The Skills That Actually Matter in the Age of AI

I've always felt under-educated compared to my friends.

Some did masters degrees. A few did PhDs. They specialized early, went deep, and emerged with credentials that signal expertise. Meanwhile, I bounced between interests — engineering degree, then growth, then product management, then solo building. A generalist path that sometimes looked like a lack of direction.

For years, I've carried a quiet insecurity about this. The world seemed to reward specialization. "Pick a niche" was the advice everywhere. And there I was, pulled toward everything, expert in nothing.

But I wasn't wandering. I was drawn to it. Every new domain scratched the same itch: how does this work, and how does it connect to what I already know?

I'm starting to think that curiosity was doing more work than I realized.

The world is moving faster than it was even two years ago. Skills that mattered in 2023 are already being automated. And the generalist path I stumbled into? It's starting to look less like a weakness and more like preparation.

There's a framework that helped me understand why: perishable vs. non-perishable skills.

The Problem with Perishable Skills

A perishable skill is one that requires constant maintenance just to stay level — and significant time investment to improve.

Take golf. If I want to get meaningfully better, I need several hours of practice a day. Even just to maintain my current level, I need 20-30 minutes regularly. The moment I stop, my short game gets rusty and my timing drifts.

Golf doesn't compound much over time. It stacks up, demanding more and more just to keep what you have.

The same is true for a lot of technical skills:

• Specific programming languages. Syntax fades without daily use.
• Particular software tools. Replaced by new versions before you've mastered the current one.
• Platform-specific knowledge. Instagram's algorithm changed again last month, and your playbook expired with it.
• Memorized procedures. Lookup beats memorization when the lookup takes 0.3 seconds.

These skills have a half-life. And that half-life is shrinking fast.

The Lindy Effect and Skills

Nassim Taleb popularized the Lindy Effect: for non-perishable things, every additional year of survival doubles the expected remaining lifespan.

A book that's been in print for 40 years will likely be in print for another 40. Shakespeare has been relevant for 400 years, so he'll probably be relevant for 400 more.

Applied to skills: a skill that's been valuable for 100 years will likely be valuable for another 100.

Writing? Lindy. Persuasion? Lindy. Systems thinking? Lindy.

Knowing how to configure a specific CI/CD pipeline? Not Lindy. It comes and goes with the platform that hosts it.

Why This Matters Now: The AI Acceleration

AI is accelerating the decay of perishable skills while amplifying the value of non-perishable ones.

From David Epstein's Range:

"The more constrained and repetitive a challenge, the more likely it will be automated, while great rewards will accrue to those who can take conceptual knowledge from one problem or domain and apply it in an entirely new one."

The data is catching up to that idea. Dice.com found that half of all U.S. tech job postings now require AI skills. A year before that, it was 14%. And the World Economic Forum projects that 39% of core workplace skills will change by 2030.

What surprised me when I dug into those numbers: the demand isn't really for "AI skills." It's for the ability to verify, edit, and explain machine-assisted work.

The perishable skill is knowing the right prompt syntax for today's model.

The non-perishable skill is critical thinking about any tool's output. Knowing when to trust it and when to throw it away.

The Non-Perishable Skill Stack

So what are the skills that compound rather than decay? The ones that transfer across domains, survive technological shifts, and become more valuable with time?

1. Learning How to Learn

James Clear put it well in Atomic Habits:

"Knowledge compounds. Learning one new idea won't make you a genius, but a commitment to lifelong learning can be transformative... As Warren Buffett says, 'That's how knowledge works. It builds up, like compound interest.'"

The skill of rapidly acquiring new skills is the ultimate meta-skill. It makes you antifragile — every disruption becomes an opportunity to learn something new faster than the people around you.

I've noticed this compounding in my own path. Each new domain I've picked up (engineering, product, growth, SEO) becomes a lens for understanding the next one faster. Product thinking made me better at growth. Growth made me better at building products people actually want. The compounding isn't just additive. It's multiplicative.

And right now, with AI tools changing every few months, the person who can pick up a new tool in a weekend has a massive edge over the person who spent six months mastering the last one. Claude works differently than GPT works differently than Gemini. The specific knowledge decays. The ability to orient quickly doesn't.

2. Systems Thinking

Understanding how parts interact to create wholes. Seeing second and third-order effects. Recognizing feedback loops before they spiral.

This applies whether you're designing software, managing a team, building a business, or figuring out why your sleep is terrible. The domain changes. The thinking transfers.

Here's where it gets concrete. Say you're building a product and someone proposes adding a new feature. First-order thinking says: "Great, more value for users." Second-order thinking asks: what does this do to our support load? If support tickets increase 30%, that means hiring another support person within two months. That changes our burn rate, which changes our runway, which changes when we need to raise again and at what terms. One feature decision just became a financing decision.

Systems thinkers catch that chain before the feature ships. Everyone else catches it when the bank account shrinks.

This is also why spinning up 10 AI agents doesn't give you 10x output. Agents create work for each other: conflicting outputs, duplicated effort, context that needs to be merged and reconciled. Without someone who understands how those pieces interact, more agents just means more mess. The bottleneck isn't compute. It's coordination. That's a systems problem.

3. Communication and Persuasion

Writing clearly. Speaking persuasively. Telling stories that move people.

These skills are thousands of years old. They'll be valuable for thousands more. But the game is shifting.

AI handles the mechanical side of communication now. It can draft emails, summarize meetings, write documentation. The commodity version of "good writing" is already free. What AI can't do is have a genuine opinion. It can't tell a story from your actual life. It can't develop a voice that's distinctly yours, one that readers recognize.

The first wave of AI-generated content was obvious and bad. It's getting better, and a lot of people assume that problem goes away when the models continue to improve. They're probably right, the slop will get more polished. But that actually makes voice and authenticity more important, not less. When anyone can produce clean, competent prose, the only differentiator is whether you have something real to say and a way of saying it that's yours alone.

The bar for "can write" drops to zero. The bar for "worth reading" goes up.

4. Judgment and Taste

The ability to assess opportunities. To know what's worth pursuing and what isn't. To recognize quality before the crowd does.

David Epstein found that the best forecasters shared one trait: "genuinely curious about, well, really everything." Broad exposure builds judgment. Narrow expertise doesn't.

I think judgment is the skill that AI makes simultaneously more valuable and harder to outsource. When an AI agent runs overnight and produces a 40-page analysis, someone has to decide: is this right? Is it useful? Does it match reality? Is it what people want? That judgment call requires domain knowledge, pattern recognition, and the willingness to throw away work that looks impressive but isn't sound.

The more AI produces, the more judgment you need to sort through it. The person who can look at a piece of AI output and immediately spot what's off? That person becomes the bottleneck in the best way.

5. Working with People

Collaboration. Negotiation. Building trust. Managing conflict. Reading a room.

Every job involves people. These skills don't expire. But I think the framing is about to change.

AI agents are getting better at collaborative tasks. They can coordinate projects, manage handoffs, handle routine back-and-forth. As more of the transactional side of teamwork gets automated, human relationships become scarce. And scarcity drives value.

The skill here isn't "managing a team" in the old sense. It's being someone people genuinely want to work with. Someone they want in the room, on the call, at the dinner. When AI can handle the logistics of collaboration, what's left is the human part: the trust, the shared experiences, the connection that makes someone pick up the phone when things go sideways.

I think human relationships will be valued for their own sake in a way we haven't seen since before Slack replaced hallway conversations.

It's always been true, but it will be more than ever in the future.

6. Integrating Across Domains

Epstein's research found that people with breadth across many areas plus depth in at least one outperformed both pure specialists and pure generalists.

"Our greatest strength is the exact opposite of narrow specialization. It is the ability to integrate broadly."

The person who can connect insights from biology to business, or from construction to software architecture, creates value that specialists can't. Not because specialists lack intelligence. They lack the adjacent reference points that make unusual solutions visible.

The Counterintuitive Part

When everything accelerates, the instinct is to chase the acceleration. Learn the new framework. Master the new tool. Get certified in the new platform.

That might be necessary, but it's a treadmill. You're running to stay in place, and the belt keeps getting faster.

The counterintuitive move is to master the skills that sit above the acceleration: learning quickly, adapting to change, forming real relationships, exercising judgment under uncertainty. These aren't "old skills" you're retreating to. They're the stable platform that lets you navigate whatever comes next.

The person who can pick up any new tool in a week, think through its second-order effects, communicate clearly about what they've found, and bring people along with them? That person doesn't need to predict which technology wins. They're positioned for all of them.

Read books that have been relevant for decades, not just the latest releases. Practice writing and speaking, not just prompting. Build relationships, not just technical knowledge. Develop judgment through wide exposure, not just deep expertise.

As Taleb might say: focus on what's Lindy.

Auditing Your Own Skill Portfolio

Here's an exercise worth 20 minutes: list the top 10 skills you've invested time in over the past year. For each one, ask:

1. Does this skill require constant maintenance? (Perishable indicator)
2. Would this skill transfer to a completely different domain? (Transferability)
3. Was this skill valuable 20 years ago? Will it be valuable in 20 years? (Lindy test)
4. Does this skill compound over time, or just maintain? (Compounding potential)

You'll probably find a mix. That's fine. The goal isn't to eliminate all perishable skills, you need them for your current work. The goal is to make sure you're also building the non-perishable foundation underneath.

The Bottom Line

Epstein calls them "wicked" environments, where the rules are unclear, feedback is delayed, and the patterns keep shifting. In kind environments (like chess), narrow expertise wins. In wicked environments (like most of modern work), breadth wins.

"Breadth of training predicts breadth of transfer... the more contexts in which something is learned, the more the learner creates abstract models, and the less they rely on any particular example."

The skills that will matter most in the age of AI are the ones that have always mattered: thinking clearly, communicating well, learning quickly, and connecting ideas across boundaries.

These skills don't decay. They compound.

Traditional jobs don't always reward this kind of breadth. Most companies want specialists who fit neatly into defined roles. That used to bother me. Now I see it differently.

If you're a generalist, if you love learning, if you're pulled toward everything, if you've never quite fit into a single specialization, the path forward might not be finding the right job. It might be building your own thing.

That's the bet I'm making. And for the first time, the lack of specialization feels less like a liability and more like exactly what the moment requires.