Skills that stay relevant beyond automation and AI

We keep asking which jobs AI will take. I think the more useful question is what’s already happening to the work we do every day.

A developer can generate in minutes the boilerplate that used to eat an afternoon. An analyst gets a first-draft report without ever opening a spreadsheet. A manager walks out of a meeting with the summary already written. I wouldn’t give any of that back.

Still, something shifts when making things gets cheap. Code isn’t impressive just because someone wrote it, and a report doesn’t earn trust by looking polished. What counts is whether the code solves the right problem and whether anyone can actually decide something from the report.

That changes how I think about careers. I’ve seen the advice to drop the technical side and lean into empathy and communication, and it sounds a bit too convenient, usually coming from people who’d rather not learn the technology. You need both, and you need to see how they connect.

The classic career loop goes like this: learn a technology, get certified, build experience, land a better role, repeat. It worked well for a lot of people, and in some fields it still does. I just wouldn’t lean on it as hard as I did ten years ago.

The World Economic Forum’s Future of Jobs Report 2025 says employers expect 39% of workers’ existing skills to change or become outdated by 2030. That’s what employers expect, not a prediction that four in ten skills will suddenly stop being useful. Even so, professional knowledge clearly spoils faster than most of us would like.¹

The usual response bothers me, though. It’s an endless chorus of “learn the next thing.” Another certification, another framework, another AI assistant, another course that promises to make you future-ready. You can spend months preparing for work without getting any better at it.

I’d rather get good at picking up something unfamiliar, working out where it fits, and deciding if it’s worth using at all. It’s a less exciting goal than mastering whatever’s trending, but it holds up better.

Say an AI assistant hands you a database migration. The code looks fine, the explanation is clear, and maybe the tests pass. Do you deploy it?

If you’ve run production systems, questions start piling up. What happens to the existing data? Can you roll it back? How long does it lock the table? What if the deployment dies halfway? Is the test environment close enough to production to tell you anything?

The AI might help answer those, and it might even catch something you missed. But you need enough technical grounding to know which questions to ask in the first place.

That’s the gap between knowledge and judgment for me. Knowledge helps you find an answer. Judgment tells you whether to trust it.

Neither comes automatically. Experienced people make bad calls all the time, and a newcomer sometimes spots the problem precisely because they aren’t attached to how things have always been done. Experience only helps if it keeps you curious instead of comfortable.

Non-technical work is no different. AI can summarise a contract, but somebody has to notice what the summary left out. It can produce a market analysis, but someone who knows the business should push back when the conclusions don’t match what’s happening on the ground.

Often the most valuable moment comes after the AI is done.

Tech has always churned. Languages, platforms and whole architectural styles have risen and faded. What’s new is how fast and how easy the churn has become.

You don’t need weeks of groundwork before trying a tool anymore, because an assistant can get you going in an hour. That makes curiosity easier to act on. It also makes shallow knowledge easy to mistake for expertise.

Picture someone who gets very good at prompting one particular AI product, then finds the next product handles the same job differently. Some of the prompts will still work, but that edge won’t last by itself.

The WEF report puts analytical thinking at the top of its core skills list, with resilience, flexibility, leadership and creative thinking close behind. Employers aren’t saying technical knowledge has stopped mattering, either. Technological literacy and AI-related skills are among the areas they expect to grow.

To me that’s a reason to reinforce what sits underneath your technical skills. Learn the new platform, sure, but learn the principles behind it too. Find out where it breaks. Know enough about the business around it to tell when the elegant technical answer is the wrong one.

Someone who understands systems can switch tools easily. Someone who only knows which buttons to press is stuck when the buttons move.

There’s a odd comfort in enrolling in another course. You get a plan, a deadline, a certificate at the end, and progress you can measure. But finishing a course and getting better at your job are different things.

I prefer learning that leaves you with something usable. Build a small app in a framework you don’t know. Dig into a problem you’ve never faced. Take a task you know cold, see if AI can improve it, and then go through carefully what it got wrong.

Those mistakes are usually more interesting than the successes, since they show where your own understanding has holes.

Demis Hassabis, CEO of Google DeepMind, has talked about “learning how to learn” as a key skill for the next generation.²

I agree, though I’d hate to see it end up on a motivational poster. Learning well includes knowing when to stop. You don’t have to follow every new model or platform. You have to spot the changes that touch your work and spend your attention there.

For a software engineer at a bank, that probably means understanding what AI does to code quality, security and data handling. It probably doesn’t mean spending every weekend reading about the latest model architecture. The same goes for people in finance, marketing, healthcare or management. Your field still matters, and AI literacy should add to your expertise, not stand in for it.

People love to say communication will stay important in the AI era. True, but we make it sound far too pleasant.

Good communication isn’t only a clear presentation or a persuasive email. Sometimes it means telling a stakeholder their preferred solution won’t work. Sometimes it means admitting your team got a decision wrong. Sometimes it’s asking the question that drags out a disagreement nobody wanted to have.Privacy is not an option, and it shouldn’t be the price we accept for just getting on the Internet.

Privacy is not an option, and it shouldn’t be the price we accept for just getting on the Internet.

The most important skills to have in life are gained through interpersonal experiences and the liberal arts.

Frank Elavsky³

Those moments are uncomfortable, and no amount of polished AI text makes them go away.

Take a project where engineering wants a technically excellent solution and the business needs something simpler by next month. Nobody’s incompetent here. The two sides are just optimising for different things. Someone has to put those differences on the table, lay out the trade-offs, and help everyone decide. That takes technical knowledge, but it also takes listening and a willingness to challenge assumptions.

Calling communication a “soft skill” has never sat right with me, as if it were less serious than engineering. Bad communication creates expensive technical problems, and good communication heads them off.

Automation usually gets introduced to fix one piece of a process, and work rarely stays in one piece.

A team automates deployments and finds testing is now the bottleneck. A company rolls out AI-written customer replies and discovers that hard cases are harder to escalate. A department speeds itself up and quietly creates extra work for everyone nearby. The local win looks great while the overall result disappoints.

That’s why I value systems thinking. You look past the task in front of you and ask what else moves when you improve it. That means understanding dependencies, incentives and consequences that no single dashboard shows you.

In software architecture, the knock-on effects land on security, reliability, running costs and the people who maintain the system. In management, they might be incentives that push people to optimise the wrong metric.

AI can help map relationships and run through scenarios. Someone still has to judge which relationships matter and whether the assumptions behind the analysis are realistic.

If your current strength is…	I’d also work on…	Why
Software development	Architecture, debugging and security	Generated code still needs someone who understands its consequences
Data analysis	Business context and critical interpretation	A correct calculation can still answer the wrong question
Project management	Conflict resolution and technical literacy	Plans become difficult when priorities collide
Cloud engineering	Cost, reliability and organisational constraints	Infrastructure decisions affect more than deployment
AI tool usage	Verification, domain knowledge and data privacy	Fluency with a tool doesn’t guarantee sound results

None of these are magic formulas. They’re just pairings where one skill makes the other more useful.

I sometimes hear that professionals should focus on empathy, creativity and communication because AI will handle the technical side. I’d ignore that advice.

You don’t need to specialise in every technology. But you should know enough to ask sharp questions, spot limitations and judge whether a result is any good.

The OECD’s 2026 report on AI and skills draws a similar line between advanced AI expertise and broader digital competence. Most workers won’t build AI models, but many will need to use digital tools, interpret data and solve problems in workplaces where AI keeps showing up.⁴

That seems a more realistic target than becoming an AI expert just because it’s fashionable. A lawyer doesn’t have to build a language model, but knowing its limits could change how they work. A developer doesn’t need every detail of model training, but should understand the risks of generated code and what it means to send data to external AI services.

Technical competence gives you somewhere to stand. Judgment helps you decide where to go.

I wouldn’t begin by asking which skill is “future-proof”? That question assumes we can predict the future of work far better than we can.

I’d start with my current work instead. Where do I lean too much on routine? Where are my decisions weakest? Which problems do I keep avoiding because I don’t understand them well enough?

The answer might be a technical course. It might be learning to analyse data, getting better at explaining hard decisions, or spending time in a different part of the business. It could even be learning to use AI better in work you already know well. Whatever it is, it should come from a real gap, not from fear of missing the next trend.

No skill is guaranteed to stay valuable forever. Even judgment can be helped along by technology, and some tasks that now take years of experience will get easier to automate. That’s worth taking seriously.

But building a career around one tool, one framework or one narrow workflow looks more fragile every year. I’d rather be someone who knows their field, can learn unfamiliar things without much hand-holding, and still questions an answer that sounds convincing. That won’t make anyone immune to change, but it gives you more ways to respond when the work shifts.

Sources

1. WEforum, “The Future of Jobs Report 2025” ↩︎
2. apnews, “Google’s top AI scientist says ‘learning how to learn’ will be next generation’s most needed skill” ↩︎
3. Pewresearch, “The Future of Jobs and Jobs Training” ↩︎
4. OECD, “AI and skills” ↩︎