In January 2016 I posted a list on Facebook of the professions most in demand in Italy. Digital Marketing was at the top, and I was pleased about it — a little vain, too, I imagine. It was the field in which I had decided to build my future, and the numbers seemed to be telling me I had chosen well.
In that same post I had written down five things, in this order: positioning, focus, strategy and objectives, tools, measurement. I had added a note too: the order isn't accidental. I would still write that today. Only now, the reason is different.
A fairly easy prediction
A few years later, shortly before the pandemic, I was in Milan in front of about a hundred entrepreneurs. I was explaining Meta Business Manager: screens, settings, campaigns, technical stuff. At one point I said something along these lines: the platforms' goal will be to remove people like me from the equation; sooner or later you'll choose what you want to achieve, how much you want to spend, and the algorithm will do most of the rest.
It wasn't a particularly brilliant prophecy. It was the natural direction of travel. If Meta and Google make money when businesses spend on their platforms, every intermediary required to make those platforms work is also friction to be reduced.
Today that direction is much clearer. In 2025 Reuters, reporting on the Wall Street Journal, described Meta's plan to progressively automate the creation and targeting of advertising campaigns by the end of 2026: the advertiser provides a product image and a budget, while the system generates creative assets and decides how to distribute them. That doesn't mean humans have already disappeared from the process; it means the industrial objective is to make an ever larger part of execution automatic.
Source: Reuters, June 2, 2025.
The part that interests me, though, isn't Meta. It's us.
For years we confused the work with the tools of the work
A large part of the economic value of many digital professions also came from how difficult execution was: setting up a campaign, writing a landing page, editing a video, building a presentation, analyzing a spreadsheet, producing a report, writing code, translating, putting together an initial piece of research or a first draft.
These were activities that took time and required a certain technical threshold to even begin. That threshold is dropping very quickly. When something that took four hours yesterday takes four minutes tomorrow, the work doesn't automatically disappear, but the reason someone is willing to pay you changes. That, to me, is the part we keep underestimating.
The most interesting data concerns the people who are supposed to be starting out
In August 2026 the Stanford Digital Economy Lab published a revision of its study Canaries in the Coal Mine?, based on administrative data covering millions of US workers and updated through June 2026.
The first finding matters precisely because it contradicts the easiest narrative: the researchers do not observe broad job displacement across the economy as a whole. What they do observe is a specific divergence. Among workers aged 22–25 in occupations highly exposed to AI, employment is about 19% lower than where it would have been had it followed the trend of their peers in less-exposed occupations. The difference appears to show up mainly through lower hiring rather than a wave of layoffs; it is also concentrated in jobs where AI tends to substitute for human tasks, while the picture is different in occupations where AI is used primarily to augment people.
The authors themselves treat these results as empirical signals to be interpreted cautiously, not as proof that every observed employment change is caused by AI.
Source: Stanford Digital Economy Lab, revision of August 12, 2026.
For me, the most interesting question starts right there. I didn't begin my career doing strategy, and I don't think almost anyone begins with the most complex work. First you do simple things badly, then you do them better; you make mistakes, watch someone more capable, copy them, discover why copying isn't enough, and accumulate cases and contexts. At some point we call all that sedimented material experience.
If AI becomes very good at precisely the tasks through which we normally built that experience, then the question isn't only how many jobs it will destroy. It is also how we will train the people who, one day, are supposed to do the work we don't want to hand over to AI. If you remove the first rung of the ladder, you haven't automatically created more senior people.
Big numbers need to be read for what they actually measure
The World Economic Forum's Future of Jobs Report 2025 is often summarized badly too. It does not say that artificial intelligence will eliminate 92 million jobs. That estimate covers the full set of technological, economic, demographic, geopolitical and environmental transformations considered by the report: by 2030, around 170 million roles could be created and around 92 million displaced, for a net gain of 78 million.
In the same report, however, 41% of the employers surveyed say they expect workforce reductions where AI can automate specific tasks. That isn't a future carved in stone either: these are expectations reported by the companies surveyed. But it is enough to understand that we are not talking about just another piece of software.
Sources: World Economic Forum, Future of Jobs Report 2025 and summary release.
The same applies to another number that became very effective in headlines. In 2026 Mark Ritson commented in Adweek on an Anthropic analysis according to which around 65% of the tasks performed by marketing professionals could, in principle, be replaced by AI operations. Tasks, not jobs. The difference is enormous: a profession is a bundle of activities and, when some of them are automated, that job may disappear, shrink, change or become more productive.
Source: Adweek, “65% of Marketing Jobs May Not Survive AI”.
What is still ours?
For years we reassured ourselves by saying machines would do the repetitive work and leave creativity to us. Then we built machines that write, draw, compose music, program, produce video and suggest ideas. So we started moving the line: creativity, empathy, strategy, intuition.
I don't know where that line will be ten years from now, and I distrust anyone who claims to know. What I do know is that, today, there are still things that are not the same as executing a task: judgment, accountability, context, experience that has never been written down, the ability to understand which problem we are actually trying to solve, and the relationship between two people who need to trust each other.
Then there is something that takes time precisely because it cannot be produced all at once: reputation.
Adobe, in its 2026 report on consumer behavior, found that one third of respondents would stop engaging with a brand if they discovered that content they believed was human had been generated by AI. That doesn't mean people reject artificial intelligence across the board; it means transparency and trust are still open questions.
Source: Adobe, AI and Digital Trends: Customer Behaviors and AI, 2026.
And there is a paradox here that I find interesting: the cheaper it becomes to produce words, images, videos, analysis and software, the less the mere fact of producing them proves anything. Publishing a book used to be a barrier; having a website was a barrier; shooting a decent video was a barrier; building software was a barrier. Many of those barriers are collapsing, and that is largely a good thing. But it also means the final product can be excellent while telling us very little about the competence of the person who generated it.
That is why trust will probably have to rest increasingly on something else: history, consistency, accountability, evidence and continuity.
So are we already dead?
Part of what I did professionally ten years ago is already dead. I see no reason to pretend otherwise just because someone used to pay me to do it.
If a machine can perform an activity better, faster and at an enormously lower cost, defending that activity out of nostalgia would be ridiculous. The point is to understand whether we were the same thing as that activity.
If all I sold was the ability to press certain buttons, I have a problem. If my value also lay in understanding which button to press, when not to press it, why to press it and in taking responsibility for the consequences, the conversation changes. That doesn't mean I'm safe; it means I need to keep moving toward work that requires more understanding and less mere execution.
In 2016 I wrote that the order of my five points wasn't accidental. Ten years later, the tools are learning to do an ever larger share of what came after them. Maybe that is precisely why what comes before has become even more important: understanding where to go, understanding why, and having someone willing to answer for the choice.