Forget the Sci-Fi Version
The popular image of AI-driven job loss is a factory floor full of robot arms. The reality playing out right now looks almost nothing like that. The jobs most exposed to today's AI systems are office jobs - the ones built around reading, writing, summarising, and organising information, because that is precisely what large language models are best at. Meanwhile, many jobs long assumed to be at risk from earlier waves of automation, particularly skilled physical trades, have turned out to be some of the most resistant.
The Roles Under the Most Pressure
Entry-level and routine knowledge work has borne the brunt so far. Basic content writing and copyediting, first-line customer support, simple data entry and transcription, junior-level translation, and paralegal-style document review are all tasks that AI tools can now do a large share of, quickly and cheaply. This doesn't mean every job in these fields disappears - it means fewer people are needed to do the same volume of work, and the entry-level rung that used to train people into a career is getting noticeably thinner. Widely cited global labour projections estimate tens of millions of roles will be displaced by AI and automation this decade - a real and disruptive number, even though the same research projects a larger number of new roles emerging over the same period.
The Roles Holding Up Well
Work that depends on physical dexterity in unpredictable environments - electricians, plumbers, HVAC technicians, and other skilled trades - remains genuinely difficult and expensive to automate, and demand for these roles has stayed strong even as automation has spread through offices. Healthcare roles centred on hands-on care, from nursing to physical therapy, are similarly resistant, since patients and regulators alike still expect a human to be responsible for care decisions. Jobs that rely on earned trust and relationship - senior sales, therapy, high-stakes negotiation - are hard to replace because the "product" is partly the relationship itself, not just the information exchanged.
The Grey Zone in the Middle
A large number of jobs sit somewhere between "safe" and "at risk" - not eliminated, but fundamentally changed. Software development is a good example: AI hasn't replaced developers, but it has changed what a productive developer's day looks like, and companies increasingly expect fewer junior developers to produce the same output with AI assistance. The same pattern is showing up in marketing, design, and financial analysis - the job survives, but the skill set required to do it well shifts, sometimes quite quickly, and people who don't adapt fall behind people who do.
What Actually Predicts Risk
Across sectors, the single best predictor of AI exposure isn't the industry a job sits in - it's whether the job's core output is a fairly predictable text, image, or piece of code produced from fairly predictable inputs. The more a role depends on physical presence, situational judgement under uncertainty, or accountability for a decision's real-world consequences, the more resistant it currently is. That pattern is likely to shift over time as AI systems improve, but it is the clearest lens available today for assessing your own exposure realistically, rather than either panicking or assuming you're immune.
