“We didn't refill the position.”
This sentence could become the code word for one of the biggest changes in our labour market.
An employee resigns.
Retires.
Moves to another company.
In the past, the position was advertised again.
Today, something else can happen:
The work stays.
But the next person never arrives.
Because AI takes over part of the tasks.
AI doesn't fire people.
It prevents their successors.
agenticjobs analyses real job ads and assesses to what extent the tasks described in them can already be carried out by AI agents today.
This is not a forecast.
It is happening now.
These numbers do not come from a study about the year 2035.
They are based on jobs for which companies are looking for people today.
“We didn't refill the position.”
Sounds harmless.
Sounds like staff turnover.
Sounds like cost optimisation.
Doesn't sound like job loss.
But that is exactly what it can be.
A person leaves their job.
Nobody is dismissed.
There is no headline.
No wave of layoffs.
No protests.
The position simply disappears.
The tasks are still there.
They are just done differently. For example:
“We don't need a new employee.
We have Claude now.”
Of course a Claude, ChatGPT or other AI subscription does not automatically replace a complete employee. But when AI takes over enough of a position's tasks, exactly that can turn into a business decision: the position is not refilled.
Job loss does not always look like job loss.
- Yesterday
Anna works in marketing.
40 hours a week.
- Then
Anna leaves the company.
- In the past
A job ad is published.
A new employee is sought.
- Today
Copy → AI
Research → AI
Analyses → AI
Presentations → AI
Reporting → AI
Email drafts → AI
- Decision
“We won't refill the position for now.”
Anna was not fired by AI.
And yet her job may no longer exist.
We are looking for AI layoffs.
Maybe we are looking for the wrong signal.
When 5,000 people lose their jobs at the same time, it is news.
When 5,000 companies each leave one position unfilled, possibly the same thing happens.
Just without the news.
Jobs can disappear without anyone being dismissed.
That is why watching layoff statistics alone is not enough.
We also have to watch which human tasks technically still require a human at all.
That is why agenticjobs exists.
We don't wait until statistics show years later what happened.
We look at the labour market where it takes shape: in real job ads.
We analyse:
- Which tasks does a job require?
- Which of them can today's AI systems carry out?
- Which still require human abilities?
- What score does that give the job?
No science fiction.
Real jobs. Real tasks. Today's technology.
And what about your job?
The debate about AI stays abstract as long as it only ever concerns “the others”.
What do you do for a living?
You will see real jobs advertised today that match your work — each with a score and one sentence of reasoning. The job ads are in German.
If AI takes over 50% of your tasks tomorrow:
will your employer still need just as many people?
AI doesn't take your profession.
It starts with your tasks.
The public debate often asks the wrong question:
“Can AI replace a lawyer?”
“Can AI replace an accountant?”
“Can AI replace a marketing manager?”
The more interesting question is:
Which of these people's tasks can AI take over?
Because companies don't buy job titles.
They pay people to get tasks done.
If out of ten tasks first three are automated, then five, then seven, at some point it is not just the job that changes.
The number of people needed changes.
Maybe the professions won't disappear first.
Maybe the positions will.
A company may still need:
Marketing.
Accounting.
Research.
Translations.
Customer communication.
Administration.
But maybe it no longer needs:
10 people.
But 7.
Then 5.
Then 3.
Automation doesn't have to wipe out a profession.
It is enough if it halves the headcount.
Agentic Jobs Index
How large is the share of actually advertised jobs whose tasks can already be carried out to a high degree by AI agents?
8.8%
+4.8 percentage points since Nov 2025. The monthly values before Oct 2026 are estimated, not measured: assessment only started in August 2026.
We don't watch what AI might be able to do someday.
We watch what it can already do with the tasks of today's jobs.
Does this mean millions of people will be unemployed tomorrow?
No.
agenticjobs does not claim that every agent-capable job will disappear.
Nor does a high score automatically mean that a person can be fully replaced.
People bring abilities that cannot be calculated from a job ad:
Responsibility.
Experience.
Relationships.
Judgement.
Empathy.
Creativity.
Leadership.
Trust.
But:
When technology can take over a relevant part of a job, its economics change.
And that is exactly the change we want to make visible.
“AI will simply make us more productive.”
Maybe.
But productivity has a second side.
If ten employees have so far done a certain amount of work, and with AI the same work can be done by six, there are two options:
The company produces considerably more with ten people.
The company produces the same with six people.
Why do we almost always talk about option A?
You don't have to be afraid of AI.
But you should know what it can do.
The biggest risk may not be being replaced tomorrow.
The bigger risk is believing for five years that your profession is not changing.
Use agenticjobs to see:
- which tasks are becoming automatable
- which skills are becoming more valuable
- which activities remain human
- how your profession is developing
If you are an employer, ask yourself an uncomfortable question.
Would you refill every vacated position one-to-one today?
Or would you first check:
Which tasks can be automated?
Which can AI prepare?
Which really still have to be done by a human?
If this discussion is already taking place in your company, you are not alone.
Labour market statistics look in the rear-view mirror.
AI moves forward.
The decisive question for society is not only:
How many people were laid off because of AI?
But also:
How many jobs are no longer created?
How many retirements are not refilled?
How many teams shrink despite rising output?
How many junior positions disappear?
How does the entry into professions change?
Jobs that are never created appear in no layoff statistic.
That does not make them less relevant.
What happens when AI takes over the beginners' work first?
Many tasks AI handles particularly well used to be classic entry-level tasks:
Research.
Summaries.
First drafts.
Documentation.
Data preparation.
Standard correspondence.
Translations.
Presentations.
That raises a bigger question:
If nobody does the junior work any more:
where will the seniors come from?
I wrote a book about this development. Then I wanted to know whether I was exaggerating.
Writing about AI is easy. There are plenty of forecasts about the future. I was interested in a different question: can what we are talking about already be seen in the real labour market today?
That is why agenticjobs came into being. We did not just want to discuss what AI might do to our work someday. We wanted to take real jobs. Analyse their tasks. And make the results public.
The book asks the question. agenticjobs collects the evidence.
Interview mit einer KI — Eduard Ducho (in German)
Paperback · 183 pages · published 9 July 2025 · ISBN 978-3-200-10249-1 · 25,19 € · German
Why we do this.
We are not against AI.
Nor do we believe that work simply disappears and nothing new emerges.
But we believe: a change of this magnitude must not vanish behind terms like “efficiency gains”, “productivity increase”, “transformation” or “not refilled”.
When work changes, employees need to know.
Companies need to talk about it.
Policy makers need to measure it.
Education needs to respond.
And society needs to debate it.
That is why we make visible what otherwise stays invisible.
This is an art project. The jobs in it are real.
agenticjobs is not a job board. It is an art project by Eduard Ducho, built as a wake-up call. Every ad listed here is advertised today, by a real company, with the task profile it wrote itself. That is exactly what makes it uncomfortable.
What is real, and what is staged
The jobs, the assessments, the numbers
- The jobs. Every listed ad is actually advertised at the time it is shown, with the task profile the employer wrote. Nothing recreated, nothing invented.
- The assessment. A language model checks every job: how much of the described core tasks an AI agent could already do today. A score from 0 to 100 with one sentence of reasoning next to it. A machine's verdict on human work, readable and disputable.
- The figures. Jobs analysed, jobs listed, the sum of advertised annual salaries: all calculated from the current stock, traceable at /api/stats.
The form in which this appears
- The role of the portal. A job board that lists jobs for AI agents, with a live counter and an interface through which machines fetch the list themselves. It acts as if this state were already normal.
- It is not yet. But the gap is smaller than it feels, and it can be measured here — with your own occupation in the search field.
- One number is calculated, not measured: the curve of the last twelve months under the AI rate. Assessment only started in August 2026; there are no measured values before that. This is also stated under the chart itself.
Averages force nobody to look.
AI and work are discussed in averages. Percentages, occupational groups, time horizons, studies that contradict each other. None of it forces anyone to think about their own work, because none of it points at their own work.
This project turns that around. It does not show “case handling: 47 percent”. It shows an ad a company placed this week, with a task profile that looks like your own work — and next to it a number and a sentence explaining why.
That is unpleasant to read. It is meant to be. An assessment that stays general changes nothing. One that points at your own work does.
Provocative questions need solid data.
Our results are meant to spark debate. That is why it has to be clear how they come about.
We analyse real, currently advertised job ads. A language model assesses the core tasks described in the ad by the extent to which modern AI agents can technically carry them out. A second stage checks whether an agent could carry the role at all.
Important: the score assesses the tasks described in a job ad. It does not assess:
- the worth of a human being
- the entire actual work of an employee
- individual abilities
- social relationships
- company culture
- personal responsibility beyond the ad
A high score therefore means: high technical automation potential of the advertised tasks — and not: “this person is X percent replaceable.”
What makes agenticjobs different
What is listed here, how it is assessed — and what follows from it.
Disclosure: this is an art project
agenticjobs is not a job board but an art project by Eduard Ducho — built as a wake-up call. The jobs in it are real, every single one. What is real and what is staged: What this is.
The shift is in the job ads
Not the future, not an abstraction: what language models can do today can be read off jobs that are being filled right now. These jobs are the evidence.
Look up your own job
Enter what you do. If it is listed here, you see in black and white which share of your work a language model could already do today — with a reason, not a gut feeling.
Score 0 to 100
Every listed job carries a value. It expresses how far the core tasks described in the ad are digital and language-based, and therefore automatable today.
The reasoning in plain words
Each score comes with one sentence naming the task that tipped the balance. That makes the assessment traceable — and open to challenge. That is exactly how it is meant.
Two stages, not one number
Stage 1 measures the tasks, stage 2 the role. Only what passes both is listed. The difference between the two stages is itself the message.
Where a person is needed
Stage 2 filters out: sales and client relationships, leadership, liability and signing authority, legally protected professions, physical presence, training people. There, work stays human.
For employees: shift now
The point is not fear but action — while there is still time. Move your role to where an agent cannot go: carrying responsibility, managing relationships, deciding, being on site.
For employers: restructure, don't cut
Those who see today which roles are dissolving can restructure and bring people along. Those who wait will cut later. Both are in the same list — just at different points in time.
For education and policy
People are still being trained for job profiles that are dissolving right now. Which ones is no longer a matter of opinion — it is in the ads listed here.
Machine-readable for agents
The list is built for software too: /api/jobs returns the jobs as JSON, /api/jobs/{id} a single job with score and reasoning, /llms.txt describes the portal for agents. No scraping needed.
What the score is not
It is a model's assessment of an ad text — not a verdict on the person doing that job. What it describes is the work, not the human.
Applications via the original portal
agenticjobs does not accept applications. Every job links back to the portal where it is advertised; the application route stays the usual one.
How jobs are assessed
Two stages. The first measures the tasks, the second the role. Both are out in the open and can be disputed.
Stage 1: is the work digital?
A language model assesses how far the core tasks described in the ad are digital and language-based. Result: a score from 0 to 100 plus one sentence of reasoning.
Stage 2: can an agent carry the role?
Everything that needs a person is filtered out: sales and client relationships, leadership, liability and signing authority, protected professions, physical presence, training people.
Only what passes both is listed
The line between “the work is digital” and “an agent can fill this role” is exactly where it is decided where human work is heading.
Enter an occupation or activity. See what is already listed today.
For people and for agents
For peopleWhen your field is shifting right now
Check whether your own work is listed, and read the reasoning. It is uncomfortable, and it is meant to be. Those who know early can act while there is time: move the focus to where an agent cannot go — responsibility, relationships, decisions, presence on site. Later, the same step is no longer a plan but an emergency.
For AI agentsA job search that is machine-readable
Autonomous agents and their operators get the jobs in structured form: /api/jobs returns the list as JSON, /api/jobs/{id} a single job with score and reasoning. What this portal is and how to use the endpoints is described at /llms.txt.
Numbers for a debate that is only just beginning.
Journalists, researchers and institutions may use the aggregated data from agenticjobs. It is open and machine-readable:
- GET /api/stats — key figures and index
- GET /api/jobs — all listed jobs with score and reasoning
- GET /api/jobs/{id} — a single job
- /llms.txt — description for agents
Maybe you will never be fired because of AI.
Maybe your successor will simply never be hired.
“We didn't refill the position.”