app or build a SaaS, the bottleneck moves to the question of, can you make people care? And that's where someone who's really good at distribution is a pro app. So the marketers who are gonna win in this agentic era are gonna be part researcher, they're gonna be part storyteller, they're gonna be part media operator, and they're gonna be a part community builder. They're basically gonna know how to take one insight and turn it into a tweet, a short form video, a YouTube title, a newsletter angle, a landing page headline, a founder story, and a sales conversation. So in a sense, it's almost like a marketer is becoming a generalist marketer. We're seeing this as a greater trend, like people are becoming generalists, they can do multiple things because if their job is to manage agents, they need to understand different components of that. That's exactly what's happening in the distribution marketing world. So the first rep I would do if I'm trying to learn distribution is I would do a distribution map. So I'd pick a niche I care about, like dentists using AI or solo consultants, maybe real estate agents, Shopify operators, whatever, could be a business that you want to start or you're building already. Then write down the 20 places their attention goes.