AI with Kyle Daily Update 192: They Took 'Er Jobs!

AI takes tasks before it takes job. Which sounds softer.

It ain’t … it’s just sneakier.

The South Park version is easier to picture. "They took our jobs!!" Deborah from accounting walks in on Tuesday and the chatbot is sat in her chair, smug little spreadsheet open, Xero login ready to go.

Nah. That ain’t it.

The actual mechanism is much quieter, more insidious. The task bundle shrinks. The company backfills slower. The junior role never opens. You don't see the missing job because it never gets a payslip.

It takes jobs BEFORE they exist.

That's the bit I wanted to dig into on the livestream: how AI takes specific tasks first and foremost. Also I’ve built a rough AI Job Risk Diagnostic to help you run this against your own role.

Very early version. Have a poke around!

Job Titles Are Too Crude

Just looking at a job title is a rubbish way to decide if it’s at risk from AI.

Accountant. Zookeeper. Marketing manager. Solicitor. Teacher. Whatever.

Those labels hide the actual work. Every job is a bundle of tasks. Admin, meetings, reporting, client work, judgement, coordination. Yes there is the core craft but it’s often the little annoying bits around the edge that somehow eat half the week…

The ILO's 2025 GenAI exposure work is useful here because it looks at the tasks inside occupations rather than the title on your LinkedIn.

Importantly it also looks pretty!

It also makes the boring-but-important point that few jobs are fully automatable with current GenAI, while nearly all occupations still contain human-input tasks.

But that does not make the risk small. It makes the risk uneven. A job doesn’t have to be fully done by AI to be at risk.

Take an SME accountant. The exposed edges are obvious: reconciliation, invoice processing, first-draft reports, spreadsheet cleanup, basic tax research. Claude in Excel is already getting useful here. Xero has been nibbling at this stuff for years anyway.

But the defensible bits are different: trust, judgement, trade-offs, business context, explaining what a client should actually do with their money.

That is what I pay my accountant for. Not the spreadsheet faff. The judgement, the experience, the “taking-it-off-my-plate”.

AI can take the sucky bits

For a lot of jobs you might want AI to take the report writing. Most people do not leap out of bed desperate to write compliance notes.

The drudgery. The terrible parts of work. AI can (and will) take these off our plate.

That’s a silver lining. AI can remove the bits people hate. Nice!

BUT…

If half the task bundle gets compressed, the organisation needs fewer hours to produce the same output. Fewer hours will require fewer people.

Maybe no one gets fired on day one. Maybe the team just stops hiring. Maybe the next vacancy sits open. Maybe five planned hires become one or two hires plus AI.

That is still labour displacement. It's just harder to see than someone being FIRED due to AI.

The Missing Job Won’t Show Up

This is why the "AI took my job" story can be a bit crap.

Layoffs are visible. They get press coverage. Like Meta’s recent (messy) round of layoffs. It puts companies under scrutiny. It raises questions.

Missing jobs do not. There’s no smoking gun. =

But absence of evidence is not evidence of absence. The damage will still be done. Just quietly and without any alarm.

If AI means a company avoids creating a junior role, there is no sad LinkedIn post. No severance. No big HR announcement. Just a job that would have existed and now doesn't.

This is also why Gen Z in particular are in trouble. As it’ll hit those entry level jobs first. Hell, it is already. This was the front page headline on the BBC yesterday.

Entry-level work has a lot of repeatable task work inside it. Legal discovery. First drafts. Research. Data cleanup. Support triage. QA. Junior analysis. The stuff everyone moans about because it is boring.

But boring work was also training. It’s what we cut our teeth on, for better or worse.

You did the rubbish bits, slowly became less useless, built judgement, earned trust, and moved up. If AI hoovers out the boring layer, the first rung of the ladder gets very wobbly. Or disappears entirely.

Stanford's Digital Economy Lab found that early-career workers aged 22-25 in the most AI-exposed occupations saw a 16% relative employment decline after GenAI adoption, with firm-level shocks accounted for. The Dallas Fed has a nice extra detail: for young workers in high-exposure occupations, the problem looks more like lower inflows into work than a big wave of layoffs.

Subtle, creeping. But dangerous.

Productivity Still Needs Customers

“Oh but Kyle we’ll just hire more people and create even more with the help of AI! Businesses will boom!”

Thank you Strawman. This is the obvious (and at first glance solid) pushback. I hear it a lot.

Fine. The company keeps headcount, gives everyone AI, and output goes nuts. Ten times the reports. Twenty times the designs. Fifty times the legal memos. Lovely!

This only works if demand follows.

More capacity does not automatically create more customers. If every accountant can process ten times the work, but the market does not magically need ten times the accountancy, prices get squeezed. Or hours get squeezed. Or headcount gets squeezed.

Probably all three, in some places.

Demand elasticity is doing a lot of work there. If the market is saturated, or customers have less cash, or everyone is producing more of the same thing, the productivity story gets ugly fast. Making more shit won’t help. More output into weak demand is not abundance.

Safe Jobs Are Not Isolated

“OK but my job is safe”.

Sorry. It’s not. This matters even if your job looks safe on paper.

Massage therapist. Physical therapist. Plumber. Dentist. Electrician. Personal trainer. Loads of physical and local work has lower direct AI exposure because robotics is still behind. It has not had its ChatGPT moment. Too expensive. Too clumsy. Too much real world mess.

For now. (Very likely more on that later!)

Physicality buys time - absolutely. You have more time to deal with AI.

What it does not buy though is immunity from the wider economy. You are still part of a market.

Let’s say you are a physical therapist. Direct impact from AI? Limited! Horray.

But your customers are bankers, accountants, marketers, lawyers, founders, or office workers… (all sitting in their lovely office chairs all day and gaining terrible postures for you to fix).. and their income gets squeezed, your demand can get squeezed too. Your task exposure might be low while your customer exposure is high.

This is second-order consequences. The stuff that is hard to see and harder to predict. You can be safe from direct replacement and still exposed to the economic blast radius. Basically if the economy goes down the shitter you won’t be immune.

Sorry!

Run The Napkin Audit

OK let’s get practical. You know I don’t deal in doom and gloom

You do not need a full tool to start. But I have built you one - cause I’m cool like that. Here’s the link.

Back of an envelope is enough.

Here we go:

Write down the top 8-10 tasks you do in a normal week. Add a rough percentage of time next to each one. Do not make this psychotic. If reconciling invoices is roughly 20%, write 20%. If client calls are 15%, write 15%. Close enough is fine.

Then mark what AI can do with each task:

Do. Draft. Assist. Check. Cannot.

That is the first pass.

Then score the defensibility. Physicality. Empathy. Non-algorithmic judgement. Accountability or legal liability. Union or collective protection if that applies.

Reconciling invoices? High exposure. Low defence. Non-physical, algorithmic, basically zero empathy. Gulp.

Client meeting? Mixed. AI can prep, summarise, draft follow-ups, maybe replace some meetings with an email if everyone is honest. But hearing what the client is not saying, earning trust, managing anxiety, reading the room… still human edge there.

Once you have the task mix, add the exposed percentage and ask the awkward question: Could fewer people now cover this job?

Or…use the tool I made here.

Move Toward The Defensible Bundle

And what do you DO with this knowledge?

Once you know what’s going to be taken by AI….let go.

Stop trying to do that stuff. That’s a dumb move. Do not spend your best energy getting better at tasks AI is going to run over.

Learn to automate those. Lean in and use AI.

If you know an industry well, that exposed edge might be the business opportunity. Build the invoice reconciliation workflow. Build the discovery assistant. Build the compliance-note helper. Build the reporting system your old team secretly wanted but never had time to make.

Nice!

But your personal skill development should move toward the defensible bundle.

Physical work where relevant. Human trust. Public speaking. Communication. Sales. Taste. Judgement under uncertainty. Client handling. Accountability. The ability to sit with someone, understand the messy context, and make a decision when there is no tidy SOP.

That is the human edge for now. Lean into this sort of messy human stuff.

To the Task,

Kyle