The most in-demand AI skills, according to the job postings
We read about 670 AI job postings in Europe. Production and teamwork come first; among AI skills, agents lead, then evals and retrieval.
Thomas Couderq, Founder of Tokn
8 min read
In this article⌄
- How we counted the skills in AI job postings
- The most in-demand AI skills in job postings
- The AI-specific skills employers ask for, ranked
- The non-AI skills employers ask for most
- Which AI skills employers want, by role
- How AI skills change with seniority
- What other AI skills reports say, and why they disagree
- Which AI skills are rising?
- Which AI skills to learn first, by goal
- Find jobs that ask for your skills
- FAQ
- What are the most in-demand AI skills in 2026?
- Is prompt engineering still in demand?
- Do AI jobs require a PhD?
- What soft skills do AI employers want?
- Which AI skill should I learn first?
The most in-demand AI skills in European job postings are not the ones on most lists. On Tokn, as of 14 October 2026, postings most often name working across teams (44%) and putting systems into production (43%). Among AI-specific skills, agents lead (1 posting in 4), then evaluation, retrieval (RAG), inference and fine-tuning. Prompt engineering appears in only 5%.
We read the core requirements of about 670 distinct AI job postings at 77 companies, collected from company career pages. This page covers capabilities; for the tools ads name, see the tools AI job postings name, and for hiring volumes, the state of the AI job market in Europe.
In short
- The most common asks are not AI-specific: cross-team work (44%), production and deployment (43%) and ownership of ambiguous problems (38%).
- Among AI skills, agents (25%) lead, then evals (15%), retrieval and inference (10% each) and fine-tuning (9%). Prompt and context engineering trails at 5%.
- The mix changes sharply by role: customer-facing roles are the largest family, and over half of them ask for pre-sales or customer-facing work.
How we counted the skills in AI job postings
We matched 50 skill tags against the requirement and responsibility bullets of every active posting, on one date.
- Base: of 802 active postings as of 14 October 2026, 668 at 77 companies have requirement or responsibility text once multi-city copies are removed. All shares use these 668.
- What we store: a shortened summary of each ad, up to five requirement and five responsibility bullets, without "nice to have" sections. Counts are floors, and they measure what employers list first.
- Concentration: Mistral AI has 86 of the 802 postings. Without it, no headline share moves by more than 2 points.
- Role families come from job titles. Below about 15 postings, read a figure as an example, not a statistic.
The most in-demand AI skills in job postings
Three general skills top the list, each named in about 4 postings in 10. The first AI-specific skills, machine learning and agents, come fifth and sixth.
| Skill | Postings | Share | Companies |
|---|---|---|---|
| Working across teams | 293 | 44% | 60 |
| Production and deployment | 287 | 43% | 65 |
| Ownership of ambiguous problems | 257 | 38% | 54 |
| Communication | 178 | 27% | 55 |
| Machine learning or deep learning | 171 | 26% | 43 |
| Agents and tool use | 168 | 25% | 51 |
| LLMs and generative AI | 165 | 25% | 51 |
| Cloud and infrastructure | 164 | 25% | 40 |
| Backend and APIs | 156 | 23% | 53 |
| Software engineering fundamentals | 151 | 23% | 48 |
| Reliability and observability | 146 | 22% | 47 |
| Business acumen | 112 | 17% | 37 |
| Evaluation (evals) | 101 | 15% | 34 |
| Customer-facing and pre-sales work | 98 | 15% | 33 |
| Stakeholder management | 98 | 15% | 32 |
The production figure counts "production" and "deploy". Adding the verb "ship", which designers and marketers use too, takes it to 52%. Overall, 63% of postings name at least one AI-specific skill, and 81% at least one human or business skill.
The AI-specific skills employers ask for, ranked
Agents lead, and they are the only AI skill present in almost every role family. The others concentrate in engineering and research.
| AI skill | Share of postings | Companies | Where it concentrates |
|---|---|---|---|
| Agents, tool use, MCP | 25% | 51 | 47% of AI and ML engineering ads |
| Evaluation (evals) | 15% | 34 | 38% of research, 30% of AI and ML engineering |
| Retrieval, RAG, search | 10% | 26 | 31% of AI and ML engineering |
| Inference and serving | 10% | 28 | 25% of infrastructure ads |
| Fine-tuning and post-training | 9% | 26 | 42% of research |
| Pre-training at scale | 8% | 20 | 41% of research |
| Prompt and context engineering | 5% | 21 | 17% of AI and ML engineering |
Production comes first. In AI and ML engineering, 65% of ads say "production" or "deploy". Artefact's Director Applied AI ad, advertised in October, put it plainly: "A model living in a notebook is a hobby, not a deliverable." Agent postings mention production or shipping 74% of the time.
Agents, evals and retrieval come as a bundle. Half of the postings that name evals also mention agents, and 40% of RAG postings mention evals. ClickHouse's AI Product Engineer ad listed the whole stack in one line: "Context engineering, tool design, evals, tracing, cost." Only 15 postings, at 13 companies, name all three, mostly AI engineer, applied AI and forward deployed titles.
Prompt engineering is smaller than the hype. It features in LinkedIn's growth lists, yet only 36 postings name prompt or context engineering, and 72% of those also mention agents. Employers treat prompting as part of building agents, not a job on its own. Rare standalone asks come from non-engineering roles, like a Legora legal engineer ad asking for "sharp prompt and use case discernment".
Model-building skills live in research. Fine-tuning and pre-training each appear in about 4 research ads in 10, a PhD in 27%.
The non-AI skills employers ask for most
Working across teams (44%) and owning ambiguous problems (38%) appear more often than any AI skill. Communication (27%), business acumen (17%) and stakeholder management (15%) follow.
They peak in customer-facing roles, the largest family on the board (115 postings): 53% name customer-facing or pre-sales work, 40% business acumen, 33% stakeholder management. Cohere's FDE ad asked candidates to "translate high-value, ambiguous business problems into well-framed agentic workflows with clear success criteria and evaluation". Our guide to forward deployed engineers covers that role in depth.
Using AI tools in your own work is named in 10% of postings and 28% of design ads; the tools article has examples.
Which AI skills employers want, by role
Engineering asks for production and agents, research for training and evals, product and marketing for collaboration and ownership.
| Role family (postings) | Most named skills |
|---|---|
| Customer-facing technical (115) | Cross-team work 54%, pre-sales 53%, production 46%, business acumen 40% |
| AI and ML engineering (106) | Production 67%, agents 47%, software fundamentals 44%, RAG 31%, evals 30% |
| Research and applied science (79) | ML/DL 57%, fine-tuning 42%, pre-training 41%, evals 38%, PhD 27% |
| Infrastructure, MLOps, inference (77) | Cloud 73%, production 71%, reliability 61%, GPU 31%, inference 25% |
| Go-to-market (58) | Ownership 47%, cross-team work 45%, communication 41%, business acumen 33% |
| Product and program management (44) | Cross-team work 68%, ownership 66%, product sense 36%, communication 34% |
| Data science and analytics (42) | Maths and statistics 62%, data analysis and SQL 57%, experimentation 33% |
See what an LLM engineer does for the engineering side and our guide to AI product managers for product. Most go-to-market and product ads name no tool at all: their skills are written in prose.
How AI skills change with seniority
In postings that state a level, mentoring jumps at lead level, and junior ads stress fundamentals over ownership.
| Skill | Junior (49) | Mid (79) | Senior (131) | Lead (178) |
|---|---|---|---|---|
| Ownership of ambiguous problems | 14% | 43% | 37% | 38% |
| Leading or mentoring | 2% | 9% | 7% | 26% |
| Communication | 39% | 29% | 25% | 28% |
| Maths and statistics | 24% | 3% | 11% | 3% |
Caveats: 35% of postings state no level, and the 49 junior ads cluster at five companies, many of them data science and consulting internships.
Languages: English is the default; a second language is tied to customer-facing roles selling into one country. Anthropic's Applied AI Engineer, Enterprise ad asked for "Native or fluent French, written and spoken" to work with French enterprise customers. Explicit asks are rare in our stored text (14 postings for French, 7 for German), plus 32 postings written in French.
What other AI skills reports say, and why they disagree
The big reports measure different things: profile growth, employer expectations, pay premiums or AI mentions across all jobs. Tokn measures what AI roles in Europe require today.
| Source | What it measures | AI-skills finding |
|---|---|---|
| LinkedIn Skills on the Rise 2026 | Growth of skills on profiles and hires, Dec 2024 to Nov 2025 | Prompt engineering and LLMs among the fastest-growing, next to cross-functional collaboration |
| WEF Future of Jobs Report (the latest edition, published January 2025) | Survey of over 1,000 employers, outlook to 2030 | AI and big data top the fastest-growing skills |
| PwC Global AI Jobs Barometer 2026 | Wage premium in over a billion job ads, all occupations, not Europe-specific | 62% average premium for AI skills |
| Stanford AI Index 2026, Lightcast data | Skills extracted from US job ads | Agentic AI skills went from 0.06% to 0.23% of US postings, 2024 to 2025 |
| Tokn | Core requirements of AI job ads in Europe, one date | Production and teamwork first; agents lead AI skills; prompting at 5% |
Which AI skills are rising?
Tokn started tracking postings on 11 October 2026, so we can't yet say which skills are rising; we'll update this page as months accumulate. For growth, Indeed Hiring Lab's open AI tracker is the best European evidence: the share of all UK postings mentioning AI went from 5.1% to 9.9% between September 2025 and September 2026, by our reading of its data, and from 3.6% to 5.7% in Germany.
Which AI skills to learn first, by goal
Here is the order we would follow, based on what each role family asks for.
Software engineer moving into AI
- Ship one LLM feature to production, end to end.
- Agents and tool use.
- Evals: measure whether it works, then improve it.
- Retrieval, then observability, latency and cost.
Data scientist
- Keep statistics and experimentation strong (62% and 33% of data science ads).
- LLM output evaluation, retrieval and production basics.
Researcher or research engineer
- Post-training and RL, evals, distributed training.
- GPU performance. A PhD and publications help here most.
Customer-facing (FDE, solutions)
- Build and demo agents; scope ambiguous business problems.
- Executive communication, and a second language if you target one market.
Product, design or marketing
- Product sense and experimentation.
- Daily AI-tool use you can show, and clear writing.
For the full path, see how to become a machine learning engineer, our overview of non-technical AI jobs and the step-by-step guide to getting a job in AI.
Find jobs that ask for your skills
To test this against your profile, start with all AI jobs on Tokn, then create a job alert for your category and city to get new matching roles by email.
FAQ
What are the most in-demand AI skills in 2026?
In Tokn's postings, as of 14 October 2026, the most named skills are working across teams (44%), production and deployment (43%) and ownership of ambiguous problems (38%). Among AI-specific skills, agents lead (25%), then evals (15%), retrieval and RAG (10%), inference (10%) and fine-tuning (9%).
Is prompt engineering still in demand?
It grows fast on LinkedIn profiles, but only 5% of the postings we read name prompt or context engineering, and nearly three in four of those also mention agents. Employers want prompting as part of building agents.
Do AI jobs require a PhD?
Rarely: 5% of postings name a PhD. In research and applied science roles it is 27% (21 of 79 postings).
What soft skills do AI employers want?
Working across teams and ownership of ambiguous problems appear in more postings than any AI skill, followed by communication. Customer-facing roles add business acumen and stakeholder management.
Which AI skill should I learn first?
For most engineers: ship one LLM feature end to end, then add agents, evals and retrieval. Researchers should focus on post-training and evals, data scientists on statistics plus LLM evaluation, non-technical candidates on product sense and visible daily AI use.
