Every few weeks a new study announces how many millions of jobs AI will take. The numbers are big, the headlines are louder, and none of it helps you, because the predictions are about jobs in general — not the specific bundle of tasks you do on a Tuesday.
A better question is whether your work has the properties machines can't reach around. Not one property — machines are good at more every year — but a specific handful that keeps showing up in the jobs that resist automation. We use four tests. None of them require technical knowledge to apply, and together they explain most of what you're watching happen to the job market.
Test 1: The work needs a body, in a place that keeps changing
A plumber walks into a different house every day. The pipes are in different places, the access is different, the previous owner's DIY is different. Machines are extraordinary in controlled environments — a factory line, a warehouse laid out for them — and struggle in places that keep changing on them. If your work happens out in the world, in unpredictable physical settings, that's protection. Not forever, but for a long time.
Test 2: A person is legally or morally answerable for the result
Someone must sign the audit. Someone must teach the class. Someone must issue the verdict. When a result carries consequences for other people, society demands a person who can be held responsible — and a model can't hold responsibility. It can draft the audit, but it can't be the auditor. Work that ends in a human signature resists automation not because machines can't do the task but because we won't accept the alternative.
Test 3: It depends on a relationship with a person
The nurse, the teacher, the therapist. In these jobs, trust is not a feature of the service — it is the service. A patient who knows a machine is delivering their care is receiving different care, even if the information is identical. If the relationship with a specific person is what people are paying for, the job has a moat that has nothing to do with technical capability.
Test 4: The role is one job, not a pile of tasks
Some roles are one coherent thing — a coherent craft that's hard to split. Others are a pile of separable tasks, and piles get dismantled: the routine 60% automated, the rest reassigned. A job that holds together as a single role resists being broken into automatable pieces. A job that's already a bundle of disconnected chores is the most exposed shape of work there is.
Reading the verdicts
Run the four tests and you get a verdict: a role Survives, is Exposed, is Emerging, or is In flux. What surprises people is that mixed is the most common outcome. One title, two jobs: the skilled half is protected, the routine half is not.
Teachers hold on tests two and three — accountability and relationship — even though machines can explain fractions. Data-entry clerks hold on almost none of the four. And AI auditors are an emerging role precisely because test two is being built into law and practice: the accountability layer is growing, not shrinking, and someone has to staff it.
No single test settles a job, and none of this is a promise about the far future. But four specific questions beat one loud prediction every time.
This is the framework we built After AI around — 24 jobs scored, every figure cited to a public source. Start with /jobs/teacher or /jobs/data-entry-clerk.
No comments yet