How to get ahead in the AI era

One post tells you software engineers are dead. The next is a guy on X claiming his AI app is making him rich.

Then Andrej Karpathy writes:

“this gap in shared understanding of LLM capability is widening.”

His October 11 post ranges from casual chatbot use to researchers with access to internal systems most of us can't test.

And you're trying to figure out what to learn, what to build, and whether any of this is actually going to change your life.

I'm writing this for people building with AI. If you code, if you don't, or if you're somewhere in between, this is for you.

Karpathy's point is that the gap comes from access, from what people attempt, and from where the models are strong. That got me thinking, because most of us spend hours arguing about AI and barely test what it can do for our own work.

I'm figuring this out too. I pay $200 a month for Claude, use Claude Code every day, and still hit usage limits. Some of my AI experiments are still side projects. But they've changed what I think I can build.

So here's the whole post in one sentence: use AI until you know what it can and can't do. Everything below is a way to do that over the next 30 days. Test it on your own work, expand what you can build, finish something useful, and make that work visible.

1. Find out what AI can do for your work

I've talked with nontechnical people trying to build websites with AI, and I've seen some rough results. I get why someone looks at that first attempt and decides the whole thing is overhyped.

But I've seen enough to know better results are possible. The model, the instructions, the context you give it, and the time you spend improving the result all change the experience. Access matters too. Someone who can keep working through a hard problem has a different experience from someone who hits a limit halfway. More usage gives you more room to work, not better results, so if your budget is the basic plan, pick a smaller task and investigate it properly.

Start with a task you understand well enough to judge. If you're a developer, that might be adding saved searches to an app you already have. Give the agent the relevant code, describe the behavior you expect, and test it on a device. Then change a requirement and see how it handles the revision. If you don't code, pick something you do by hand every week and ask the agent to automate it.

Write down how long it took to get a result you trust, including your review and fixes, and where the tool needed help. That's how you learn which work you can delegate and which decisions still need you. Make the experiment small enough to finish in the time you actually have.

There's a bonus here. Once you know the limits firsthand, you can tell when someone is overselling AI and when a product is a thin wrapper around a model. That alone will save you money.

Your first advantage is firsthand evidence. Pick one task, define what good looks like, and find out how much of it you can get done with the tools you have.

2. Expand what you can build

I worked as a software engineer at Walmart using React Native. Later, at Expo, I worked with developers and the tools they use to build apps every day. React Native and Expo became the way I understood mobile development, because they were what I used.

That gave me a lot. It also shaped what I was willing to try.

Here's what surprised me when I finally started building with SwiftUI and C++ this year. They look a lot like JavaScript and TypeScript. Reading code is easier than writing it, so if you already know one language, understanding another is faster than you'd expect. And honestly, I haven't written a single line of SwiftUI or C++ by hand. I prompt the agent, and I read the code when I want to understand a decision, which is less often than I thought it would be.

I still use React Native and Expo. I'm also enjoying having more options. What I know about navigation, data, device capabilities, and user experience carries over, and AI fills in the parts I'm missing while I learn them.

I've felt this outside app development too. While working on a little display for my desk, I downloaded Blender and asked AI to design a case for it. It did. One small experiment opened up a whole other part of making a product.

Bigger teams are retesting their assumptions too. Shopify is moving its mobile apps from React Native to Swift and Kotlin because coding agents changed the cost of maintaining separate codebases, and DHH showed six native HEY apps about a week after starting them. I shared my take on the Shopify decision in this post. The signal isn't that one stack won. It's that the cost of trying something changed, and Shopify was clear that native expertise still mattered.

Pick one constraint you've accepted and test it. Rebuild a familiar screen in a different framework, try a device API you've avoided, or automate a task you do by hand. Write down what carried over and what you still need to learn.

3. Turn an experiment into something someone can use

Starting a project doesn't mean signing away the next six months of your life.

Build the weird app. Try the idea that sounds ridiculous. Start five tiny experiments in a day if you want. Two hours later, you can decide it's going nowhere and delete it.

You still learned something: what AI can do, where it breaks, or what you actually enjoy building. That might become useful on a completely different project a year from now. So give your ridiculous ideas a couple of hours before you talk yourself out of them.

Give each experiment a question to answer. Is this interaction possible? Does this automation actually save me time? Do I enjoy working in this stack?

When one of them is promising, take it through one complete workflow that another person can use.

Inkigo, my AI tattoo preview app, is my own example of this. It's a real app in the App Store with real customers, doing around $650 a month. It's small, but it's the loop working end to end, and I wrote about how it got there.

This is the Builder's Loop I keep coming back to: idea, understand, build, ship, earn, iterate. AI helps with the work at every step. You still choose the problem, check the result, and learn from the person using it.

So put your first version in someone's hands. Watch whether they understand it, whether they come back, and what they ask for. Their behavior shows you work the demo didn't. Use it to improve the app, change the idea, or stop. If they keep using it and the problem matters, you have a reason to find out whether others would pay.

Finish one useful workflow and let someone test it. An internal tool, a small app, or a reliable automation all count. Make one improvement based on what you see.

4. Give people evidence of what you can do

Some of the clearest AI success stories I see are creators teaching people how to use the tools. Part of that is visibility, because creators show up in my feed and someone quietly improving their work with AI is easy to miss.

Still, there's a real lesson there. When people are confused, a clear explanation is valuable. A tutorial that feels basic to you might be exactly what someone else needs. You already know which tool to open, what to give it, and how to recognize a bad result. A beginner may be missing all three.

Naval has a line that connects building with communicating:

“Learn to sell, learn to build, if you can do both, you will be unstoppable.”

I've lived this with Code with Beto. I started sharing what I was learning about React Native while preparing for interviews. That work opened doors at Walmart and later at Expo, and teaching became a much bigger part of my life than I expected.

You already have material from the previous three sections. Show the screen you rebuilt. Explain where the agent got stuck. Demo the workflow someone is now using and what you changed after their feedback. Include the task, the tool, the context you gave it, the result, and the part that still needed your judgment. If you only tested it for an afternoon, say so. People can tell, and honesty is what makes them trust the next thing you share.

Write a short post, record a demo, or share it with one community. You can start by helping one colleague. Pay attention to the questions people ask, because they usually point at the next problem worth solving.

Over time this gives people reasons to trust your judgment, hire you, collaborate with you, or try something you build. You become easier to recommend because they've seen your work.

Make one useful result easy to find and understand. Show what you did, what you learned, and invite someone with the same problem to try it.

Your next 30 days

Use one problem for the whole month so each week builds on the last. Keep it small enough to finish in the time you actually have.

At the end of the month, look at the evidence: what you can build now, what you can delegate reliably, what someone found useful, and what you can show for it.

That's the whole idea. Use AI until you know what it can and can't do. Nobody can hand you that understanding, and once you have it, it compounds. You'll build faster, you'll stop paying for products that are just a wrapper, and you'll be able to tell what's real the next time a post says engineers are dead.

Start this week. Pick one thing you do by hand and automate it for yourself. That's how you get ahead of this.

If you want a map of what these models can actually do before you start, I put together a free LLM learning roadmap. And if you want to take an idea all the way to the App Store, my book walks through the whole lifecycle, whether you code or not.

From idea to the storeFrom Idea to App Store with AI Coding Agents

The mental model and the full lifecycle of shipping a mobile app with an AI coding agent, including how to get through App Review.

Let's keep building.

Beto