The Fastest Way To Learn LLMs Is To Stop Reading About Them And Start Building Toys
Most leaders approach AI like a strategy course. Six months later they can talk about AI competently and cannot do anything with it. That gap is starting to matter.
I have watched a lot of senior leaders try to learn AI. Most of them approach it like it is a strategy course. They read the McKinsey report. They listen to two podcasts. They sit through a vendor demo. Six months later they can talk about AI competently and they cannot do anything with it. That gap is starting to matter.
Naval put it well on the nav.al/code podcast. He called vibe coding the unbounded video game. You sit down, you have an idea, you tell the LLM what you want, you get something working in an hour. No team to align. No sprint to plan. No JIRA. Just you and the model and the thing you wanted to make. He is right that this is the most fun he has had with software in twenty years and he is right that this is the way in.
Here is the framing I give every executive who tells me they want to get fluent. Stop trying to learn AI. Start trying to build a toy.
Pick something tiny. A script that summarizes your inbox each morning. A small dashboard that pulls your team's status updates and flags the ones that need follow up. A bot that drafts your weekly note based on your calendar. None of these are going to ship to a customer. That is the point. The toy is the textbook.
The first hour is going to feel stupid. You will not know what a virtual environment is. You will copy paste error messages into Claude and feel like a fraud. Do it anyway. By hour three you will have something that works. By hour ten you will start to feel which prompts the model handles well and which it gets lost on. That feel does not transfer through a slide deck. You only get it from your hands on the keyboard.
Naval makes another point that is worth repeating. He says the command line is back. For twenty years the industry pushed everything toward graphical interfaces because typing was hard. LLMs make typing the most powerful interface again because language is what they speak. If you can describe what you want clearly, you can drive a system. The leaders who get comfortable describing things to machines in plain English are the ones who will move fastest in the next five years.
A few rules for your first ten toys.
One. Build for yourself, not for your team. Personal pain teaches faster than corporate pain.
Two. Ship in a weekend or kill it. If you cannot get something usable in two days, the scope is wrong.
Three. Show your work. Every toy you build, demo it to one person. The act of demoing locks the learning in.
The leaders who treat AI as a reading list are going to spend the next three years feeling vaguely behind. The ones who treat it as a hobby workshop are going to look around in 2027 and realize they are the most fluent person in any room they walk into. That is a real edge and it costs you a few weekends.
Sources: Naval Ravikant on nav.al/code podcast.