Artificial intelligence is reshaping the job market at speed. It is creating new tools and even new roles, but it is also cutting the need for work that is repetitive, data heavy, and routine. That pressure is showing up first in entry level hiring, where companies often look for quick efficiency gains. At Reed, one of the larger employment agencies, listings aimed at junior candidates reportedly fell from more than 180,000 in 2021 to about 50,000 last year.
With numbers like that, it is no surprise that people keep asking whether any career is truly safe from automation. James Reed, the head of Reed, argues that the most resilient roles are the ones grounded in human connection, real world judgment, and accountability. In his view, it is not enough to simply “know” something when a job requires you to read a room, assess risk, make a call, and take responsibility for the outcome. Using job ad analysis and broader market movement, he singled out 11 roles he believes are far harder to fully automate.
The first group he highlights sits in health care and social care, where technology can support diagnosis and paperwork but cannot replace trust, empathy, and constant adaptation to individual needs. Caring for vulnerable people also involves responsibility that is difficult to hand off to a machine, especially when conditions change minute to minute. Projections for England suggest social care will need hundreds of thousands of additional workers by 2040, with Skills for Care putting that figure at around 470,000. In other words, demand is expected to keep growing even as tools get smarter.
Another role Reed points to is cybersecurity, because digital threats evolve constantly and attackers change tactics fast. Tools can detect patterns, scan logs, and automate some response steps, but high risk decisions still require human judgment and ownership. A breach is rarely just a technical puzzle, since it can involve legal exposure, business tradeoffs, and hard choices under time pressure. That blend of context and responsibility is part of why the field remains difficult to automate end to end.
Education also makes his list, especially teaching roles that depend on reading students and managing a classroom. AI can draft lesson plans, summarize material, and generate practice questions, yet it cannot replace the teacher who notices a student struggling and adjusts approach in real time. Discipline, relationship building, and the social dynamics of a room are central to learning outcomes. Those are human skills that rely on nuance rather than perfect data.
Emergency medical work is another category he calls out, because crises rarely arrive with tidy information. In an emergency, responders must act physically, choose priorities quickly, and keep patients calm while conditions shift. Reed’s point is that this combination is hard to package into a machine, since there is often no clean script to follow. It is a job where split second judgment and bedside reassurance matter as much as technical skill.
Reed also includes employment advisers and recruiters, which is notable because recruitment already uses algorithms heavily. He argues that strong recruiters evaluate motivation, team fit, potential, and ethical dilemmas, and that some of this work draws on intuition built through experience. Even when software helps screen resumes, the final call often rests on conversations and human signals. That is especially true when a hiring decision affects team dynamics and long term performance.
Real estate valuation is another role he sees as resistant to full automation, because valuation is not only a spreadsheet exercise. It often requires an on site visit, noticing risk factors, reading neighborhood nuance, and judging property condition. Those conclusions can carry legal and financial consequences, which raises the bar for accountability. A model can assist, but a person is still expected to stand behind the report.
Skilled trades such as electricians, plumbers, and carpenters are also on his list, largely due to unpredictable physical environments. Every home, breakdown, and installation presents different constraints, and the work happens in the real world rather than a controlled digital space. AI may help with planning and diagnostics, but the actual fix still requires hands on problem solving. As the argument goes, a robot that can reliably handle every odd corner case in a cramped basement is not here yet.
Hospitality and service roles round out another section of Reed’s picks, because customer experience depends heavily on human warmth and improvisation. Reservations and ordering can be automated, but guests remember how problems were handled when something goes wrong. A flexible response, a sincere apology, and a tailored solution are difficult to reduce to a script. Human contact is often the product as much as the service itself.
He also separates property agents from valuers, emphasizing how emotional buying or selling a home can be. Negotiation, trust building, and understanding personal needs are central in these transactions, especially when stress is high. People often want reassurance, not just information, during a major life purchase. That emotional component is hard to hand off to an algorithm without losing what clients actually value.
Early childhood educators appear on the list for similar reasons, with an added layer of safety, privacy, and regulation. Supporting young children requires emotional sensitivity and constant situational awareness, which cannot be replicated by a chatbot. There are also accountability requirements that place clear responsibility on adults. In fields involving children, trust is not optional, it is foundational.
Finally, Reed points to entrepreneurs, arguing that AI can supply data and ideas but cannot take the leap of accountability. Building a business demands deciding “at your own risk,” trusting intuition, and turning an unclear concept into something that works in the real market. Entrepreneurship also depends on persuasion, leadership, and resilience through uncertainty. Those are human traits that do not come from prediction alone.
More broadly, this debate sits inside a long history of technological change and shifting work. Automation has repeatedly replaced specific tasks while creating new ones, which is why many experts focus on skills rather than single job titles. Roles that combine social intelligence, physical presence, and responsibility tend to be more resilient because they demand context and trust. At the same time, many occupations will still change dramatically as AI tools become everyday coworkers rather than outright replacements.
If you are thinking about your own career, it can help to look at what parts of your work are routine and what parts rely on judgment, relationships, and real world problem solving. Learning to use AI tools can still be a big advantage, even in roles that Reed believes are safer, because productivity gains often go to the people who adapt first. The practical takeaway is not to panic, but to build skills that machines struggle to mimic, like empathy, negotiation, leadership, and decision making under uncertainty. Share your thoughts on which jobs you think will stay most human in the comments.
