The Next Rung · a reference note
Human Reserved: who decides, and by what criteria
Bill Gates proposed a category of work we keep for people, then said plainly that he could not answer who decides what goes in it, or on what basis. This is an attempt at an answer.
"The idea of Human Reserved raises a host of questions I don't have answers to. Who gets to decide what we reserve for humans? What criteria should we use?"
"These will need to be worked out in public as part of the transition plan."
Bill Gates, 26 August 2026
What Gates actually proposed.
The idea came out of watching his father's care. In the later stages of his Alzheimer's, Gates writes, his father was looked after by paid carers "who understood him even when he struggled to express himself." Something in that work, he says, was "irreplaceably human. No robot could or should have done it."
He is careful about the edges. The reserve would evolve rather than be fixed. Some domains, he suggests, would be mixed, "with a human in charge who's using the technology to extend what they can do". The lines would vary by country, because Japan's shortage of young carers is not Britain's problem.
Alongside it he proposes taxing AI tokens and robots, on the grounds that the current system "nudges you toward replacing people with machines" by taxing payroll while letting a company write off a machine.
It is a more serious proposal than the coverage has suggested, because it concedes something most technology commentary will not: that some allocations of work should be decided rather than left to whatever the price of compute happens to be that year.
When too many people cannot move
"We may do it because allowing machines to take over a certain role will displace a large number of people who can't easily change jobs. You can't tell a 55-year-old who has worked in construction their whole career that they need to go work at an elder care facility and expect them to find it fulfilling."
When a machine could, and shouldn't
"In health, for example, imagine a robot giving you the awful news that you have an incurable disease. There's no technical reason why it couldn't. Yet it shouldn't."
"I like the phrase Human Reserved because it makes me think of nature reserves—places where we could put buildings and roads, but we choose not to because the loss would be too great."
Bill Gates, Gates Notes, 26 August 2026Where his own criterion runs out.
The working test in the essay is whether something is irreplaceably human. It is a good test. I would sharpen it slightly: the question is whether you have to be in the room.
That covers more than care work. You have to be in the room to understand an electrical fault in a hundred-year-old house. A model cannot know what a century has done to a wall by looking at a photograph of it. It is the same quality as a carer who does not need an algorithm to work out that the person in the bed wants another pillow. It is not the cognition that is irreplaceable. It is the presence.
Apply that honestly, though, and it does not extend to people who take meetings for a living. Orchestration work, meaning the chasing, scheduling, formatting, summarising and keeping-the-thread-alive layer that most junior professional roles are built from, fails the in-the-room test completely. Nothing about it requires a body in a room.
Which is exactly why it goes first. And the employers agree. The Institute of Student Employers surveyed 144 organisations for its 2026 Student Development Survey and found 87 per cent expect AI to reshape graduate and apprentice roles, with 29 per cent expecting significant change and 43 per cent reporting that roles have already shifted without being formally redesigned.
Read those two lists next to each other. The tasks on the way out are how people used to acquire the capabilities on the way in.
So under the criterion as Gates states it, the reserve protects care, trust and physical presence, and it leaves the bottom of the professional ladder entirely outside the boundary. That is not an oversight in his argument. It is what the criterion says. Which is why I think it needs a second one alongside it, not instead of it.
Inside the boundary
- Caring for someone who cannot tell you what they need
- Delivering a diagnosis a person has to hear from a person
- Diagnosing a fault in a building with a century of history in it
- Work where the judgement depends on being physically present
Outside it, on his test
- Chasing, scheduling and coordinating
- Formatting and summarising
- Keeping a thread alive between other people
- What the ISE survey calls "routine admin, basic data and writing tasks"
A second criterion: reserve for development, not only for dignity.
I should be clear about where I am standing. I build the software that removes this work. Orca is an AI assistant for executives and small teams, and it does not usually replace a role directly. It much more commonly prevents one from being hired.
A rung removed, with nobody at fault
In the first quarter of this year, a small digital marketing agency, a team spending a large share of every week orchestrating. Orca took a chunk of that away. They were growing, so rather than hire, they grew into the hours that came back. An office coordinator role had been planned to take up the slack. It was never opened.
Nobody lost a job. Nothing appeared in any redundancy figure. The removal was entirely reasonable at the level where the decision was made.
The thing that is not on a curriculum
An office coordinator does not stay an office coordinator. That role is the usual route into operations management, then operations direction, or across into project management.
What makes someone good at those jobs is knowing why things run late. Why some weeks move and others do not. Why a plan survives contact with one client and falls apart with another. You learn it by being the person the mess passes through.
Two years of doing it badly
We hired at the family garage in June, a vehicle technician qualified on MOTs and servicing. Apprentices there start at the bottom: sweeping, making the tea, oil changes, the repetitive jobs.
On paper that is the most automatable work in the building. In practice the repetition, and the mistakes made on things where a mistake does not matter much, are exactly what let someone handle a difficult job three years later.
I have built the alternative, which is why I am sceptical of it. Last year I built the simulation tools at Enterprise Skills, a business I have since exited, designed to teach exactly these capabilities through probabilistic scenarios. They worked better than nothing. But you do not sweat in a sandbox. You can fail a simulation and feel nothing, because nothing was at stake. Competence you can read for. Judgement needs consequence.
There is now economics behind this rather than just anecdote. Acemoglu, Kong and Ozdaglar published work through the NBER in February 2026 on what they call knowledge collapse: when AI absorbs a task, the output carries on, but the learning that used to come attached to the task disappears. The juniors never do the foundational grind that builds judgement.
When you automate the doing, you also automate the learning.
The Next Rung
Can the competence at the next level up be learnt any other way?
If it can, automate the work and lose nothing but the cost. If it cannot, that rung is load-bearing, and removing it does not save a salary. It removes a stair.
The test, worked through.
Reserve it: paralegal work
- You can learn the law by reading it.
- You cannot learn how a matter actually behaves, which client goes quiet, which deadline really moves, what a clerk will accept, without having handled the volume.
- There is no other route to the judgement a solicitor needs. The rung is load-bearing.
Do not reserve it: a reconciliation task
- A task that exists only because two systems do not talk to each other.
- Nobody's future competence is built on top of it.
- Automate it without ceremony.
The difference between the two criteria matters more than it sounds. Reserve for dignity and you protect the work people already do. Reserve for development and you protect the way in. A reserve built only on dignity can be large, generous, well-intentioned, and still leave nobody able to reach it.
What the entry-level numbers actually show.
The figure usually reached for is Adzuna's: UK entry-level roles, meaning graduate jobs, apprenticeships, internships and junior positions, down by almost a third since ChatGPT launched in November 2022, from 28.9 per cent of the job market in 2022 to 25 per cent.
I am not going to present that as an AI number, because the people closest to it say it is not one, or not yet.
They are right, and taking them seriously makes the argument stronger rather than weaker. If the bottom rung is already thinning for ordinary economic reasons, the developmental hole is already opening, and it was opening before any of this. What AI changes is the repair. A downturn ends and firms start hiring juniors again, because they need the hands. Automation does not end. It makes not rebuilding that rung permanently rational.
The cleaner number, for what it is worth, is narrower and better sourced: junior employment down roughly 7 to 13 per cent at firms adopting generative AI, with corroborating analysis showing falls of around 13 per cent for workers aged 22 to 25. Gates makes the same observation in his own essay, that after generative AI was widely adopted "employment fell significantly among young workers in jobs that are especially vulnerable to replacement, but not among their older colleagues".
The honest caveat
When that research was published, Stephen Isherwood, joint chief executive of the Institute of Student Employers, said it was "too early for AI to impact graduate job numbers" and that "a flat economy combined with rise in employers' national insurance contributions is having the biggest impact".
Indeed's senior economist Jack Kennedy said the same thing, that the fall "can't be attributed solely to the rise of AI".
So who decides.
An individual is accountable for their own judgement. Nobody can acquire it on their behalf. An organisation is responsible for providing the conditions in which it can be acquired. A reserve that inverts those will protect a great many jobs and still produce nobody capable of doing them in fifteen years.
Which points at the answer. It should not be a new committee, and it should not be government drawing up a schedule of protected occupations. It should be whoever is already accountable for the supply of judgement in a given profession, and in most professions that body already exists and is already watching the problem arrive.
The Law Society and the SRA
Already define what counts as qualifying work experience.
The accountancy institutes
Already define practical experience requirements and sign them off.
Medical royal colleges
Already define supervised hours.
Apprenticeship standards
Already define what an apprentice must actually have done.
The Monday morning version
None of those bodies needs a new power. They need to answer one question and publish the answer: which parts of qualifying experience can now be produced by a machine and still count, and which cannot?
That is enforceable in a way a general labour-market rule is not, because these bodies already hold the thing firms actually want, which is the ability to certify people.
What breaks if you get it wrong.
If you have never been a junior anything, you cannot become a senior anything. That is the entire argument in one line, and it is not clever, it is just arithmetic with a delay on it.
The delay is what makes it dangerous. Nobody announces this. A team of ten becomes a team of five over a few years as people leave and are not replaced, and the work gets absorbed by whoever stays, because people absorb work in order to keep their jobs. Nothing is cut. Nothing is declared. The team simply stops hiring below itself, and the organisation reads the whole thing as efficiency, because at every individual decision point that is exactly what it was.
Now put the reserve on top of that. In the GeekWire interview published alongside the essay, Gates described working the idea through with Claude, "thinking through different ways to get the share of work reserved for humans up to 40%, using shorter workdays and earlier retirement to spread what's left around".
Both of those levers redistribute work that already exists among people who already have it. Earlier retirement takes people off the top of the ladder. Shorter working days thin every rung a little. Neither one creates a training position. Reach 40 per cent that way and you have built a large, humane reserve staffed entirely by people who got their experience before the reserve existed, with no mechanism for producing the next cohort.
Fifteen years of that and you get professional atrophy: nobody senior enough to teach anybody junior enough to take over. That is not a problem arriving in the 2040s. It is a hiring decision being made this quarter, in a few thousand small firms, by people who are not doing anything wrong.
"The ladder does not just lose a step at the bottom. It loses the training ground that used to produce the people who knew how to climb."
The Next Rung
The questions this raises that I cannot answer either.
Gates listed four open questions. I have taken the first two. The other two are genuinely hard, and I would rather say so than perform an answer.
Gaming
Any boundary creates an incentive to relabel work so that it sits on the profitable side of the line. Write a reserve around supervised junior experience and you will get supervision that is nominal, logbooks signed off for work a model did, and roles that are junior in title and empty in content.
This is harder than the equivalent problem in a nature reserve, because you can photograph a building. You cannot photograph whether anybody learnt anything. I do not have a mechanism, and I am sceptical of anyone who says they do.
Arbitrage
If one country reserves the training rung and another does not, the work moves. Worse, the training rung is the most portable part, because it is the most codified: the parts of a job you can write down are both the easiest to automate and the easiest to send somewhere else.
That is Gates's fourth question and it is the one I find hardest, because the answer probably has to be international, and international agreement on labour standards has not historically been fast.
And what I would like to be wrong about. The optimistic case is straightforward and might well be right: new roles appear that nobody can currently name, as they always have. I would want that. My doubt is about rate, not direction. The question is whether new work appears as quickly as old work disappears, and what happens to the people standing in the gap. It could be a short gap. It could be most of a working life. It will be uncomfortable either way, and the discomfort lands on people whose only mistake was arriving at the labour market at the wrong time.
Evidence that cuts against me
In the same ISE survey, 40 per cent of employers said they expect AI to replace none of their entry-level roles, and a further 42 per cent expect it to affect between 1 and 10 per cent of them.
That is a substantial body of practitioner opinion saying this is being overstated. It is here because a page that only cites the evidence agreeing with it is not worth citing.
Frequently asked questions.
What is Human Reserved?
Human Reserved is a term coined by Bill Gates in an essay published on 26 August 2026. It describes work that machines will be fully capable of doing, which we choose to keep for people anyway. Gates draws the comparison with nature reserves, on the grounds that we could build on that land but judge the loss too great to allow it.
Who decides which jobs are Human Reserved?
Gates said openly that he does not know, and that the question would have to be worked out in public. One answer is that it should not be a new committee. It should be whoever is already accountable for the supply of judgement in a given profession: the professional bodies, regulators and awarding organisations that already define what counts as qualifying experience through training contracts, supervised hours, logbooks and apprenticeship standards. They already make and enforce this kind of decision. What they would need to add is a ruling on which experience can be produced by a machine and still count, and which cannot.
What criteria should be used to reserve a job for humans?
Gates offers two. The first is economic, where automating a role would displace a large number of people who cannot easily change jobs. The second is intrinsic, where the work is irreplaceably human, such as caregiving or delivering a terminal diagnosis. A third is proposed here: reserve the work that is the only route to the judgement above it. The test is whether the competence at the next level up can be learnt any other way. If it cannot, that rung is load-bearing, and removing it does not save a salary, it removes a stair.
Does Human Reserved protect entry-level jobs?
Not as currently framed. Gates's working test is whether work is irreplaceably human, which protects care work and roles requiring physical presence and trust. Entry-level orchestration work, meaning coordinating, chasing, scheduling, formatting and summarising, fails that test completely. That is precisely why it is being automated first. Under the criterion as stated, the bottom rung of the career ladder is not reserved.
What is the difference between reserving work for dignity and reserving it for development?
Reserving for dignity protects work because of what it means to the person doing it, which tends to protect established roles and the top of a career ladder. Reserving for development protects work because of what it teaches the person doing it, which protects the route onto the ladder. They produce very different reserves. A reserve built only on dignity can be large, well-intentioned, and still leave nobody able to reach it.
How would Human Reserved be enforced?
This is genuinely unresolved, and Gates lists it as one of his own open questions. Any boundary creates an incentive to relabel work so it sits on the profitable side of the line. A reserve written around supervised junior experience invites supervision that is nominal and logbooks signed for work a model did. The difficulty is that you can photograph a building inside a nature reserve, but you cannot photograph whether anyone learnt anything. There is no credible enforcement mechanism on the table yet, including in this piece.
Sources.
- Bill Gates, "The turbulent AI era is here. The choices we make now are critical.", Gates Notes, 26 August 2026. gatesnotes.com
- Todd Bishop, "'I am very concerned': Bill Gates says the world needs a plan to deal with AI, and he has three ideas to start", GeekWire, 26 August 2026. geekwire.com
- Institute of Student Employers, Student Development Survey 2026, 144 respondents, published 7 May 2026. ise.org.uk
- Adzuna entry-level jobs research, published 30 June 2025, with commentary from the Institute of Student Employers and Indeed, reported in HR magazine, 1 July 2025. hrmagazine.co.uk
- Hosseini and Lichtinger, on junior employment at generative-AI-adopting firms, 2025, with corroborating Stanford and ADP analysis for ages 22 to 25.
- Acemoglu, Kong and Ozdaglar, on knowledge collapse and the loss of the learning externality, NBER, February 2026.
The long version is a book.
The Next Rung: Why AI is breaking the career ladder and how to build the next one. What AI is doing to the career ladder, and what that does to the economics and the society built on top of it. Underneath all of it sits the question of whether the economy is supposed to work for people, or people for the economy.
Pre-order The Next RungPublished by Practical Inspiration Publishing on 19 January 2027.