AI mostly moves your job rather than deleting it, shifting the work up a level of abstraction, from performing a task to directing and judging it. The domain expertise stays valuable; what changes is that you spend less time doing and more time deciding. But the honest version has a sting the optimists skip: the work that erodes first is the entry-level, execution-heavy rung, which is how people used to acquire the judgement the higher rung requires. The role moves up. The ladder loses its bottom.
Ross Jones, Founder, The Hopium Lab. Last modified 22 July 2026.
An evidence-led read that is neither doom nor cheerleading, because both are lazy.
Is AI actually deleting roles?
Not at the rate the headlines claim, and the data on where it bites is more specific than "jobs". Analysis of hundreds of thousands of real AI-assisted work sessions found humans still making the large majority of the planning decisions while the model handled execution, and management-shaped work scoring highest for effective AI use rather than lowest. The pattern is augmentation of judgement, not replacement of the person holding it.
The labour data points the same way, with a sharp edge. Payroll analysis found the measurable employment decline concentrated in 22-to-25-year-olds in AI-exposed occupations, while experienced workers in the same occupations stayed flat or grew. That is not "AI takes jobs" evenly. That is AI eroding the entry rung specifically, which is a different and more awkward finding.
AI is not deleting roles across the board. It is dissolving the bottom rung of the ladder while leaving the top rungs standing, which looks like good news until you ask how anyone reaches the top.
What does "moving up a level" actually mean?
It means your value migrates from executing the task to deciding what the task should be and judging whether it was done right. A lawyer spends less time drafting and more time deciding what position to take and whether the draft is sound. A developer spends less time typing implementations and more time on architecture and review. A marketer spends less time producing copy and more time on judgement about what should be said and to whom. The domain does not change. The altitude does.
This is why domain expertise holds its value while raw execution loses it. The judgement, what matters here, what good looks like, where the non-obvious risk is, takes years to build and does not transfer to a model, because it was never written down for the model to learn. The execution is exactly what the model is good at. So the work redistributes: the model does more of the doing, the human does more of the deciding, and the person who only ever did the doing is the one exposed.
The work moving up a level is good news for the person whose value was always their judgement, and bad news for the person whose value was their throughput. AI is a throughput machine.
What is the wage and demand signal?
The market is paying up for the higher-altitude version of the skill, which is the clearest sign the work is moving rather than vanishing. Analysis across more than a billion job postings found a substantial AI wage premium, rising sharply where more than one AI skill is required, and, tellingly, most AI-skill postings now sit outside the IT function entirely. The premium is largest for roles like architects and specialists, not for pure coders.
That is the signal of a skill spreading up and out, not concentrating or disappearing. AI capability is becoming a modifier on existing senior roles across every function, rewarded because it multiplies judgement rather than replacing it. Same domain, higher-altitude version of the job, priced higher. Same same, but different.
| The old shape of the role | The AI-shifted shape | |
|---|---|---|
| Where time goes | Doing the task | Directing and judging the task |
| What is scarce | Throughput and skill | Judgement and taste |
| Who is exposed | Nobody in particular | The pure-execution / entry rung |
| Wage signal | Flat | Premium for the augmented senior version |
| The catch | , | The rung that built juniors is the rung AI does |
What is the part nobody will say?
The part nobody in either camp will say is that eroding the entry rung breaks the mechanism by which people become seniors. The optimists say "AI moves work up, everyone levels up" and skip how a junior acquires the judgement the higher rung needs, when the grunt work that used to build that judgement is exactly what the model now does. The doomers say "AI takes jobs" and miss that seniors are largely fine. Both dodge the actual problem: a profession that automates its bottom rung stops producing its top rung a decade later.
This is a real, unglamorous, structural problem and it does not have a tidy answer. If juniors no longer do the repetitive execution that taught the pattern-recognition, the pathway to senior judgement has to be rebuilt deliberately, through different kinds of exposure, deliberate apprenticeship, or roles designed to build judgement without the throughput work as the vehicle. Companies hollowing out their juniors to cut cost are, without noticing, defunding their own future senior pipeline. That bill arrives late and it is large.
What are the failure modes?
Four, split between what individuals get wrong and what organisations do.
Staying at the task level. For an individual: continuing to define your value by the execution AI now does, rather than moving up to the judgement it does not. The role moved; the person did not.
Assuming senior means safe. The abstraction ratchet does not stop at your current altitude. The layer above yours can be moved up too, later. Safe-for-now is not safe-forever, and the response is to keep moving up, not to assume you have arrived.
Hollowing out the juniors. For an organisation: cutting the entry rung for immediate cost, and discovering in a decade that there is no one qualified for the senior roles because the pipeline that produced them was cut. Cheapest today, most expensive later.
Believing the job count nets out. Assuming AI creates as many roles as it moves, evenly. It does not move evenly, it hits the entry rung hardest, and "it'll all balance out" is a forecast, not a fact, made by people who will not carry the cost if it is wrong.
What do you actually do about it?
Move up the abstraction deliberately, and if you run a team, rebuild the pipeline you are about to break.
IS YOUR ROLE MOVING UP OR BEING AUTOMATED?
The Hopium Lab · v1.0 · 22 July 2026 · take it, fork it, argue with it
FOR YOU
1. WHERE IS YOUR VALUE?
[ ] In doing the task -> that is the part moving to the model. Move up
[ ] In deciding what the task is and judging it -> that is the durable part
2. ARE YOU CLIMBING OR STANDING?
[ ] Learning to direct and judge AI output -> climbing
[ ] Competing with AI on throughput -> a race you lose
3. THE HONEST QUESTION
[ ] If a model did the execution, is what remains still YOUR job,
or was the execution the whole job?
FOR YOUR TEAM
4. THE PIPELINE
[ ] Cutting juniors because AI does their work? Where do seniors
come from in ten years?
[ ] Have you designed a way to build judgement without the grunt work
as the vehicle? If not, you are defunding your own future
THE TEST: strip the execution out of your role and describe what is left.
If it is a real job made of judgement, you are moving up. If there is
nothing left, the execution was the job, and up is the only direction
that is not down.
The comfortable framings are both wrong. AI is not coming for everyone, and it is not a free level-up for everyone either. It moves the work up a rung, rewards the judgement that was always the scarce part, and quietly removes the rung people used to climb. The move up is real and it is good for those who make it. The missing bottom rung is the problem the confident takes on both sides are avoiding, and it is the one actually worth thinking about.
Ross Jones, Founder, The Hopium Lab. Last modified 22 July 2026.