A good plumber in most American cities can charge around $150 an hour and still be booked out two weeks in advance. No chatbot is coming for that job, and it is worth understanding exactly why — because the reasons are a map you can apply to your own work. “Safe from AI” is not really a list of lucky job titles. It is a set of traits, and you can measure any job, including yours, against them. Here is how to do that measurement, step by step.
Step 1: Break your job into tasks, not a title
Nobody’s whole job gets automated at once. What actually happens is that individual tasks inside a job get absorbed by software, one at a time, and the job quietly changes shape around what is left. So the honest unit of analysis is the task. Take a notebook and list what you genuinely do in a week: not your job description, but the real hours. Writing summaries. Answering routine questions. Crawling under a house to find a leak. Calming down an angry customer. Deciding which of three bad options is least bad.
Expect a mix. Almost every job — including the “safe” ones below — contains some tasks a machine already does well, and most contain a few that would surprise you in the other direction.
The question is never whether your job contains automatable tasks. It is what remains after those tasks are gone, and whether what remains is a job a person would still be paid to do.
Step 2: Score each task against the four automation tests
Run every task on your list through four questions. Is it digital — done entirely on a screen, with information in and information out? Is it predictable — similar each time, with a right answer that follows a pattern? Is it low-stakes if wrong — an error is annoying rather than dangerous or expensive? And is it unsupervised-friendly — could it run without anyone being responsible for the outcome?
Tasks that score yes four times are the exposed ones: data entry, routine report writing, first-draft translations, basic bookkeeping, standard customer queries. That is the territory where change is already visible — the debate in will AI replace customer service jobs is really about an occupation whose task list scores high on all four. Tasks that fail even one test — because they are physical, unpredictable, dangerous to get wrong, or need a responsible human attached — are the durable ones. Your job’s safety is simply the balance between the two piles.
Step 3: Identify the work that fails the tests by design
Look at the traits that keep failing the automation tests, because they cluster into recognizable kinds of work:
- Physical work in unpredictable places. Every house, vehicle, body, and building site is different, and the real world does not arrive in tidy data. Plumbers, electricians, mechanics, and HVAC technicians work with their hands in spaces no two of which are alike.
- Work where a human must be accountable. Surgery, childcare, courtrooms, and cockpits keep a licensed, responsible person in the loop partly by law and partly because society insists someone answerable is there. Regulation is a genuine moat.
- Work built on trust and presence. Therapists, nurses, teachers of young children, and good salespeople are partly selling the fact that a person is there. Automating the tasks around them does not remove the reason they are hired.
- Work at the messy edges. Emergencies, repairs, negotiations, and crises are defined by being the cases the pattern does not cover — which is precisely where pattern-based tools are weakest.
Job-exploration tools reflect the same logic: browse O*NET’s list of occupations with a bright outlook and you will see it is heavy with hands-on healthcare, skilled trades, and in-person roles — the categories whose tasks fail the four tests most often.
Three jobs, scored honestly
The method is easier to trust with examples. Take a paralegal: document review, deadline tracking, and standard form preparation score high on every automation test — digital, predictable, pattern-based. But interviewing a distressed client, spotting the one odd clause in a contract, and knowing which local judge wants what are judgment and trust work. Same title, two very different futures depending on which pile the person builds a reputation in.
Take a retail manager. Scheduling, inventory reports, and stock ordering are already substantially software work. Handling the Saturday-morning crisis — a delivery missing, two staff out, a queue forming — plus coaching a struggling teenager through their first job, is physical, present, and human. The manager who is “good with the system” is replaceable by the system. The one who holds a store together on its worst day is not.
And take the electrician, close to our plumber. Even here there are digital, predictable tasks: quotes, invoices, scheduling, parts ordering — increasingly handled by apps, and electricians are glad to hand them over. What remains is fault-finding in a fifty-year-old house where the wiring matches no diagram, working live-adjacent work safely, and being trusted alone in someone’s home. That pile barely shrinks at all. Notice the pattern in all three: the exposed tasks were mostly the ones nobody loved doing anyway. Automation, viewed through the task list, looks less like a thief and more like a very aggressive editor of your week.
Step 4: Move yourself toward the judgment end of your own job
You do not need a new career to use this map; you need a direction inside your current one. In nearly every occupation there is a spectrum from routine execution toward judgment, relationships, and responsibility. The bookkeeper who only enters invoices is exposed; the one who spots the cash-flow problem and tells the owner what to do about it is not. The support agent who answers FAQs is exposed; the one trusted with furious customers and retention decisions is not.
Concretely: volunteer for the tasks involving people, ambiguity, and physical presence. Learn the parts of the job that require a license, a signature, or a site visit. Become the person who checks the machine’s output rather than the person whose output the machine checks. These moves are unglamorous and they compound.
Step 5: Learn to run the tools instead of racing them
The final step sounds paradoxical and is not. In most workplaces, the first person whose productivity jumps is the one who adopts the AI tools for the automatable pile — drafting, summarizing, first passes — and spends the saved hours on the human pile. Employers notice who those people are. The realistic risk for most workers is not “a robot took my job”; it is “a colleague who uses these tools well now does my job plus half of another one.”
A five-minute exercise: read a job ad through the task lens
Here is a way to make all of this concrete before your next career decision. Take a current job advertisement for the role you want — yours in five years, or the one you hold now — and read the responsibilities list line by line, scoring each against the four tests from step two. “Prepare monthly reports”: digital, predictable, pattern-based — exposed, whoever holds the job. “Present findings to leadership and defend recommendations”: judgment, presence, accountability — durable. “Maintain records and update systems”: exposed. “Handle escalations”: durable. Most real postings split roughly in half, which is the single most useful fact in this article — the future of that job is not the whole posting surviving or vanishing, but the durable half becoming the job and the exposed half becoming software the jobholder supervises. When you choose training, projects, and promotions, you are really choosing which half of that list your name is attached to. Choose deliberately, a line at a time, and the automation question stops being weather you worry about and becomes a direction you are already walking.
An honest caveat before you relax
“Safest” means slowest to change, not immune. Trades face labor shortages and aging workforces more than they face robots, but no occupation’s task mix is frozen, and today’s judgment work can become tomorrow’s pattern. Geography matters too: the same job title can mean a routine, centralized, highly automatable role in one setting and a varied, on-site, judgment-heavy role in another — the hospital technician and the small-clinic technician share a title and very little else. Score the job you actually have, in the place you actually have it, not the occupation in the abstract. The durable strategy is not picking a fortress job at twenty and hiding in it for forty years. It is the habit built in the steps above: knowing which of your tasks are exposed, steadily trading them for work that fails the automation tests, and treating each new tool as something to operate rather than something to fear. Do that, and you do not need the future to be kind. You just need it to keep needing people for the parts that were never really about information at all.
2 Comments