AI skills for 2030: what to learn now, according to the data

What counts as an AI skill (and what is a tool feature)

Most lists of "AI skills" are lists of buttons. Knowing where the summarise command sits in one vendor's product is a feature, and features change every quarter. A skill is something that transfers when the tool changes: writing a brief an agent can execute, designing a check that catches a wrong answer, knowing which tasks sit inside a model's competence and which just outside, connecting two systems so a task runs without you, and combining any of those with a domain you already know.

That distinction is the organising idea of this page. The ranking below rates each skill by durability, the measured labour-market data is kept separate from our forecast, and the learning plan at the end prefers free material because the skills, not the subscriptions, are what employers pay for.

The demand is measurable. Our keyword data shows "ai skills" at about 3,600 US searches a month, up roughly 400% year on year with a three-year compound growth rate near 100%; "ai literacy" sits at 2,900 a month, growing 113% a year; in the UK the term draws 720 searches (up 238%) and in Germany "KI-Kompetenzen" 590 (DataForSEO, retrieved 2026-11-09). A note on vocabulary: "skills" has also become a product term for packaged agent instructions, and queries such as "claude skills marketplace" (5,400 a month) belong to that meaning, not to this page.

The wage premium and who gets it

PwC's 2025 AI Jobs Barometer, an analysis of close to a billion job advertisements across six continents, finds that workers with AI skills earned an average wage premium of 56% in 2024, up from 25% the year before, and that productivity growth in the industries most exposed to AI nearly quadrupled, from 7% in 2018–2022 to 27% in 2018–2024.^[1] A

Two cautions before you plan a career around that number. First, it is a premium in advertised salaries for roles that list AI skills, not a measured raise for learning them; roles that ask for AI skills also tend to be senior, technical and in high-paying sectors. Second, the premium is unlikely to persist for skills everyone has. The controlled studies suggest where the durable value sits: in a quasi-experiment across 5,172 customer-support agents, a generative-AI assistant raised issues resolved per hour by about 14–15% on average, over 30% for the least experienced and close to zero for the most experienced.^[2] A In a preregistered experiment with 453 professionals, ChatGPT compressed the quality gap between weaker and stronger writers.^[3] A When a tool raises the floor, the premium moves to whatever the tool cannot do: choosing the task, checking the output, and knowing the domain well enough to tell good from plausible.

Measured demand: what WEF and ILO actually count

Before the ranking, the labour-market numbers, each with its grade and its limits.

Figure	Source	What it measures	Grade
39% of workers' current skills transformed or obsolete by 2030	WEF Future of Jobs 2025 (employer survey, 1,000+ firms)[4]	Employers' expectations, not observed change	B
59 of every 100 workers need reskilling or upskilling by 2030; 11 unlikely to receive it	WEF Future of Jobs 2025[5]	Survey extrapolation	A
170 million jobs created, 92 million displaced by 2030 (net +78 million)	WEF Future of Jobs 2025[5]	Survey extrapolation	A
25% of global employment in occupations with some generative-AI exposure; 34% in high-income countries; 3.3% in the highest tier	ILO refined exposure index, May 2025[6]	Task exposure, not displacement	A
56% wage premium for AI skills (2024)	PwC AI Jobs Barometer 2025[1]	Advertised salaries	A

These are the facts. Everything in the "Our forecast" section further down is a model output, and we keep them apart on purpose.

Skills ranked by durability

The durability rating is our editorial judgement, informed by the studies cited; it is not a measured quantity. "Durable" means the skill survives a change of vendor and a generation of models.

Rank	Skill	Durability	Why, with evidence	Free way to practise
1	Verification and acceptance testing	Durable	Users of AI were 19 points less accurate on tasks just outside the model's competence and did not notice;[7] automation bias falls with training and accountability[8]	Sample and trace three claims in every AI output you use this week
2	Task selection (knowing the frontier)	Durable	Gains concentrate inside the frontier and among novices;[2] [7] human–AI teams lose on decision tasks[9]	Keep a two-column log: tasks where AI helped, tasks where you corrected it
3	Specification and briefing	Durable	The successor to prompting: goal, inputs, constraints, done-criteria, budget, escalation	Write one six-field brief per delegated task; see our agent playbook
4	Domain expertise plus AI	Durable	The premium moves to judgement when tools raise the floor;[3] experienced developers lost time with AI on code they knew well, which is a warning about where not to apply it[10]	Apply AI to the parts of your field you know least, check with the parts you know best
5	Workflow and automation design	Durable, tool-shaped	Connecting triggers, steps and approvals transfers across platforms; the platforms do not	Build one scheduled automation with a free plan or a self-hosted instance
6	Data literacy	Durable	Reading a table, a distribution and a base rate is how you catch a confident wrong number	Re-derive one statistic from its source each week
7	AI literacy (legal and ethical)	Durable, rising	Required of deployers under the EU AI Act since February 2025[11]	Read Article 4 and your organisation's AI policy
8	Model and tool fluency	Tool-specific	Necessary today, obsolete by the next release	Vendor documentation; expect to relearn quarterly
9	Prompt techniques	Fading as a term	Absorbed into briefing and into the models themselves; search share falling (next section)	Learn the five patterns, then stop collecting them
10	Building agents from scratch	Specialist	Valuable for engineers; most knowledge workers need to manage agents, not code them	Optional; use a no-code platform first

Prompting is fading as a term: what replaces it

Google Trends data, read as each term's share of its five-term group to cancel a sampling shift in 2025, shows "prompt engineering" losing 6.2 percentage points of share over the past twelve months after peaking in the week of 15 February 2026, while "ai skills" gained 6.8 points and "ai literacy" 2.1 (DataForSEO Trends, retrieved 2026-11-09). Raw interest in prompting is still higher than two years ago; it is simply growing more slowly than the words around it.

The reason is visible in the studies. The failures that matter are not badly worded requests; they are tasks handed over on the wrong side of the frontier and outputs accepted without a check.^{[7] [8]} Three things replace prompting as the skill:

1. Specification. A brief with a goal, inputs, constraints, acceptance criteria, a budget and escalation rules. It is a management artefact, and it outlives any model.
2. Context design. Deciding what the model is allowed to see: which documents, which examples, which tools. Permissions are part of the skill.
3. Evaluation. Writing the check before the task runs, sampling the output, logging the error rate. This is where the durable premium sits.

The EU AI Act literacy duty (Article 4)

Since 2 February 2025, Article 4 of the EU AI Act requires providers and deployers of AI systems to take measures to ensure a sufficient level of AI literacy among the staff who operate them, taking into account their training, experience and the context of use. The Act's high-risk obligations for workplace uses, which include systems that allocate tasks or evaluate workers, now apply from 2 December 2027 after the 2026 Omnibus amendment.^[11] A For an employee the practical consequence is that AI training is now something your employer must provide or enable, not a personal hobby; for a manager it is a documentation task with a date. Our explainer on AI literacy covers the definition, the duty and what counts as sufficient. This page is not legal advice.

Our forecast for skills demand

In our Future Demand Radar, the category "future-of-work skills and AI literacy" scores 45.8 out of 100 in the baseline scenario (p10–p90 band 41.4–50.2), which places it in the "monitor" class; the conservative scenario gives 42.5 ("monitor") and the accelerated scenario 48.2 ("validate in 12 months"). Our baseline scenario expects US search demand for the category to reach about 1.5 times today's level by 2028, 2.2 times by 2030 (band 1.7–2.8) and 2.8 times by 2035. The assumptions behind that path are the WEF churn figures above, the PwC premium and the AI Act's literacy duty as a compliance driver; the main downside risk is commoditisation, as every platform bundles its own training. Forecasts are model outputs with stated uncertainty, not facts, and the three scenarios for knowledge work to 2035 show what each path looks like in practice.

A 90-day learning plan, with free and paid options

The plan assumes four hours a week. Free options come first in each block because the skills above are learned by doing them on your own work, not by watching. We have no affiliate relationship with any course platform and have not reviewed specific courses; where paid courses are mentioned, they are options to evaluate, not recommendations.

Days 1–30: foundations and frontier-finding.

• Free: read the WEF press release and the ILO exposure paper linked below, then list the tasks in your own job and sort them into "inside the frontier", "outside", and "not sure".^{[5] [6]}
• Free: work through the official documentation of one assistant you already have access to (every major vendor publishes guides), then write six-field briefs for three real tasks.
• Free: run our Tool Finder to pick a stack to practise on, filtered by budget and privacy.
• Paid option: a structured introductory course on a platform such as Coursera or Udemy if you learn better with a syllabus; check the publication date, since material older than a year describes tools that have changed.

Days 31–60: automation and verification.

• Free: build one scheduled automation with an approval step, on a free plan or a self-hosted instance of n8n, Zapier or Make. Keep a run log: runs, failures, minutes spent checking.
• Free: adopt the acceptance-test habit on every AI output you use. Sample three claims, trace them to a source, record the error rate.
• Paid option: a course on workflow automation or data literacy, again only after you have one automation running, so you know which gaps are yours.

Days 61–90: domain and judgement.

• Free: apply AI to the part of your field you know least, and check it with the part you know best. Note where it fails; that is your personal frontier map.
• Free: read Article 4 of the AI Act and your organisation's AI policy; ask what training your employer provides under the duty.^[11]
• Free: teach the briefing-and-testing method to one colleague. Explaining it is the fastest test of whether you have it.
• Paid option: employer-funded training. Under Article 4 the question "what AI literacy training do we provide" is one your organisation should be able to answer.

At day 90 you have a frontier map, a brief archive, a running automation with a log and an error rate you can quote. That is a portfolio, and it is worth more in an interview than a certificate.

What to do next

Open the Tool Finder to choose a stack to practise on; the point is to learn the durable skills on real work, and the finder filters tools by budget, team size and privacy tier so the practice environment matches your constraints. The complement to this page is human skills in an AI economy, which covers judgement, taste, trust and coordination with the exposure evidence behind each. The ranking above is reviewed when we re-run the model on new data; subscribe below for the update.

Frequently asked questions

What are the most important AI skills to learn for 2030?

By durability: verification and acceptance testing, task selection (knowing which tasks sit inside a model's competence), specification and briefing, domain expertise combined with AI, workflow and automation design, and data literacy. Tool fluency and prompt techniques are necessary today but change with every release. The ranking is our editorial judgement informed by the controlled studies cited on the page.

Is the 56% AI wage premium real?

It is a measured figure from PwC's analysis of close to a billion job advertisements: roles listing AI skills advertised salaries 56% higher in 2024, up from 25% a year earlier.^[1] A It describes advertised pay for roles that demand AI skills, which are often senior and technical, not a raise for completing a course, and it is unlikely to persist for skills everyone acquires.

Is prompt engineering still worth learning?

Learn the handful of patterns, then move on. In Google Trends the term has lost 6.2 percentage points of share within its group over the past year after peaking in February 2026, while 'ai skills' and 'ai literacy' gained. What replaces it is specification (a full brief), context design (what the model may see) and evaluation (the check you run before trusting the output).

Does my employer have to train me in AI?

Under Article 4 of the EU AI Act, applicable since 2 February 2025, providers and deployers must ensure a sufficient level of AI literacy among staff operating AI systems, taking account of their role and context.^[11] A How that is met is up to the organisation. This is not legal advice; our AI literacy explainer has the detail.

Which courses do you recommend?

None by name. We have no affiliate relationship with course platforms and have not reviewed individual courses, so we list Coursera and Udemy only as places where structured options exist. Prefer material less than a year old, and start a course only after you have practised on your own work for a few weeks, so you know which gap you are filling.

How does measured demand differ from your forecast?

Measured demand is what the WEF, ILO and PwC counted and what our keyword database recorded on 2026-11-09, each with a grade. The forecast is our radar model's output: a baseline score of 45.8 for the skills category with a 41.4–50.2 band and an expected 2.2-fold rise in search demand by 2030. The first set are facts with stated limits; the second is a model with stated uncertainty.

Sources and evidence

1. 2025 Global AI Jobs Barometer (press release: AI linked to a fourfold increase in productivity growth) — PwC (2025), report B www.pwc.com/gx/en/news-room/press-releases/2025/ai-linked-to-a-fourfol (retrieved 2026-11-09)
2. Generative AI at Work — Brynjolfsson, Erik; Li, Danielle; Raymond, Lindsey (2025), observational A doi.org/10.1093/qje/qjae044 (retrieved 2026-11-09)
3. Experimental evidence on the productivity effects of generative artificial intelligence — Noy, Shakked; Zhang, Whitney (2023), RCT A doi.org/10.1126/science.adh2586 (retrieved 2026-11-09)
4. Jobs market at a crossroads: which are the fastest growing and declining jobs? (summarising WEF Future of Jobs 2025) — Euronews (2025), report B www.euronews.com/business/2025/02/01/jobs-market-at-a-crossroads-which (retrieved 2026-11-09)
5. Future of Jobs Report 2025 – press release (French edition) — World Economic Forum (2025), report B reports.weforum.org/docs/WEF_Future_of_Jobs_2025_Press_Release_FR.pdf (retrieved 2026-11-09)
6. Generative AI and Jobs: A Refined Global Index of Occupational Exposure (ILO Working Paper 140) — International Labour Organization (2025), report A www.ilo.org/publications/generative-ai-and-jobs-refined-global-index-o (retrieved 2026-11-09)
7. Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of Artificial Intelligence on Knowledge Worker Productivity and Quality — Dell'Acqua, Fabrizio; McFowland III, Edward; Mollick, Ethan R.; Lifshitz-Assaf, Hila; Kellogg, Katherine; Rajendran, Saran; Krayer, Lisa; Candelon, François (2023), RCT B doi.org/10.2139/ssrn.4573321 (retrieved 2026-11-09)
8. Complacency and Bias in Human Use of Automation: An Attentional Integration — Parasuraman, Raja; Manzey, Dietrich H. (2010), systematic review A doi.org/10.1177/0018720810376055 (retrieved 2026-11-09)
9. When combinations of humans and AI are useful: A systematic review and meta-analysis — Vaccaro, Michelle; Almaatouq, Abdullah; Malone, Thomas (2024), meta-analysis A doi.org/10.1038/s41562-024-02024-1 (retrieved 2026-11-09)
10. Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity — Becker, Joel; Rush, Nate; Barnes, Elizabeth; Rein, David (METR) (2025), RCT B doi.org/10.48550/arXiv.2507.09089 (retrieved 2026-11-09)
11. AI Act | Shaping Europe's digital future (Regulation (EU) 2024/1689, application timeline incl. 2026 Omnibus amendment) — European Commission (2026), report A digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai (retrieved 2026-11-09)