Is My Job Safe From AI? A Five-Question Test You Can Run on Any Headline

If you have typed “is my job safe from AI” into a search bar at 1am, you already know the internet has an answer for you, and it is designed to make your stomach drop. There is always a fresh list — “50 jobs AI will kill by next year,” “the careers that won’t exist in a decade” — and your role is somewhere on it, ranked, colour-coded, sentenced. These lists get millions of clicks precisely because they trade certainty for fear. But almost none of them can survive a few honest questions. So instead of handing you another ranking to lose sleep over, I want to give you something more useful: a test you can run yourself, on your own job and on any scary headline that crosses your feed.

Why the viral “jobs AI will kill” lists mislead

The first problem with those lists is that they confuse tasks with jobs. A job is a bundle of many tasks, some routine and some not, held together by judgement, context, and responsibility. When a headline says “AI will replace paralegals,” what it usually means is that AI can now draft a first-pass document summary — one task inside a role that also involves chasing clients, reading a room, spotting the thing that isn’t written down, and being the person a partner trusts at 6pm. Automating a task is real. Automating the person who owns the whole bundle is a much taller order, and the lists quietly slide from one to the other.

The second problem is timing. “Will be replaced” is doing enormous work in those headlines with no date attached. A capability existing in a demo is not the same as it being cheap, reliable, regulated, insured, and trusted enough to deploy across an industry. There is often a decade or more between “a model can do this in a lab” and “your employer has actually restructured around it.” A prediction with no timeframe cannot be wrong, which is exactly why it is so easy to publish.

The third problem is accountability — nobody who writes these lists is on the hook for being wrong. There is no cost to declaring your career doomed and moving on to the next slideshow. That asymmetry should make you suspicious. If you want a more grounded version of these questions, I have written separately about will AI take my job and about which office jobs AI replaces first, because the honest answer is always more specific than a headline can be.

A prediction with no timeframe cannot be wrong — which is exactly why it is so easy to publish.

The five-question test

Here is the diagnostic. Run it on your own role first, then keep it in your pocket for the next headline. None of these questions gives a yes-or-no verdict; they give you a direction and a set of things to strengthen. Score each one loosely — “a lot,” “somewhat,” “barely” — and pay attention to where you land.

1. How routine and rule-based are my core tasks? The more your day is made of repeatable, well-defined steps with a clear right answer, the more exposed those steps are. If most of what you do could be written as a checklist someone else could follow with no context, that is the part a machine reaches first. If your work constantly involves ambiguous inputs, exceptions, and “it depends,” it is far harder to encode.
2. How fully digital is the work — does it need a body, a place, or a person in the room? Software automates software cheaply and atoms expensively. Work that lives entirely inside a screen — text in, text out — is squarely in reach. Work that needs hands on a physical thing, presence in a specific location, or a human being physically trusted in the room is protected not because it is more sophisticated but because it is more expensive and awkward to replace. A plumber and a therapist are safe for very different reasons than a data-entry clerk is exposed.
3. Does a mistake carry real accountability someone must own? Ask what happens when the work is wrong. If an error costs money, breaks the law, ends a life, or lands someone in court, then somebody has to be accountable for the decision — and a model cannot be sued, licensed, struck off, or sent to prison. In high-stakes work, AI tends to become the assistant that drafts and flags, while a named human stays on the hook for signing. The higher the consequence of being wrong, the stickier the human role.
4. How much of my value is judgement and relationships rather than output? Some of what you are paid for is the artefact — the report, the code, the design. But a lot of it is the trust that got you the brief, the read on what the client actually wants, the ability to say “this is the wrong thing to build” before anyone wastes a quarter on it. Judgement about which problem to solve, and relationships that make people bring you the problem in the first place, are the least automatable things in almost any role. If your value is mostly volume of output, you are more exposed than if your value is knowing what should be produced at all.
5. Who captures the savings if my tasks are automated — and does that give someone an incentive to cut me? This is the question the lists never ask, and it is the one that actually decides your fate. A task becoming automatable does not automatically remove you; someone has to choose to remove you, and they do it when your removal puts money in their pocket. If automating your tasks means your employer keeps the savings, there is a direct incentive to reduce headcount. If the savings would flow to customers, or get eaten by coordination costs, or if you are the person who operates the very tool that does the automating, the incentive points the other way.

Notice how the last question changes the shape of the whole test. Questions one through four are about capability — can the machine do the task. Question five is about power — who decides, and who benefits from the decision. You can be brilliant at an un-automatable job and still be cut because someone above you found a way to book your salary as a saving. And you can do fairly routine work and stay employed for years because nobody with the authority to replace you has a reason to.

Running the test on a headline

The same five questions work on any “AI will kill X” article, and they turn a scary claim into a checkable one. When you next see a list, interrogate it:

• Task or job? Is the headline describing one automatable task, or has it quietly promoted that to the whole role?
• When? Is there an actual timeframe, or just the timeless “will be replaced”? Is the capability shipping in real workplaces, or living in a demo?
• Accountability? Does the work in question carry consequences someone must legally own — and does the article account for that friction?
• Judgement and relationships? Has the writer reduced a relationship-heavy, judgement-heavy job to its most mechanical slice to make the claim land?
• Who benefits? Who is telling you this job is doomed, and what are they selling — a course, a tool, a consulting engagement, a stock story? Follow the incentive behind the prediction.

Most doom lists fail three of these five on contact. They describe a task as a job, attach no date, and ignore both accountability and the incentives of whoever is publishing. That does not mean nothing changes — it means the change is slower, narrower, and more negotiable than the headline wants you to feel.

The pattern underneath the panic

Once you start asking who benefits, you notice something that goes back long before AI. The story we are told about a new technology is almost always that it is an unstoppable force of nature — the tide coming in, the weather changing — and that our only choice is to adapt or drown. But technology is not weather. It is built, funded, priced, and deployed by people who make choices, and those choices reliably favour whoever already holds the capital. This is the capture pattern, and I keep returning to it because it is the same move, a new machine, every time. The power loom, the shipping container, the spreadsheet, the algorithm — in each case a genuine capability arrives, and then a narrative of inevitability arrives right behind it to make sure nobody asks the more dangerous question: who is capturing the gains, who is paying the cost, and who gets a say.

The sharper question is not “will the machine do my task?” but “who benefits when it does?”

That reframe is why I am not interested in selling you either doom or dismissal. The people who say “AI changes nothing, relax” are wrong, and the people who say “everything is over, your job is gone” are also wrong — and, not coincidentally, both are usually selling something. The truth is that a task being automatable is the start of a negotiation, not the end of one. Whether it costs you your livelihood depends on ownership, incentives, regulation, unions, and public choices — on all the messy human machinery that decides who keeps the savings. This is also why the technology genuinely can create jobs even as it destroys tasks: new work appears wherever the productivity has to be turned back into something a person still has to do, decide, or be trusted with.

So the most important thing you can do with the question “is my job safe from AI” is to refuse the passive version of it. Do not ask it the way you would ask about the weather. Run the five questions, and then hold on to the fifth one hardest, because capability gets all the headlines while power quietly makes the decisions. If you want to understand why the inevitability story keeps working on us, I have written more about how technology gets captured — the mechanics of how a tool that could have gone many ways gets steered toward the few who already own the most.

Your job is not a line item on someone’s slideshow. It is a bundle of tasks, judgement, and trust, sitting inside an organisation full of choices that have not been made yet. The machine can do more of your tasks every year — that part is real, and pretending otherwise helps no one. But whether that fact ends with you cut loose or with you doing more of the work only a person can do is not settled by the technology. It is settled by who benefits when the machine does the task. Ask that question loudly, keep asking it, and you turn a headline that was designed to scare you into a decision that someone — maybe you — still gets to influence.