work & your next five years · 3 min
Will AI come for your job — or just the first rung?
Not 'will AI change work' in the abstract — will it change yours, and when?
The economy-wide numbers are calm. The entry-level numbers are not. Where you stand decides which story is yours.
Picture an observer hovering above Earth, reading everything humans publish on this. Two camps shout past each other: one says the robots are coming for everyone, the other says relax, they always said that. Both are partly right — and mostly arguing about different things.
Here is the shape of it in plain English: who says what, who is probably closer to the truth, and how sure anyone can honestly be.
What you're usually told
The worker's warning
Unions and many economists on the left
AI is quietly automating the bottom rungs and handing the gains to whoever owns the software. The IMF's own 2024 work found AI is likely to widen inequality — and entry-level openings have already thinned, with more 'junior' roles quietly asking for years of experience.
The transition story
The IMF and most policy institutes
Around 40% of jobs worldwide — and about 60% in rich countries — are 'exposed' to AI. But exposed is not doomed: the IMF reckons roughly half of those could be made more productive rather than replaced. The honest headline is upheaval and reskilling, not a switch flipped to off.
We've seen this film
Market economists and tech optimists
Every wave of automation since the loom was going to end work, and every time the new jobs outnumbered the old. Expect a productivity boom and kinds of work we cannot yet name. So far the whole-economy numbers sit closer to this calm view than to the panic.
The contrarian read
The 'it's overhyped' read
Daron Acemoglu (MIT) and fellow sceptics
A leading economist ran the numbers and found the whole-economy effect is real but small — under about 0.7% added productivity over ten years, perhaps less. On this view the genuine harm is not mass unemployment; it is concentrated and unfair, landing hardest on juniors and new graduates while the aggregate barely moves.
Our read
Most likely, this decade: no economy-wide jobs apocalypse — but a real, uneven squeeze that lands first and hardest on people just starting out.
Our confidence: moderate (≈60%). A judgement from current evidence, not a forecast.
Put the camps together and the contradiction mostly dissolves. The macro economists are probably right that AI will not vaporise work across the whole economy this decade — measured productivity gains are still modest. The worried camp is probably right that the damage is concentrated: in 2026, recent-graduate unemployment is running above the national rate, and entry-level postings have thinned sharply.
So the truthful version is neither 'robots take all the jobs' nor 'nothing to see here'. It is redistribution — toward whoever owns the models, and away from new entrants. Less dramatic than the headlines, and if you are twenty-two and job-hunting, considerably worse.
- ≈60%Uneven squeeze — modest economy-wide gains, a hard entry-level market, most jobs changed rather than vanished.
- ≈20%Faster bite — a capability jump automates whole roles before new ones appear, and displacement outruns retraining.
- ≈20%Broad lift — AI raises productivity enough that new work absorbs the displaced and wages rise widely.
How sure are we?
Beware attribution: recent-grad hiring is also weak for ordinary cyclical reasons (a post-2021 over-hiring hangover), so not all of the squeeze is AI. 'Exposed to AI' means tasks can be touched, not that the job disappears. Every percentage is a considered judgement.
Where this comes from
- IMF (2024) — Gen-AI: Artificial Intelligence and the Future of Work Where the 40% / 60% job-exposure and the inequality finding come from.
- Daron Acemoglu (2024) — The Simple Macroeconomics of AI The 'modest macro effect' estimate (<0.7% added productivity over 10 years).
- Stanford Review (2026) — The Class of 2026 is struggling to find jobs — and it's not because of AI The honest counter — how much of the entry-level squeeze is cyclical, not AI.
Go deeper — read the book
A thousand-year history of who actually captures the gains from new technology — and how that gets decided.
The heavyweight on this question — a Nobel economist's thousand-year evidence base. If you read one book on it, this is the one.
Written with AI and editorially curated. Talescout always labels AI text (EU AI Act, Art. 50). Numbers are considered judgements, not measurements.