Will AI Take My Job? Audit Your Week, Not Your Job Title
We are not going to give you a percentage. Every tool that returns one for your occupation has invented it, because two people with the same job title spend their weeks doing entirely different things, and the exposure lives in the things. Nobody has measured your week.
What can be assessed is which of your own tasks a machine can already do. That is a real question, you can answer it in four minutes, and unlike a forecast it tells you something you can act on this month.
Name the things you actually did last week, one per row. Be specific: not marketing, but wrote the monthly customer email. Then tick whichever of the five apply to that task. Nothing is stored or sent anywhere.
The five tests, and why each one matters
These are not a proprietary index and there is nothing hidden in them. Exposure rises with a digital output and a checkable answer, and falls with accountability, physical presence and relationship. Disagree with any of it and adjust your own reading, which is easier to do when the reasoning is on the page.
The output is digital
This is the threshold test and it decides more than the other four combined. Work whose product is a document or a file is work a model can already produce a first draft of. That is not the same as doing your job, but it is the part of your job that changed.
There is a broadly correct answer
Tasks with a checkable answer automate first, because success can be measured without you. Tasks that require choosing between defensible options under uncertainty resist much longer, since the difficulty is deciding what good would even mean.
Accountability attaches to it
This is the strongest protection of the five, and the most underrated. Where a named person must answer for the outcome to a regulator, a court, a board or a client, the work does not disappear, it changes into review and sign-off. The task survives in a different form and the person who can defend the call is the one still doing it.
It is physical
Robotics moves far slower than software, and unpredictable physical environments remain genuinely hard. This is real protection today and the least likely of the five to be the reason your particular week is safe, because most office work fails it.
It depends on a relationship
Not the polite version of this, which automates immediately, but the real version: the client who tells you the true reason for the deadline, the colleague who warns you before the meeting. That runs on trust with a specific person and does not transfer to a tool.
What actually changes your position
In rough order of how much difference each one makes.
1. Move toward the accountability, not away from the tools
In every field where a named person must answer for the outcome, the work becomes review and sign-off rather than production. The person who can defend a decision to a regulator, a court or a client is the one still in the room. Refusing to use the tools does not get you there, and being able to say exactly why an output is wrong does.
2. Become the person who verifies, in your own domain
Anyone can generate a draft. Very few people can say with authority that a particular output is wrong, and why, in a field where being wrong has consequences. That capability is scarce, it is rising in value, and it is built from domain knowledge rather than from tool knowledge.
3. Write down what you can actually prove
The labour market is moving from credentials that record attendance to evidence that records capability. A list of courses attended answers a question nobody is asking. A record of problems solved, decisions defended and work you can be examined on is the thing that survives the next reorganisation.
4. Learn the failure modes, not the prompts
Prompt technique ages in months. Knowing where these systems reliably fail, why they state false things fluently, and which of your tasks they should never touch is knowledge that keeps its value and is exactly what makes your review worth trusting.
One thing worth knowing before you buy any training
In the United States, no federal law and no state law requires your employer to train you on AI, and nothing requires you to hold a certificate. We wrote the jurisdiction by jurisdiction answer with citations, including the part where the honest answer is no. Anyone telling you a rule obliges you to buy their course is describing a rule that does not exist.
Questions people ask
Will AI take my job?
Nobody can answer that about a job title, and any tool that gives you a percentage for your occupation has invented it. Exposure is a property of tasks, not titles, and two people with the same title spend their weeks doing different things. What can be assessed is which parts of your own week a machine can already do, which is why this page asks you to list what you actually did rather than what you are called.
Which jobs are safest from AI?
Rather than a list of titles, the properties that protect work are consistent: someone must be accountable for the outcome in a way that carries a signature, a licence or liability; the work needs a body in an unpredictable place; or it runs on trust built over time with specific people. Accountability is the strongest of the three and the most underrated, because where it exists the task usually becomes review and sign-off rather than disappearing.
What should I do about AI taking jobs?
Move toward the accountable end of your own field rather than away from the tools. The durable position is being the person who can say an output is wrong and defend why, in a domain where being wrong has consequences. That is built from domain knowledge and from knowing where these systems fail, not from prompt technique, which ages in months.
Is it true that AI only automates tasks, not jobs?
It is closer to the truth than the alternative, and it is not comforting on its own. A job is a bundle of tasks, and removing most of the bundle changes the job even when the title survives. The useful version of the idea is that the bundle can be rebuilt deliberately: if you can move your week toward the tasks that resist, you have changed your position without changing your employer.
Do I need an AI certificate to stay employable?
No law requires one, and in the United States no state or federal rule requires employers to train employees on AI at all. What employers increasingly ask for is evidence of capability rather than attendance, which is a different thing from a certificate. A record of what you can actually do, that you could be examined on, answers the question a certificate only gestures at.
Nothing you type here is stored or sent anywhere. This is a way of thinking about your own work, not a forecast, and it deliberately produces no probability, because a probability would imply a measurement nobody has made.