Nobody can tell you exactly which jobs will look the same in five years. Anyone who claims to know is guessing.
What we do have is evidence. We know which skills are gaining value as AI spreads through offices, and which habits quietly wear people down. The World Economic Forum’s Future of Jobs Report 2025 surveyed more than 1,000 employers representing over 14 million workers across 55 economies. Those employers expect about 39% of workers’ existing skill sets to be transformed or become outdated between 2025 and 2030. That’s a lot of change. It’s also a lot of room for anyone willing to move first.
The 15 skills below come from that kind of evidence: job-ad analyses covering more than a billion postings, field experiments inside real companies, and a couple of cautionary tales. A few are about using AI well. Most are about the things AI still does badly, or the things it can make you worse at if you’re careless with it.
1. Get fluent with the AI tools in your own field
Start with the obvious one, because the numbers behind it are hard to ignore. PwC’s 2026 Global AI Jobs Barometer, released in June 2026, analysed more than one billion job ads across 27 countries and territories. Here’s what it found:
| PwC 2026 Global AI Jobs Barometer | Figure |
|---|---|
| Growth in jobs requiring specific AI skills | 69% |
| Growth in the overall jobs market | 9% |
| Average wage premium for AI skills (2026 edition) | 62% |
| Same premium in the previous edition | 57% |
The premium swings a lot by sector, from as high as 118% in consumer markets to 16% in government and public sector work. Keep in mind what these figures are. They describe what employers ask for and pay for in job ads. They don’t promise that one online course will lift your salary by 62%.
Fluency doesn’t mean becoming an engineer. It means knowing which tasks in your job AI handles well, which ones it fumbles, and how to get from a vague request to something you’d actually use.
Give it a go: pick one recurring task this week, like a meeting summary or a first draft, and run it through an AI tool three times with three different sets of instructions. Write down what changed and which version you’d send.
2. Think critically about every confident answer
AI rarely sounds unsure. That’s a problem, because people tend to borrow its confidence.
A team from Microsoft Research and Carnegie Mellon University, led by Hao-Ping Lee, surveyed 319 knowledge workers who use generative AI at work and collected 936 real examples of AI-assisted tasks. In the study, presented at CHI 2025 in April 2025, higher confidence in the AI went along with less critical thinking. Workers who were more confident in their own abilities did the opposite and engaged more. One caution: this is a self-reported survey, so it shows an association, not proof that AI weakens anyone’s mind.
Still, the pattern is worth noticing in yourself. The moment you stop wondering whether an answer is right is the moment you’ve handed over the part of the job that was yours. Sharpening your reasoning skills is the long fix.
Next time, do this: before you accept an AI answer on anything that matters, ask yourself what would have to be true for it to be wrong. Then spend two minutes looking for exactly that.
3. Treat quality control as your job, not the tool’s
Workers already sense this shift. Microsoft’s 2026 Work Trend Index surveyed 20,000 knowledge workers who use AI, across 10 markets, between February and April 2026. Asked which human skills matter more as AI takes on work, 50% picked quality control of AI output, the top answer. Critical thinking came second at 46%. And 86% said they treat AI output as a starting point and stay responsible for the thinking.
Saying it is easier than doing it. In July 2025, Deloitte Australia delivered a 237-page report to the country’s Department of Employment and Workplace Relations, under a contract worth about A$440,000. Sydney University researcher Chris Rudge then found fabricated references and a made-up quote attributed to a Federal Court judgment. The revised version disclosed that a generative AI tool had been used, and in October 2025 Deloitte agreed to a partial refund. The department said the report’s recommendations didn’t change. The reputational damage was done anyway.
Quality control is mostly unglamorous attention to detail at work. That’s exactly why people skip it.
Start small: make a personal rule that every number, name, quote and citation in AI-assisted work gets traced to a real source before it leaves your hands.
4. Go deep in your field
Here’s the chain of cause and effect, step by step.
Generative AI learns from huge amounts of existing work, so it’s good at reproducing what competent people usually do. That lifts the floor. When the floor rises, average work gets cheap, and the value moves to people who can tell when the usual answer is wrong for this client, this law, this patient or this codebase. That judgment comes from depth.
There’s solid evidence for the first link. Erik Brynjolfsson, Danielle Li and Lindsey Raymond studied the staggered rollout of an AI assistant to 5,179 customer support agents. In the working-paper version, access to the tool raised issues resolved per hour by 14% on average and by 34% for novice and lower-skilled agents, with minimal effect on the most experienced. The authors suggest the tool spread the habits of the best agents to everyone else. The peer-reviewed version, published in the Quarterly Journal of Economics in 2025, keeps the same pattern. The productivity effect is well identified. The career conclusion, that depth becomes the differentiator, is an inference from it, so treat it as a reasoned bet rather than a measured fact.
Depth also needs fuel, and that fuel is staying curious as an adult long after school stops forcing you to learn.
Your move: choose one narrow corner of your field and read primary material on it this month. That might be the regulation itself, the original study or the technical standard. Summaries don’t count.
5. Learn in a way that keeps the knowledge in your head
AI can make you look competent while you learn very little. That’s the uncomfortable finding from a field experiment led by Hamsa Bastani at the Wharton School, published in PNAS in June 2025.
Nearly 1,000 high school maths students in Turkey practised with either a standard ChatGPT-style tool, a version with tutoring guardrails that gave hints instead of answers, or no AI at all. The standard tool helped during practice. Once access was removed, those students scored about 17% lower on exams than students who never had AI. The guardrail version largely avoided that harm. These were teenagers studying maths in one school, not professionals, so take the exact numbers with care. The mechanism is easy to recognise in adult work, though: if the tool does the thinking, the skill never forms.
The basics of improving your memory, like retrieval practice and spacing, still do the heavy lifting.
A quick experiment: next time you learn something new at work, struggle with it alone for ten minutes first. Then ask AI only for a hint, never for the full answer.
6. Communicate clearly with people who lack your context
Technology keeps spreading, and the ability to explain things keeps rising in value right alongside it. LinkedIn’s Skills on the Rise 2026 list, published in February 2026, ranks skills by year-over-year growth in two measures: skills members added to their profiles, and skills held by members who got hired. The data compares December 2024 to November 2025 with the year before, across 12 markets. Executive and stakeholder communication, public speaking and cross-functional coordination are rising, next to technical skills like prompt engineering and large language models. LinkedIn’s own summary stresses clear communication during uncertainty.
This makes practical sense. Briefing a colleague and briefing an AI tool use the same muscle: state the goal, give the context, name the constraints, say what done looks like. People who do that well get better results from both. If you want a warm-up, the habits behind having a good conversation with anyone carry straight over to work updates.
Test it this week: rewrite one status update using three lines only. What changed, why it matters to the reader, and what you need from them.
7. Protect your original thinking
AI makes you a bit more creative and everyone a lot more similar. That’s roughly the picture from a meta-analysis by Niklas Holzner, Sebastian Maier and Stefan Feuerriegel, posted in May 2025.
They pooled 28 studies with 8,214 participants. AI on its own scored about the same as humans on creativity tasks. People working with AI did somewhat better than people working alone, a small effect. But the diversity of ideas dropped sharply when people collaborated with AI. Two caveats: it’s a preprint that hadn’t been peer-reviewed when I checked, and the diversity finding rests on a small number of studies, mostly text tasks with GPT-4.
Think about what that means for a career. If your whole team brainstorms with the same tools, the “good” ideas will start to look alike. The person who brings the odd angle becomes valuable precisely because it’s rare. Thinking more laterally is a trainable habit.
Here’s a simple way in: write down your own five ideas before you open any AI tool. Then ask the AI for ideas that are deliberately different from yours, and keep the strangest one on the list.
8. Get good at the human moments people remember
Klarna’s customer service story is a useful reality check. In 2024 the company said its AI assistant was handling two-thirds of customer service chats and doing the work of about 700 agents. In May 2025, CEO Sebastian Siemiatkowski told Bloomberg that cost had been too dominant a factor in the rollout and that the result was lower quality. Klarna began recruiting human agents again so customers could always reach a person. The company also told Fortune it was still very much AI-first, so this was a rebalancing, not a retreat.
The lesson isn’t that AI failed. Routine questions went to the bot just fine. The friction showed up in the moments that needed patience, reassurance and a bit of judgment, which are the moments customers remember and tell friends about. That’s where building your emotional intelligence pays off, especially in client-facing roles. Pay attention to the next frustrated person you deal with at work and notice what actually calms them down, because that’s the part of your job no chatbot has nailed yet.
9. Learn how other functions think
A pre-registered field experiment at Procter & Gamble, led by Fabrizio Dell’Acqua with researchers from Harvard, Wharton and ESSEC, gave 776 professionals real product-innovation problems. They worked alone or in pairs, with or without AI. The working paper came out in April 2025 and was later published in Organization Science.
Individuals using AI matched the performance of two-person teams working without it. Without AI, R&D people proposed more technical solutions and commercial people more market-focused ones. With AI, both produced more balanced proposals regardless of background. Participants using AI also reported more positive emotions while working. All of this comes from one company and one type of task.
So if a solo person with AI can match a pair, what still sets you apart on a team? Partly, it’s knowing how the other functions think, so you can spot where the AI’s tidy, balanced answer misses something real. Good social skills are what get you invited into those rooms.
Give it a go: book thirty minutes with someone from a function you rarely work with. Ask what they wish your team understood about their work.
10. Build resilience for constant change
The WEF’s Future of Jobs Report 2025 doesn’t only track technical skills. Alongside AI and big data, employers expect resilience, flexibility and agility to rise in importance by 2030, together with curiosity and lifelong learning. Analytical thinking remains the top core skill, rated essential by about 70% of employers.
The same report puts numbers on the churn. It expects new jobs equal to 14% of today’s employment (about 170 million) and displacement of 8% (about 92 million), a net gain of 78 million. It also estimates that 59 of every 100 workers will need training by 2030, and that 11 of those 59 are unlikely to get it. These are employer expectations, not certainties, but they tell you roughly how fast the ground is moving.
Resilience in practice is less heroic than it sounds. Mostly it’s about recovering quickly when a tool, a process or a role changes under you, and a lot of that comes down to emotional regulation skills.
Start small: the next time a new system lands at work, volunteer to be one of the first to try it properly, even though you’ll feel clumsy at first.
11. Decide what to hand to AI and what to keep
Think about a pilot and an autopilot. The autopilot flies most of the journey, and that’s a good thing. But the pilot stays responsible, decides when to switch it on and off, and keeps enough hands-on practice to take over when something unusual happens.
The analogy holds in one important way: the tool handles routine work while you keep the judgment and the skills for exceptions. It breaks down in another. Aviation has strict rules, checklists and simulator training that force pilots to stay sharp. Your workplace almost certainly doesn’t, so you have to build those habits yourself.
Microsoft’s 2026 Work Trend Index suggests the most advanced AI users already do this. It identified a group it calls Frontier Professionals, about 16% of the 20,000 surveyed, who use AI agents for complex work and redesign their workflows around it. They were more likely than other users to pause and decide what AI versus a human should do (53% vs 33%). They were also more likely to do some work without AI on purpose to keep their skills sharp (43% vs 30%). That’s a correlation, not proof the habit causes success, but it’s a telling one. Once you’ve set your rules, sticking to your decision matters more than the rule itself.
Next time, do this: before a big task, sort its parts into three piles. AI does it, AI drafts and I fix, I do it myself. Then follow your own sort.
12. Measure whether AI actually speeds you up
Feeling faster and being faster are different things. A randomised controlled trial by Joel Becker, Nate Rush and colleagues at the research group METR, published in July 2025, showed how far apart they can drift.
Sixteen experienced open-source developers completed 246 real tasks on large projects they had worked on for years. Each task was randomly assigned to allow or forbid AI tools, mostly Cursor Pro with Claude 3.5 and 3.7 Sonnet. Before starting, the developers expected AI to cut their time by 24%. Afterwards, they believed it had cut their time by 20%. The measured result: with AI allowed, tasks took 19% longer. They accepted fewer than 44% of the AI’s code suggestions, and many reported heavy clean-up work.
The authors are careful about limits. It’s a small group, in one demanding setting, using early-2025 tools that have since improved. The broader point still stands. Your sense of your own productivity is a weak instrument. And if your AI workflow means constant back-and-forth between prompting and doing, it’s worth knowing how to focus on one thing at a time.
A quick experiment: time two similar tasks this week, one with AI and one without. Count the minutes, including the checking and fixing.
13. Get real reps early in your career
If you’re starting out, this skill is about getting experience on purpose. Erik Brynjolfsson, Bharat Chandar and Ruyu Chen at the Stanford Digital Economy Lab analysed payroll data from ADP, the largest payroll provider in the United States. In the November 2025 version of their paper, workers aged 22 to 25 in the most AI-exposed occupations saw a 16% relative decline in employment since generative AI spread widely, even after controlling for firm-level shocks. Older workers in the same occupations held steady or grew. The declines were concentrated in jobs where AI automates work rather than augments it. This is a working paper, and an earlier August version put the decline at 13%, so the estimate is still moving.
PwC’s 2026 Barometer adds a related signal. Among 2.4 million US entry-level jobs, the most AI-exposed ones were seven times more likely to ask for traditionally senior skills like leadership, creativity or face-to-face interaction. Put simply, the easy starter tasks are drying up, and the remaining entry-level jobs expect more. If that makes your twenties feel shaky, you’re not imagining it, and a quarter-life crisis is a common response.
Your move: volunteer for the messy, people-facing, judgment-heavy work nobody wants, like client calls, incident reviews and stakeholder meetings. That’s where the experience AI can’t hand you comes from.
14. Use AI responsibly and know the rules
Responsible AI use is turning from a nice idea into a legal expectation, at least in Europe. Under Article 4 of the EU AI Act, organisations that provide or use AI systems have had an AI literacy duty since 2 February 2025. That covers ordinary companies whose staff use chatbots at work. In July 2026 the Digital Omnibus on AI, Regulation (EU) 2026/1744, softened the wording. Organisations no longer have to ensure a “sufficient level” of AI literacy. Instead, they must take measures to support it. The duty itself stayed.
Most of this lands on employers, not individuals. Still, it changes what good employees look like. The person who knows which data can’t go into a public tool, when AI use should be disclosed, and where a human has to sign off is the person managers trust with sensitive work. This is general information, not legal advice. Rules vary by country, so check what applies where you work.
Here’s a simple way in: find your employer’s AI policy and read it properly this week. If there isn’t one, ask your manager what the expectations are, and write the answer down.
15. Teach and mentor others
Matt Beane, a technology management researcher at the University of California, Santa Barbara, has spent years studying how skills get passed on in workplaces full of robots and AI. In his 2024 book The Skill Code, he argues that skill has always grown through the bond between experts and novices. Intelligent technology often lets experts work alone, which quietly cuts novices out of the work they used to learn from. He identifies challenge, complexity and connection as the three conditions skill needs, and he urges managers to protect direct contact between experts and beginners.
That argument rests on Beane’s field research rather than a single trial, and the book is now a couple of years old. Newer data points the same way. LinkedIn’s Skills on the Rise 2026 list includes mentorship and team management among growing skills. Teaching also forces you to explain your judgment out loud, which is one of the best ways to sharpen it. If you manage people, it’s a core part of being a good boss.
Test it this week: invite a junior colleague to sit in on a piece of work you’d normally do alone with AI. Talk them through why you accept, reject or rewrite what the tool suggests.
What to do when you slip
You will slip. You’ll paste an AI answer you meant to check. You’ll let the tool write something you wanted to learn yourself. You’ll spend a month not touching any of these skills because work got busy.
How you respond matters more than the slip. In four experiments published in Personality and Social Psychology Bulletin in 2012, Juliana Breines and Serena Chen at UC Berkeley found that people who responded to a failure with self-compassion were more motivated to improve. In one experiment, after failing a difficult test, they spent more time studying for the retake than people who had been encouraged to boost their self-esteem. That study is old but foundational. A more recent set of four studies with 1,846 students, by Kévin Rigaud and colleagues in Personality and Individual Differences (2025), adds a realistic limit. Self-compassion after setbacks was linked to well-being, but not to academic performance. Being kind to yourself helps you stay in the game. It won’t build the skill for you.
So when you slip, fix the output first, name what happened without drama, and set one tiny rule to stop it repeating. If your inner voice turns harsh, stop hating yourself for it and get back to the next task.
Conclusion
You can’t future-proof a career with a single course or a single tool. What you can do is keep building the skills that compound: using AI well, checking it hard, knowing your field deeply, and getting better at the human parts of work that machines still handle clumsily.
Pick two skills from this list. One that uses AI and one that protects you from leaning on it too much. Work on just those two for the next month. Then come back and pick the next pair.
For more practical, research-backed ideas on building a career that lasts, explore our work articles.
