⏲️ The 20-Minute Job Safety Check

Worry about AI and your job has a strange quality: it expands to fill whatever time you give it. Give it a sleepless night and it’ll take the whole night. Give it a scroll through the news and it’ll ruin the scroll. Left unstructured, it never finishes, because it never has anywhere to land.

So we’re going to give it somewhere to land: twenty minutes, a timer, and four questions. At the end you will know, in writing, where your job actually stands and what your next move is. Not a feeling… a page.

I think of it like an MOT. You don’t MOT your car by lying awake wondering if the brakes are fine. You put it on the ramp, run the checks, and get a sheet back: here’s what passed, here’s the advisory, here’s what needs booking in. Twenty minutes of structure beats six months of vague dread, every time.

(If you want the thinking behind why jobs don’t get replaced the way headlines suggest - tasks go first, whole roles rarely - that’s Is AI Coming For My Job? An Honest Gut-Check. This guide is the practical: the exercise itself. They pair nicely.)

Put the kettle on, set a timer, and let’s put your job on the ramp.

The Four-Question MOT

Four questions, five minutes each. Do them in order - each one feeds the next. You can do this on paper, but it works best with Claude open (ChatGPT works too), because question two onwards uses AI to check your own blind spots.

Question 1: What do I actually do all week? (5 minutes) This is the inventory. Write down everything you did last week as separate tasks - not your job description, your actual week. How: brain-dump fast, aim for 15 to 30 items, and deliberately include the small stuff: the chasing, the formatting, the “quick” calls, the meeting that eats Tuesday. The non-obvious bit: your job description is the worst possible source for this. Job descriptions describe the role someone imagined years ago; your calendar and sent folder describe the job you actually have. If you’re stuck, open last week’s sent emails and calendar and read them like a detective.

Question 2: Which of these tasks can AI do? (5 minutes) This is the exposure check, and you don’t do it yourself - you ask the AI directly, because it knows what it’s capable of better than a news headline does. How: paste your task list in and ask it, honestly, which tasks it could do today, which it could half-do, and which it can’t touch (the prompt below does this for you). The non-obvious bit: instruct it to be blunt. AI assistants default to reassuring, and a reassuring safety check is a useless safety check. You want the mechanic who shows you the worn pad, not the one who waves you off the forecourt.

Question 3: What’s my moat? (5 minutes) Your moat is the part of your role that would still need you even if your workplace went all-in on AI tomorrow. Relationships you hold. Trust you’ve banked. Judgement calls with your name on them. Context nobody wrote down - why the big client is touchy in March, which supplier says yes if you ring rather than email. How: list five things under “this needs me, specifically”, then test each against one hard question: does it need a human, or does it currently just involve one? The non-obvious bit: “I’ve always done it” is not a moat. Habit isn’t a moat. A moat is something someone would miss within a fortnight if you stopped.

Question 4: What’s my next move? (5 minutes) The whole point of an MOT is the sheet at the end: fine, advisory, fix now. How: from everything above, pick exactly one action for the next fortnight - usually either “learn to run AI on my most exposed task before someone else does” or “spend more deliberate time on my moat”. Write it down with a day attached. The non-obvious bit: one move, not five. A five-item plan after an exercise like this is a decorative plan. One action with a date gets done, and doing it earns you the next one.

The whole exercise, as one prompt

Here’s the entire MOT as a single guided sequence. Paste it in, and the AI runs the exercise with you - timing, bluntness and all. You just answer its questions.

Nick this one, fill in the blanks:

I want to do a 20-minute job safety check with you. Act as a blunt but kind career mechanic: your job is to give me an honest MOT on my role, not to reassure me. Run these four stages IN ORDER, one at a time, and wait for my answer before moving on.

About me: I'm a [JOB TITLE] at [TYPE OF ORGANISATION - e.g. "a mid-sized housing association", no company name needed]. I've been in this kind of role for [YEARS].

STAGE 1 - INVENTORY (5 min): Ask me to list everything I did last week as separate tasks, big and small. If my list looks thin or too grand, prod me for the boring bits: chasing, formatting, routine replies, recurring meetings.

STAGE 2 - EXPOSURE (5 min): Sort my tasks into three groups: (a) AI could do this today with light checking, (b) AI could do a first draft and I'd judge it, (c) AI can't touch this. Be blunt. Do not flatter tasks into group (c). Estimate roughly how many hours of my week sit in group (a).

STAGE 3 - MOAT (5 min): Ask me what parts of my role would still need me specifically if my employer went all-in on AI. Challenge every answer with: "does this NEED a human, or does it currently just involve one?" Help me end up with 2-3 genuine moat items, and tell me plainly if something I've claimed isn't really a moat.

STAGE 4 - VERDICT (5 min): Give me an MOT-style result sheet:
- PASSED: what's solid about my position
- ADVISORY: what's exposed but not urgent
- FIX NOW: the single most important thing to act on
Then help me pick ONE action for the next 14 days - specific, small, with a suggested day to do it - and one sentence on why that action beats worrying.

Keep the whole thing moving - this should feel like 20 focused minutes, not a therapy session.

What good input looks like

The exercise lives or dies on Stage 1, so here’s what a strong inventory looks like. Say you’re a customer service team leader at an insurance firm. Weak: “manage the team, handle escalations, reporting.” Strong: “did the Monday rota, ran two one-to-ones, handled four escalated complaints (two on the phone), wrote the weekly stats report, updated the complaints log, sat in the ops meeting, coached Priya through a hard call, chased IT about the headset order, covered the queue over lunch twice.”

From the weak list, the AI can only shrug. From the strong one, it can tell you something real: the stats report and complaints log are exposed, the escalation calls and the coaching are moat, and your one move is probably learning to run AI on that report before the quarterly review.

What the MOT can’t tell you

Twenty minutes gives you a snapshot, not a prophecy - roles and tools both shift, which is why this check is worth redoing every six months or so, same as the car. It can’t see your workplace’s politics, your boss’s plans, or your industry’s economics, so treat the verdict as “where do I stand” rather than “what will happen”. And it will be confidently wrong occasionally about what AI can and can’t do, in both directions… which is rather fitting, and a good reminder that everything it says gets your judgement on top.

Mostly, though: it won’t work as something you read. It only works as something you do, with the timer actually running.

If today only has five minutes in it

Do Stage 1 tonight. Just the inventory - fifteen honest tasks from last week, typed into a note. That list is the raw material for everything else, and most of the calm arrives the moment the vague cloud of “my job” becomes a list you can point at.

And when you’ve got your verdict sheet, the natural next read is the reframe that sits underneath it all: Is AI Coming For My Job? An Honest Gut-Check. Book yourself in, friend. The ramp’s free. ✨