Here is the direct answer, because I know you came for it: some jobs are genuinely resistant to AI automation, and it is not random which ones. The jobs that hold up share four structural traits — physical unpredictability, real-time emotional stakes, legal accountability, and genuine creative judgment. Most jobs have some of all four. The ones that are truly safe have most of all four.
This post covers which jobs, why each one resists, and what to do if you run a business and want an honest picture of where your team stands.
TL;DR
If your job requires you to show up in an unpredictable physical space, read a room in real time, carry personal liability for the outcome, or make a call where nobody has written the playbook yet — AI is not replacing you. It is, however, going to handle your admin while you get on with the actual work. That is a good trade. Take it.
Photo: Pexels
What AI is already replacing
Start here, because pretending otherwise is how people end up surprised. AI is not looming on the horizon. It is already doing this, right now:
- Data entry and document classification — anything where the input is structured and the output follows a pattern
- Templated writing — standard reports, first-draft emails, meeting summaries, product descriptions
- Basic customer service — scripted queries, appointment scheduling, FAQ responses
- Junior research work — compiling information from multiple sources into a structured summary
- Transactional legal work — document review, contract comparison, first-draft standard agreements
- Boilerplate code — CRUD operations, test coverage for well-specified functions
Nine times out of ten, when a business owner tells me AI is going to replace their team, what they actually mean is that AI is going to do the things their team used to hate doing. That is not a replacement. That is a gift — if you manage it correctly.
In 100+ automation projects since 2018, the pattern I keep seeing: the employees who are "replaced" by automation were doing tasks, not work. The ones who were doing actual work — judgment, relationships, accountability — are still there. Usually faster, and considerably less annoyed.
The invoice lady and what she taught me
I want to tell you about one project before I get to the frameworks, because it is the best illustration I have of what job security actually looks like in the age of automation.
A mid-sized company I worked with was processing 30–50 invoices per week the old way: print, document manually, then a finance employee walked around the office physically collecting approvals from different departments. Every single invoice. Every day.
When I showed her the automation — digital approvals, no more walking — she went quiet. Not impressed quiet. Worried quiet.
She thought the automation was going to take her job. Or worse: take away the only part of her day where she actually talked to people.
I told her: "You will have way more time to talk to people now. You will just do it after the invoices are processed, not instead of processing them."
After the demo, she got it. She now handles 50–100 invoices per week — double the volume, same person, no more walking. About 15–20 hours a week freed up for actual relationship work. Her job did not disappear. It got better.
I reckon "AI will replace your job" is lazy fear-mongering. The real story — the one backed by the 100+ implementations I have watched up close — is that AI replaces tasks, not people. People who learn to work alongside it get stronger. Everyone else panics and waits to be surprised.
Right. Frameworks. Here we go.
The 4 skills AI cannot replace (in any role)
Before the job-by-job breakdown: the skills AI cannot replace are not job-specific. They are structural. Any role that relies heavily on these four is resistant to automation — regardless of industry. Understanding the framework matters more than memorising the list, because the specific safe jobs may shift as technology develops, but the underlying skills do not.
Physical presence in unpredictable environments
A plumber diagnosing a leak in a 1930s building navigates corroded pipes, unexpected access constraints, water damage that matches no blueprint, and a homeowner standing two feet away asking questions. No robot does that reliably at scale in 2025, or in 2030.
Real-time emotional judgment with genuine stakes
A therapist reading a patient who is not saying what they mean. A teacher who notices a child has stopped engaging and figures out why. A crisis negotiator adjusting their tone based on micro-signals. These are not tasks that decompose into a prompt.
Moral and legal accountability that cannot be offloaded
A doctor who misdiagnoses is liable. A judge who rules wrongly is subject to appeal. An architect whose building fails carries professional consequences. AI systems cannot hold professional accountability — and the legal frameworks that govern high-stakes professions require a human to own the outcome.
Creative direction under genuine ambiguity
There is a meaningful difference between asking AI to write a blog post and asking it to figure out what a brand should stand for and what it should never say. The first is a task. The second is a judgment call that requires synthesising incomplete information and making a bet. AI assists the execution. It cannot make the strategic call.
Photo: Pexels
Healthcare workers
Surgeons, nurses, paramedics, physiotherapists. These jobs score high on all four pillars simultaneously, which is why the McKinsey Global Institute consistently identifies healthcare as one of the sectors with the lowest automation potential.
AI is already reading scans, flagging anomalies, and drafting discharge summaries. What it is not doing is placing a hand on a patient's shoulder and making a judgment call about whether to proceed with surgery on a 78-year-old with three complicating factors. That combination of physical unpredictability, emotional load, and legal accountability is genuinely hard to decompose.
The US Bureau of Labor Statistics projects healthcare employment to grow 12.6% between 2021 and 2031 — adding roughly 2 million positions — during the exact period when AI in healthcare is advancing fastest. The two trends are not contradictory. They are both true at the same time, which tells you something useful about how this actually works.
Photo: Pexels
Skilled tradespeople
Electricians, plumbers, HVAC engineers, carpenters. These are the jobs economists consistently underrate because they do not require a degree. They are also the ones that consistently fail to automate for one simple reason: every job is different.
An electrician rewiring a listed building from the 1950s encounters something unexpected every hour. A plumber fixing a burst pipe works in a physical space that was never designed with their access in mind. The robotics needed to handle that variability at scale — across the full diversity of real-world environments — does not exist and is not close.
Meanwhile, the skilled trades in most developed economies are already experiencing labour shortages that AI will not solve. (If you have ever watched a plumber contort themselves into a space a person should not physically fit into and somehow make it work — that is not a job description. That is a superpower. Skynet is not ready for it.)
Photo: Pexels
Teachers and educational counsellors
AI tutoring tools are genuinely good at information transfer. Patient, available 24 hours a day, infinitely scalable. They are not, however, a teacher who notices that the student who was engaged last week has gone quiet, connects it to something the form tutor mentioned about a situation at home, and adjusts the entire dynamic of a lesson accordingly.
Teaching — actual teaching, not just information delivery — is fundamentally relational. The same is true of educational counselling, special needs support, and the pastoral care that holds together the experience of being in an institution. AI handles the content layer adequately. The human layer — the one where someone decides a child needs a different approach today than they needed yesterday — is not a prompt. It is a skill built over years of watching people.
Photo: Pexels
Mental health professionals
AI companion apps are increasingly sophisticated. Some people find real value in them and I am not dismissing that. They are not a replacement for a trained therapist working with someone through serious trauma, addiction, or psychiatric crisis — and the professional bodies that govern mental health practice are not going to permit it to become one anytime soon.
Beyond the regulatory protection, there is a functional one: the therapeutic relationship is the mechanism. It is not a container for information. A therapist who can be present, who can sit in difficult silence, who notices what a patient is doing with their body while they talk — that is not a language model capability. It is a human one, built slowly, over real interactions with real people.
Trial lawyers and judges
Legal research? Being automated. Document review? Already done. Contract generation? Largely automated, and legitimately so — it was always overpriced pattern-matching that required a qualified lawyer only because that was the rule, not because the task required judgment.
What is not being automated: advocacy, judgment, accountability. A trial lawyer reads a jury, adjusts an argument in real time, and makes calls under pressure where the stakes are a person's liberty or livelihood. A judge applies legal principle to a specific combination of facts that has never existed in exactly this configuration before. AI can process precedent at scale. It cannot carry the moral weight of a sentencing decision. Those are different things.
Photo: Pexels
Software developers — the nuanced one
This is the category where I get the most pushback, so I want to be specific. Junior developers writing boilerplate code are already being compressed. AI coding assistants are genuinely good at well-defined problems. That is a real and current pressure on entry-level developer roles.
What they cannot do: architect a system from genuinely ambiguous requirements, make trade-offs between competing constraints, or take accountability when production goes down at 2am. Senior engineers and architects are safer. The gap between junior and senior is widening.
The rule of thumb: the more your development work is "translate a clear specification into code," the more exposed you are over the next three years. The more your work is "figure out what the specification should be, then own the consequences of getting it wrong" — the safer you are. Build toward the second kind.
Photo: Pexels
Business strategists and consultants
Here is the one where the nuance matters most — and where I will be honest about my own lane, since I am a consultant writing about consultants.
Research, report-writing, status updates, first-draft strategy documents — all of this is being automated within consulting. If your value proposition is "we will research this and tell you what we found," that proposition is eroding. It used to cost money because it required people-hours. It no longer does to the same degree.
What is not being automated: the judgment call in a room where everyone has incomplete information, two of the reasonable options are politically impossible to implement, and someone has to make the call and own the consequences. That is what leadership is. It is also what good consulting is, when it is being done properly rather than when it is being done as a cover for delivering a deck that nobody acts on.
In my experience, the consultants and managers who survive every technology shift are the ones who compete on judgment, relationships, and accountability — not information access or data synthesis. AI has not changed that rule. It has just made the consequences of ignoring it faster.
What to do if you run a business
The relevant question for most business owners is not "is my job safe?" It is "which parts of every role in my team are pattern-based, and which require genuine judgment?"
Here is the practical audit. For each role on your team, estimate the split between:
- Pattern-based, repeatable work — data entry, report compilation, standard emails, templated analysis, scheduled admin
- Judgment, relationship, and physically complex work — client calls, strategic decisions, technical problem-solving in the field, mentoring, anything where the right answer depends on context
Roles that are 70%+ pattern-based will look significantly different in three years.Not necessarily gone — but smaller, faster, or done by one person instead of three. Roles that are 70%+ judgment and relationship work are largely stable — but the people in them will be faster if they learn to use AI tools for the other 30%.
Honestly, the businesses I worry about are not the ones with high automation exposure. They are the ones that are slow to automate the parts that can be automated, because those are the ones whose competitors will be operating at meaningfully lower cost within 24 months. That is the actual risk. Not automation. Slow adoption of automation while everyone else moves.
I write more on this across the blog if you want to keep going. The posts on automation ROI and business process automation cover the practical side of what to automate and when.
Frequently asked questions
Which jobs are most at risk from AI right now?
Data entry clerks, basic copywriters producing templated content, customer service representatives handling scripted queries, junior analysts compiling reports, and paralegals doing document review are all seeing real displacement in 2025. They share the same trait: the work follows consistent patterns that a language model can learn. If your role is mostly that, start building the skills that aren't.
Will AI replace software developers?
Junior developers writing boilerplate — partially, and already. AI coding assistants are genuinely good at well-defined problems. What they cannot do is architect a system from ambiguous requirements, make trade-offs between competing constraints, or take accountability when production fails at 2am. Senior engineers and architects are considerably safer. The gap between junior and senior is widening fast.
What skills make a job resistant to AI?
Four things: physical presence in unpredictable environments, real-time emotional judgment with genuine stakes, moral or legal accountability that cannot be transferred to a machine, and creative direction under genuine ambiguity. If your work requires most of those four, you are structurally safe. If it requires none of them, pay attention to what is happening in your industry.
Is consulting safe from AI?
Research and synthesis within consulting — no, that is being automated rapidly. Judgment, relationships, and accountability — yes, those remain. The consultants who are exposed are the ones whose value proposition is 'we know things you don't.' The ones who are safe are the ones whose proposition is 'we have seen this fail 40 different ways and we will be in the room when you make the call.'
How do I know if my job is safe from AI?
Ask yourself: could this role be described as a clear set of rules someone could follow without judgment? If yes, it can probably be automated eventually. If the honest answer involves 'it depends,' 'you have to read the situation,' or 'someone has to own the outcome' — you are considerably safer. The rule of thumb: specificity of task plus accountability for outcome equals job security.
What should I do if I think my role is at risk?
Stop waiting for certainty and start building the skills that are structurally resistant: physical capability, emotional intelligence, genuine expertise with accountability, creative judgment under ambiguity. Also: learn to use AI tools well. PwC puts the wage premium for workers who use AI effectively at 56%. That number is only moving in one direction.
Tijdo Koster
Automation consultant since 2009. 100–200 projects. Still answers his own emails.
If you have made it to the bottom and your main takeaway is that Tijdo has a Terminator reference in a blog post about AI, fair. My 17-year-old says it is cringe. I informed him that cringe, as of this morning, is now called timeless. He disagreed. We agreed to disagree.
There is more on the blog if you want to keep reading. And if you want to see which AI tools are actually worth adopting, the products page has the opinionated shortlist.