Is AI creating entirely new jobs? I read 17,513 postings. It adds one duty.

If you work outside engineering, your feed has probably shown you this story in 2026. AI is creating lots of entirely new jobs: Head of AI Transformation, AI Enablement Lead, AI Marketing Manager. And it seems legit: "a whole new era", "AI is changing the job market". These are jobs you can actually imagine.

If a new profession was forming next to each of us, I was curious what it was and whether I should move into it. Let's see what I found.

Before counting, I expected three things:

• AI jobs take a growing share of hiring.
• The US is ahead of Europe.
• An AI job is a new profession, with its own duties and skills.

So I counted. Three sources:

• Company career pages. 5,856 job postings with AI, ML or LLM in the title from 930 companies, collected between July and October 2026.
• Job postings, 12 countries. In October I collected 17,513 postings from the US, the UK and ten more European countries: roles with AI in the title and ordinary roles from matching job families.
• The duties. For 1,273 postings a language model read the text and marked which duties it gives the hire: building AI tools, training people, selling an AI product and so on. I call these marks duty labels. The model worked from a codebook, my written rules for each duty. To check the labels, part of the postings was labelled again without access to the first result.

Assumptions:

• An AI role has AI, ML, LLM, GenAI, artificial intelligence or machine learning in the title. An ordinary role comes from a matching job family, such as marketing, product or sales, with no AI in the title: Marketing Manager against AI Marketing Manager.
• I compare non-engineering roles. Where engineering roles are counted too, I say so. Titles with engineer, scientist, developer, architect or programmer are out on both sides. On career pages I keep forward deployed and solutions engineers and architects with the non-engineering roles, since they work with clients: 276 of the 2,100.
• "Mentions AI" means at least one sentence with AI, LLM, machine learning, GenAI, agentic, GPT, Copilot or prompting. I call this word search the AI marker. It reads English, so I compare countries on English texts, and in Table 3 with a marker that also reads local words.
• Employers fall into three types: tech companies; consultancies, which include agencies, IT services and recruiters; everyone else, from banks and insurers to retailers and hospitals.
• The 12-country postings are a sample, so I compare postings with each other and calculate no market shares from them.

None of the three expectations was confirmed. I start with the first one: does the share of AI jobs grow?

AI titles stay flat. AI mentions grow.

To answer it, I followed 348 companies whose career pages I tracked from early August. Six times between 5 August and 17 October I counted their open AI roles and divided them by all their open roles.

On 5 August they had 61.5 AI roles per 1,000 openings. On 17 October, 59.3. On the four dates in between the share stayed from 61.4 to 63.1. Outside engineering the share stayed between 16 and 19 per 1,000: 16.2 on 5 August and 16.2 on 17 October.

The number of companies with an AI role barely moved too: 106 in August and 113 in October, out of 317 companies with no large one-off jump in postings during the window.

Indeed counts far more postings than I do: the share of all postings on Indeed that mention AI. In a year it went from 3.7% to 7.3% in the US and from 5.1% to 9.9% in the UK. In my window, from 5 August to 30 September, it still grew: from 6.4% to 7.3% in the US and from 9.6% to 9.9% in the UK. Indeed also counts machine learning and data science terms, so I show the two numbers side by side and keep them apart.

So AI in job descriptions keeps growing, and the share of AI titles at the companies I follow stayed flat. The first expectation is not confirmed.

At the companies I follow there are about as many AI titles per 1,000 openings in October as in August. If you look for AI work on career pages, the growth is inside ordinary postings, so I looked there next.

Ordinary jobs mention AI too, mostly at tech companies

I took the ordinary roles from the 12-country postings, the ones with no AI in the title, and checked how many mention AI anywhere in the text. At tech companies it is 68%: 854 postings out of 1,265. At consultancies 49%, 400 out of 812. At other employers, from banks to hospitals, 30%: 561 out of 1,896, about three in ten.

Table 1. Ordinary roles that mention AI, by employer type and region. English-language postings, non-engineering roles; number of postings in brackets. "All 12 countries" also includes Cyprus and Turkey.

Tech companies lead in every region: by 1.8 times in the UK and up to 2.7 times in the US.

For example, take two of the US marketing manager postings from October, neither with AI in the title. One is from a tech company and lists "familiarity with AI-driven marketing tools" in the requirements. The other is from a credit union and has no mention of AI.

How much AI is in a posting that mentions it? In a typical one, 3 sentences out of about 50 at a tech company and 2 at other employers. The difference is in the postings that talk about AI at length.

At tech companies 31% of ordinary postings have 4 or more sentences about AI: 394 out of 1,265. At other employers it is 8%, 150 out of 1,896, about one in twelve. So at a tech company AI is a visible part of an ordinary job description, more than a single word.

Part of these mentions sits in the company description: "we are an AI-powered platform". An "About us" block could be found and cut out in 44% of tech postings, 30% at consultancies and 31% at other employers.

Table 2. Where AI sits in ordinary postings that have an "About us" block. English-language postings, non-engineering roles, all 12 countries.

At a tech company 39% of the sentences about AI describe the company itself. Dropping the postings that mention AI only there moves tech companies from 68% to 60%, consultancies from 49% to 45% and other employers from 30% to 28%. In postings without an "About us" heading I can't separate the two. If they look like the ones with it, the shares come to about 51%, 38% and 25%. The order of the three types holds.

An AI mention in an ordinary posting says little about how much of the job is AI. In my view, at a tech company it says that the company is bringing AI into its work, and into many roles at once: more than half of its ordinary postings mention AI. Which part of that is yours, and what the posting means by "AI tools", it may be useful to ask at the first interview. At other employers this is still rare: one ordinary posting in twelve talks about AI at length.

The US and Western Europe are close inside each employer type

My second expectation was that the US is ahead of Europe. Table 1 already answers it. Inside each employer type the US and Western Europe are close: 72% and 71% at tech companies, 56% and 51% at consultancies, 27% and 31% at other employers.

Language changes the picture inside France and Germany. About a third of their postings are in English (29% and 34%), and those mention AI as often as in the US and the UK, or more. Postings in French and German mention it less.

Table 3. Ordinary roles that mention AI, by language of the posting. The marker reads AI in English and in the local language (KI, IA and the local words for artificial intelligence). Non-engineering roles; number of postings in brackets.

Part of the gap comes from who writes in English. In France tech companies posted 42% of the English-language postings and 22% of all postings. In Germany it is 47% and 26%. Tech companies mention AI more often than other employers, so English-language postings do too.

With all languages counted, the four countries sit between 41% and 49%. On English texts the US and Western Europe are within five points of each other inside each employer type. So for AI in ordinary roles the second expectation is not confirmed. AI roles themselves differ more, and Table 7 below shows it.

In practice the employer and the language of a posting tell you more than its country. In France and Germany an English-language posting mentions AI at least as often as one in the US, and it more often comes from a tech company.

Outside engineering an AI title is a familiar job

That leaves the last expectation: is an AI job a new profession? To check it, I sorted all 5,856 AI roles from the career pages by their titles. These are roles with AI, ML or LLM in the title from 930 companies, collected from the full job lists of their career pages.

Close to two thirds of them are engineering: 3,756 roles, or 64%. The other 2,100, or 36%, are non-engineering roles from 597 companies.

Engineering roles

Table 4. Engineering AI roles on company career pages, July to October 2026. 3,756 roles; groups come from rules on title words. Shares add up to 100%, give or take a point from rounding.

The largest group builds apps and agents on top of language models: 38% of engineering AI roles. Classic machine learning and platform work add another 40%. Most of these roles ask for an engineering background, so the rest of the article looks at the 2,100 non-engineering AI roles, 36% of all 5,856 AI roles I reviewed for this article.

Non-engineering roles

Here I split AI vendors, companies whose product is AI, out of tech companies. The first row holds forward deployed, solutions and presales roles. They work with a client: before the sale they show how the product fits, and after it they set the product up.

Table 5. Non-engineering AI roles on company career pages by job family and employer type. Read down each column: a column is one employer type and adds up to 100%, give or take a point from rounding. "All" includes 79 roles from employers of unknown type.

Counts behind Table 5

Non-engineering AI roles	AI vendors	Tech companies	Consultancies	Other employers	Unknown type	All
Forward deployed, solutions, presales	145	191	46	35	8	425
Product (PM, owner, design)	33	241	15	63	9	361
Marketing, growth, content	33	135	45	52	25	290
Sales, accounts, partnerships, customer success	33	145	13	21	5	217
Transformation, enablement, adoption, strategy	46	78	32	29	5	190
Governance, risk, legal, safety	19	58	13	27	6	123
AI operations and automation	6	36	13	17	2	74
Analytics and data	2	31	10	21	6	70
Program management, HR, AI leadership	9	45	8	7	2	71
Other titles	44	128	22	74	11	279
Roles	370	1,088	217	346	79	2,100

Almost every group here is a familiar profession: product, marketing, sales, helping clients deploy a product. The titles from my feed, transformation, enablement and adoption, form one group: 9% of non-engineering AI roles, 190 out of 2,100. What this group is asked to do comes in the next section. What changes from employer to employer is the mix. At AI vendors 39% of non-engineering AI roles help clients deploy the vendor's product. At tech companies the largest group is product, 22%. At consultancies two groups share the lead with 21% each: marketing, and solutions and presales.

So outside engineering an AI title usually means a familiar job. The word AI says what the job is attached to: an AI product the company sells, or bringing AI into how the company works. The duty labels show which of the two you get.

• Working on an AI product the company sells: marketing it, selling it, managing it.
• Bringing AI into how the company works: building automations and agents, rolling them out, setting the AI roadmap, measuring the return, training people, picking vendors.

Table 6. What non-engineering AI roles are asked to do, by employer type. Duty labels on 792 postings from the 12-country sample and company career pages. A posting can carry several duties, so rows add up to more than 100%. AI vendors are counted with tech companies.

At a tech company half of the AI roles work on an AI product. At other employers 86% bring AI into how the company works and 11% touch a product. A consultancy does the second job for its clients: 56% of its AI roles work with clients.

So with an AI posting it helps to look first at who posted it. At a bank or a hospital the job is most likely about bringing AI into the company. At a tech company it can be either one, and the text of the posting tells you which.

The one duty the word AI adds

If the job depends on the employer, does the word AI add anything by itself? It does, and it is one duty: bringing AI into how a company works. I counted, separately for AI roles and for ordinary roles, how many postings mention adoption, change management or upskilling.

Table 7. Share of postings that mention adoption, change management or upskilling, counted separately for AI roles and for ordinary roles. Each column is its own base, so the two do not add up to 100%. English texts, non-engineering roles; "All 12 countries" also includes Cyprus and Turkey.

In ordinary postings these words often describe something else, such as customers adopting a product: 493 of the 1,239 ordinary postings that use them have no sentence about AI. The duty labels below count the duty itself.

What does this duty consist of? The duty labels split it into parts.

Table 8. Duties written in the posting, AI roles against ordinary roles. Duty labels on 801 AI roles and 452 ordinary roles, non-engineering, from the 12-country sample and company career pages. Each column is its own base; a posting can carry several duties, so the columns add up to more than 100%.

The last row is the most interesting. Using AI tools in your own work shows up in 10% of AI roles and 8% of ordinary ones, close. The difference is in bringing AI into how the company works: deciding where it goes, rolling it out, writing the rules, measuring the result, training people. Each of these appears in a quarter to a half of AI roles and in almost no ordinary roles. At other employers the shares are higher still: 69% of AI roles there set AI strategy, 61% roll AI out, 52% own AI policy and risk, 46% measure the impact.

So it is one duty with several parts: bringing AI into how a company works, its teams, processes and rules.

Here is how it reads in a posting. A US clearing house hires a Principal, AI Enablement & Change, and gives the role "the change management that determines whether AI capabilities… actually get used."

The obvious objection: some AI roles in the sample were picked by titles like AI Adoption and AI Enablement, so of course they talk about adoption. True, and that is why I checked inside each job family. In marketing, these words show up in 48% of AI roles and 18% of ordinary ones (166 and 742 postings). In product it is 59% against 38%, in governance 48% against 20%. In the duty labels the part "Roll AI out", counted without transformation roles, appears in 33% of AI roles against 6% of ordinary ones (612 and 384 postings). So the gap holds in roles whose titles say nothing about adoption.

Other additions are smaller and depend on the role:

• Code. Python in 18% of AI roles against 3% of ordinary ones (all texts, 2,440 and 5,854 postings). The gap sits in automation, transformation and solutions roles. AI marketing roles ask for Python in 10% of postings and AI product roles in 5%.
• Prompts. 20% of AI roles mention prompts or prompting, against 1% of ordinary ones.
• Seniority, in the US. The median AI role asks for 8 years of experience, the ordinary role for 5. In the UK and Western Europe both ask for 5.

So in most AI postings, whatever else the job is, the hire is asked to bring AI into how the company works.

In my view this is the part to check against your own experience. If you have brought a new tool into a team, such as a CRM or an analytics platform, trained people on it and measured what changed, you have done three parts of this duty with a tool other than AI. Whether an employer counts that for an AI role, the postings I read do not say.

Few AI roles are old jobs with a new label

The sceptic's answer to the third expectation is the opposite one: an AI job is an old job with a new word in the title. If a company closes its Content Manager posting and soon after opens an AI Content Manager with the old text, that is a new label on an old job.

I checked this in the career page data, inside each company. For every non-engineering AI role with a usable text I looked for an ordinary role at that company that closed between 60 days before and 30 days after the AI role appeared. Then I scored how alike the two texts are, from 0 for no words in common to 1 for word for word. Sentences that repeat across a company's postings, such as menus and "About us" paragraphs, came out first, since they make any two postings look alike. To have a number to compare with, I ran this test again on ordinary roles followed by other ordinary roles.

• AI roles: a very close predecessor, with a score of 0.7 or more, for 1% of the 330 roles that had an ordinary role closing nearby.
• Ordinary roles: 4% of 1,898.

So an AI role copies an earlier posting less often than an ordinary role does.

The duty labels agree: in 4% of the labelled AI roles, 35 of 801, the duties hold nothing about AI.

So the third expectation is not confirmed. An AI role outside engineering is a familiar job: product, marketing, sales or delivery, and the mix depends on the employer. A copy of an old posting under a new title is rare. What the title adds is one duty that ordinary roles rarely carry: bringing AI into how a company works, its teams, processes and rules.

So, is AI creating entirely new jobs?

Mostly no. It adds one duty to jobs that already exist.

• The share of AI titles stays flat, 61.5 and 59.3 per 1,000 openings, while AI keeps spreading into ordinary job descriptions, most of all at tech companies.
• The US and Western Europe are close on AI in ordinary roles once you compare within one employer type.
• An AI title outside engineering is a familiar job: product, marketing, sales or delivery, and rarely a copy of an old posting (1% of the AI roles I could test against 4% of ordinary roles). At consultancies and other employers it is mostly about bringing AI into how a company works, in 81% and 86% of AI roles. At tech companies the roles split: 25% work only on an AI product, 36% only bring AI into the company, 27% do both, and the other 13% carry neither duty.
• The title adds one duty: bringing AI into how a company works. In the duty labels 41% of AI roles roll AI out, against 8% of ordinary roles. The words adoption, change management or upskilling appear in 60% of AI postings and in 31% of ordinary ones.

The new profession forming next to each of us turns out to be our own job, with one more duty attached.

What to do with this

If you are looking for work. Read the employer before the title. At consultancies and other employers the job is mostly bringing AI into the company: the product part is rare, 9% and 11%. At a tech company it can be either one or both, so read the duties too: 25% of AI roles there work only on an AI product, 36% only bring AI into the company and 27% do both. In my view it is worth preparing one story for the rollout part: what you rolled out, to how many people, and what changed in the numbers. Code matters mostly in automation, transformation and solutions roles.

If you search by title. Most AI titles are engineering roles: among AI roles of any kind that I collected, engineering roles are 89% of the AI titles in the UK, 75% in Germany and 94% in the US (roles counted once per company and title). Non-engineering AI roles show up when you search for them by name: AI Enablement, AI Transformation, AI Marketing. And at tech companies more than half of ordinary roles already mention AI, so it helps to search the text too.

If you are hiring. At other employers, from banks to hospitals, an AI role is mostly about bringing AI into the company: 69% of these postings ask the hire to set AI strategy and 61% to roll AI out. It may help to write the posting around those duties and to ask candidates what they have rolled out before. At a tech company AI roles split between the AI product and the company's own work, so it helps to say in the posting which of the two the role is, or that it is both.

How I might be wrong

• The career page data grew during the window. The counts behind the share went from 619 to 822 AI roles and from 10,060 to 13,859 openings. Most of that growth came from 31 of the 348 companies, where the number of postings jumped during the window. At the other 317 companies both counts went down a little, and the share was 61.7 and 60.9 per 1,000.
• My company list leans to tech. I built it around marketing roles, so tech companies are overrepresented. The tables compare employer types with each other for that reason.
• The AI marker reads words. An English-only marker undercounts local-language postings: it puts France at 22%, while a marker that also reads French gives 41%. That is why the country tables use English texts or a marker that also reads local words. With local languages counted, the US and Western Europe stay close at tech companies (73% and 68%) and at other employers (27% and 27%). At consultancies the gap grows to ten points, 56% and 46%.
• Company descriptions inflate mentions. Where I could not cut out the "About us" block, I cannot tell "we are an AI company" from "you will work with AI". At tech companies the real share of ordinary roles that mention AI may sit closer to 51% than 68%.
• Employer types are rough. A model checked 80 employers against their stated sector and company description and found 23 in the wrong type; I corrected those. Outside the small group of AI vendors the error rate was about one in five, 12 out of 58. The errors go both ways and blur the gap, so the real gap is likely a bit wider.
• Job families come from title words. On a check of 150 roles one in ten was in the wrong family and two more in ten were borderline.
• Duty labels come from models reading each posting with a written codebook. After an independent audit I tightened the codebook and labelled five main duties again. A separate labeller, without access to the existing labels, then agreed on each of them in 97% to 98% of 100 postings. For the other duties in Table 8 the agreement was 91% to 99%. All of these checks were models too.
• The labelled sample is built with quotas per job family, region and employer type: transformation roles are 24% of the labelled AI roles and 9% of AI roles on career pages. So it helps to read Table 8 as a comparison of its two columns.
• The rebrand test sees only part of the history. 58% of the AI roles in this test were already open on the first day I watched their company, against 51% of ordinary roles. A predecessor that closed before that day is invisible on both sides. Only 330 AI roles had both a usable text and an ordinary role with text closing nearby. The test ran on the full posting texts, which the open datasets do not include.
• The window is short. July to October 2026.

The data

All three datasets are open, with column descriptions: job postings for 12 countries, AI roles on company career pages, and the duty labels. Each package has summary tables behind the tables and charts in this article: mariascales.com/research/ai-jobs/data. The Indeed numbers come from Indeed Hiring Lab, CC BY 4.0.