How to Become Successful and Beautiful in the AI Era

A few things that seem increasingly important when intelligence and execution are becoming cheap.

The obvious advice for surviving the AI transition is to learn how to use AI. I think we’re already past that.

Using ChatGPT is rapidly becoming the equivalent of knowing how to use Google. Useful, certainly. Competitive advantage, not really. The more interesting skill is learning how to build systems out of AI.

A year ago, most of my AI work meant talking to one agent in one chat. Today, a normal task can mean hours of autonomous work, dozens of spawned subagents, different models doing different jobs, verification loops, tools and workflows. I don’t necessarily know what every subagent is doing, nor should I. It doesn’t work perfectly on the first attempt. But once it works repeatedly, it starts looking suspiciously like a superpower.

Automate Production, Not Thinking

Anything involving information production should increasingly be delegated. Research preparation. Documents. Code. Tables. Administrative work. Data processing. Routine communication. There is little nobility in manually spending three hours formatting something a machine can produce in three minutes. But there’s an important boundary.

Don’t outsource goals and quality judgement together with execution.

If AI decides what should be done, does it, evaluates itself and then tells you everything went wonderfully, congratulations: you’ve built a very efficient slop machine. The human job moves upward.

Your Time Is More Expensive Than Tokens

I increasingly don’t care whether an agent uses $0.30 or $3 worth of tokens if it saves me two hours. Tokens keep getting cheaper. Human time doesn’t. Automation for the sake of automation is mostly an entertaining engineering hobby. But turning an eight-hour workflow into four hours has enormous value.

For one person, that’s four extra hours of life. For another, it’s twice as many experiments, products or client problems solved by the same company. Both are perfectly reasonable uses of technology.

Be Broad, but Be Deep Somewhere

AI makes being a generalist dramatically more powerful. An investor can suddenly write small tools, analyze datasets, prototype products and conduct research. An engineer can work on design, market research, sales and finance. A designer can build functioning software. Wonderful.

But this only works if there is still something you understand better than the machine’s default answer.

Otherwise, how do you know whether the result is good? So I’d optimize for a slightly strange combination:

deep expertise in one domain + AI-assisted competence across neighboring domains.

You need a home territory from which to invade everything else.

Choosing the Problem Becomes the Bottleneck

Historically, execution was expensive. You needed employees. Capital. Specialists. Months of development. So organizations spent enormous effort managing scarce execution capacity.

AI changes that equation. One person can increasingly produce the output of what used to require a small team. Which makes a different resource scarce:

knowing what is worth doing.

What problem matters? Where is the money? Why now? Which hypothesis should we test? What’s the cheapest way to discover whether we’re wrong? How do we measure success?

AI can produce 100 plausible answers to almost anything. Generating option number 101 isn’t particularly valuable.

Judgement Is the New Scarcity

The valuable skill is increasingly judgement.

Which answer is actually correct? Where is the hidden assumption? What sounds impressive but is banal? Which solution works generally but fails in this particular situation? What did the model misunderstand? What shouldn’t we build at all?

We spent decades rewarding people for knowing answers. We’re moving toward rewarding people who can evaluate answers. And that leads directly to another increasingly expensive human capability.

Taste

You can now generate 100 logos before lunch. Or 100 illustrations. 100 product concepts. 100 headlines. 100 songs. 100 interface designs. 100 business strategies. Therefore, producing another variation is rapidly approaching zero economic value.

The scarce part is being able to look at those 100 outputs and say:

This one. Delete the other 99.

That’s taste. And unlike generation, taste doesn’t automatically become better because the model got another trillion parameters. It comes from seeing thousands of examples, making decisions, being wrong, developing standards and actually practicing something long enough to understand what good looks like.

Be Good at Something

This may become more important for reasons that have nothing to do with economics. Be a master of something. Running a hedge fund. Cooking. Mathematics. Painting. Writing software. Dancing on TikTok. Doesn’t particularly matter.

People have always disproportionately valued the best practitioners. AI probably makes this power-law distribution even more extreme because mediocre production becomes almost free. But mastery also does something else. It creates taste, patience, identity and a relationship with difficulty.

That may become psychologically important in a world where economic usefulness is gradually becoming a worse foundation for human self-worth.

Your job doesn’t need to be the only evidence that you deserve to exist. Probably shouldn’t have been anyway.

And Occasionally Make Life Harder on Purpose

Technology has spent the last century systematically removing friction. AI takes this to a rather absurd conclusion. You don’t need to remember. Search. Write. Calculate. Navigate. Plan. Wait. Sometimes you barely need to formulate the question. This is mostly fantastic. But some useful human qualities are produced precisely by friction.

So I think we’ll increasingly need to create some deliberately. Write something yourself occasionally. Read the difficult paper instead of its summary. Draw badly. Cook dinner from ingredients. Learn something without asking Claude every thirty seconds. Walk somewhere without navigation. Spend several hours without your phone. Not because doing things inefficiently is morally superior. It isn’t. But because the easiest possible life and the best possible human aren’t necessarily the same optimization problem.

That, I think, is the strange bargain of the AI era. Automate aggressively. Build systems. Save your time. Expand what you can do. But use the resulting leverage to become better at choosing, judging and mastering things, rather than merely becoming the person who presses Enter while increasingly elaborate machines do everything else.

The machines are getting very good at execution. We should probably spend the saved time becoming good at the parts that remain ours.