Does AI take your job? The right question is a different one

In brief. The question "does AI take your job" is badly framed: an occupation is a bundle of different tasks, and AI absorbs some of them, not the whole package. At an energy reseller we worked with, the task that disappeared was keeping in mind who to call back: the salesperson stayed, and second contact on hot leads went from 50% to over 90%. The 2026 data (Anthropic Economic Index) confirms that the unit being automated is the task, not the occupation, with different effects role by role. The useful question for a company is not whether to cut, it is which task inside the job should be removed because it was never the job.

An energy reseller, twelve people, a mundane problem

Before we arrived, the salespeople at an energy reseller with twelve people on the team called leads from memory. Who had been contacted last, who was hot, who needed calling back today: it all lived in each person's head, or in a spreadsheet nobody updated consistently. The oldest leads rotted at the bottom of the list without anybody noticing. The measured result was that half the leads they had paid for never got a second call.

Think about where that lead goes: the company bought it, somebody paid for it, and it ends up forgotten, not because the salesperson is lazy, but because nobody can hold hundreds of priorities that change every day in their head.

This is the point where the question "does AI take salespeople's jobs" shows its limits. It is not the question that helps. The useful one is: which piece of that job was never the job?

Tasks, not occupations: why the distinction changes the answer

An occupation is not one solid block. It is a bundle of different tasks stacked together. The reseller's salesperson did sell, certainly, but before that they sorted a list, remembered the status of every contact from memory, decided who to call first with a mix of instinct and luck. Only the last part, the conversation with the customer, was really the job.

With MARF, the task of sorting and remembering passed to the system. The salesperson no longer decides who to call by feel: the system says, based on who is hottest and how long they have been waiting. The salesperson concentrates on the conversation, not on organising the list. Second contact on hot leads went from 50% to over 90%, with the same number of people on the team.

The task that disappeared, holding every lead's status in your head, was not the value that company sold its customers. It was the work around the job, not the job. The task that stayed, the conversation, is exactly what that company pays a salesperson a salary for, and with more time free from sorting they do it better.

The task that passes to the system. Not "who calls better", but "whose turn is it on the phone now". The decision stays with the salesperson, the memory does not.

This distinction, between an automatable task and the real job, is more useful than any forecast about how many positions AI will remove or create. And it is verifiable, because we watched it happen in a real company, not on a projection sheet.

What the data says, with the caution it needs

On the net balance of jobs, nobody has a crystal ball, and whoever promises a precise number is selling smoke. The most solid picture today is the one that looks at the task, not the whole occupation. The Anthropic Economic Index, in its January 2026 report on "economic primitives", measures precisely this: which tasks a model covers inside a role, and at what level of required skill. The interesting result is not "how many jobs disappear", but that the same type of task, removed from a role, can lower the skill required in one case (a travel agency that loses complex planning and is left with ticket issuing alone) and raise it in another (a property manager who loses routine bookkeeping and is left with negotiation and stakeholder management). Same type of automation, opposite effect, depending on what is left of the job.

The following report, from March 2026 on labour market impacts, adds an honest caveat worth quoting in full: there is no evidence of a systemic rise in unemployment among the workers most exposed to AI from late 2022 to today, but hiring of workers aged 22 to 25 in the most exposed roles slowed by around 14% after ChatGPT arrived, without a similar slowdown for more senior workers in the same roles. The pain, where it exists, concentrates at the entry door of a career, not evenly across everybody.

In Italy company behaviour confirms the direction: AI adoption among businesses went from 8.2% to 16.4% in a year (Istat, 2025). The main brake for those left behind, in that data, is skills, not fears about jobs.

How to spot the task at risk in your own role

Before fearing for a whole occupation, look inside it, task by task. A practical way:

• Do you do it from memory or habit? If you are holding a status, a priority, a list that changes every day in your head, that is the first candidate for automation. It is not where your value sits, it is the noise around it.
• Does it require reading a person or deciding under ambiguity? Negotiating a price, calming an angry customer, judging an edge case: here AI stays an assistant, not a replacement.
• Do you take responsibility for the outcome? If the final signature, the choice and its consequences are yours, that task does not get delegated to a system, it gets supported by one.

At the energy reseller the first point was sorting leads. In your role it may be something else: the first draft of a document, data entry, repeated reporting. The method for finding it does not change: separate what you do out of habit from what you do out of judgement.

What a company can do, besides worrying

If you run a company, the operational question is not "how many people do I cut", it is "where is my team losing time on a task that is not their job". Three concrete moves, in the order we do them with clients:

• Measure where value is lost, before buying any tool. A precise Value Leak is worth more than a feeling of inefficiency.
• Automate the task, not the role. A system with a clear scope (what we call an AI Employee) takes on the sorting or the repetition, and leaves the judgement to people.
• Train an internal reference point per department, the AI Champion, instead of a generic course for everybody. That is how the practice stays in the company even when the supplier or the tool changes.

Since 2 February 2025, article 4 of the European AI Act requires organisations to ensure an adequate level of AI literacy among the people working with it. It is no longer just good practice, it is a requirement.

The position, not the moral

AI does not take whole occupations, almost ever. It takes the tasks that occupied your head without generating value, and leaves, often strengthened, the part that occupation exists for. The real risk is not being replaced by a system, it is being the last one doing by hand a task the rest of the market has already stopped doing by hand.

The real risk is not being replaced by a system. It is being the last one doing by hand a task the rest of the market has already stopped doing by hand.

If you want to see where your team is losing time on tasks that are not your job, look at the AI Champion programme: it is how we bring this distinction into a real company, department by department. If instead you prefer to start from the number, you can quantify in euros where the task to remove is hiding with the ROIometer.