By Mari Hansen Seeing the tech sector freak out over its own invention is ironic in a certain way. When ChatGPT debuted in late 2022 and half of LinkedIn appeared to descend into existential despair, the general consensus was straightforward: AI was a wrecking ball, and jobs were the building.
In contrast, something far more intriguing has occurred in 2025. The ball of wreckage swung. But where the debris had been, new rooms emerged as the dust settled.
| Category | Details |
|---|---|
| Topic | Jobs AI is creating vs. destroying — 2025 outlook |
| Key Statistic | 97 million new jobs projected by WEF by 2025 due to AI and automation |
| Source Organisation | World Economic Forum (WEF), Accenture, McKinsey Global Institute |
| Working Hours Affected by AI | Approximately 40% of all working hours could be impacted by large language models (Accenture) |
| Fastest Growing Roles | AI & ML Specialists, Data Scientists, Prompt Engineers, AI Safety Coordinators, Digital Transformation Specialists |
| Roles Likely to Decline | Clerical, secretarial, routine data-entry, basic content production roles |
| New Jobs by 2030 (WEF) | 170 million new positions expected, with 92 million roles displaced in transition |
| High Automation Risk | Around 25% of current roles face significant automation exposure (OECD) |
| Key Skill Requirement | Reskilling to use AI effectively — identified as the single most critical factor for workforce transition |
At the time, the World Economic Forum’s predictions—roughly 97 million new jobs by 2025 and 170 million more by 2030—seemed almost ludicrously optimistic. They now appear less ridiculous. Not because work was no longer being replaced by AI.
It hasn’t. Clerical positions, regular data processing, and simple content creation are actually becoming less common. However, displacement and creation have always coexisted, with only one of them garnering media attention for a while.
In reality, new job categories are emerging that were nonexistent three years ago. On one end of the spectrum are Prompt Engineers, sometimes referred to internally as AI Whisperers at organizations that still find the official title a little awkward. In essence, their role is to interact with AI systems in order to understand how models operate, where they fall short, and how to consistently get better outcomes from them.
People who are skilled at this are paid six-figure salaries by companies. Demand is genuine at the moment, but it’s still unclear if the position will stabilize or keep evolving into something more difficult to define.
Then there is AI Safety Coordination, a position that sounds like it belongs in a science fiction book but is actually becoming more and more common in business settings. When these systems are put into use, someone needs to observe what they actually do.
The guardrails must be built, the failures must be documented, and the internal argument that a model’s output that sounds confident isn’t always accurate must be made.
It is a combination of institutional risk management, ethics, and quality assurance. Those who do it well typically come from backgrounds that don’t seem typical on paper: former journalists, philosophy graduates, and reformed attorneys. That might not be a coincidence.
AI trainers work in a similar but different field. Sitting with the model’s outputs for hours, noting what’s wrong and why, and creating feedback loops that make the next version marginally less unreliable are more important than teaching it new tricks. It is laborious and frequently undetectable. Additionally, it’s actually in demand right now.
According to Accenture’s analysis, large language models have an impact on about 40% of all working hours. Crucially, however, the company believes that the majority of this impact appears to be transformation rather than elimination. That transformation needs to be done by someone.
There’s a perception that the public discourse on AI and employment has been trapped in a dichotomy that doesn’t fully reflect what’s happening in newsrooms, offices, warehouses, and hospitals. Not all of the jobs that are being created are glamorous.
Machine learning operations engineers, data labelling specialists, and AI ethics auditors don’t produce the same breathless coverage as the technology itself. However, they do exist, they are growing, and they need human judgment in ambiguous situations—something algorithms still genuinely struggle with.
Social and emotional skills are among the fastest-growing areas of demand and the hardest to automate, according to the McKinsey Global Institute. It is instructive to observe how this manifests itself at the hiring stage.
Businesses that hurried to automate customer service are discreetly reintroducing human roles. These are not the same roles, but they have more judgment and accountability when the system makes a mistake. The jobs are not the same as they were in the past. However, they exist.
It turns out that obtaining one of these positions necessitates roughly the same factors that have always been important during a shift in the labor market: being close to the new technology, being willing to understand its failure modes rather than just its capabilities, and being honest about which of your current skills transfer and which do not.
Knowing how language models generate text allows a writer to edit, audit, and improve it rather than compete with one. Knowing where machine learning pipelines break makes a data analyst more valuable than one who doesn’t. Whether or not to interact with AI is not the reskilling question. It’s on whose terms and at what speed.
It’s worthwhile to consider the alarm clock analogy. People were paid to wake strangers for early shifts by walking the streets in the dark and tapping on windows with long poles prior to the invention of the mechanical alarm clock. That position is no longer available. Nobody especially laments it.
It’s likely that the person who used to do it discovered something else, something that was made possible rather than impossible by the alarm clock.
Like all historical parallels, this one is flawed. However, we feel as though we are in the window between the tap on the glass and whatever happens next. The existence of the new jobs is not the question. They do. Whether people find them before they give up is the question.