Bill Gates has put a more direct policy question into the discussion about AI, robotics and work. In a long essay reported by TechCrunch on 26 August, he argues for a “robot tax” and for roles or tasks that society might deliberately reserve for people. He calls the second idea “Human Reserved”.
The framing matters. Gates is not announcing a tax policy, and he is not claiming that a named list of jobs will disappear. He is proposing a way to think about the choices made when machines can perform a growing share of economic activity. His worked example is a 55-year-old construction worker who has spent a career in a trade and is told to retrain for elder care.
The example moves the argument away from an abstract question about productivity. It asks who bears the cost when a task is automated, what retraining can realistically achieve, and whether all technically possible substitution is socially desirable.
The robot-tax argument starts with the current incentive
Gates’s starting point is the different treatment of hiring people and buying automation. “Right now, if you’re an employer and you hire someone, you pay payroll taxes on their earnings,” he wrote. “But if you buy a robot, you can usually write it off right away as a business expense.”
His claim is that the tax structure can make replacement with machines more attractive than retaining workers. A tax, he wrote, “would slow the rush away from human labour a little and raise money for retraining and a stronger safety net.”
That is a policy argument rather than an established outcome. It depends on how a tax would be designed, which technologies it would cover, who would pay it, and what would be done with the revenue. Yifan Powers of the Tax Policy Center has identified one core difficulty: defining a robot tax.
Automation is not a single category. Some technology substitutes for human tasks. Some complements a worker’s activity. Other systems alter a workflow without making a simple one-for-one replacement possible. A rule would need to distinguish among those effects without treating every productivity tool as a replacement for a person.
Michael J. Ahn, writing for Brookings, argues that a robot tax could help support people displaced by automation and prompt more deliberate consideration of deployment. He also points to problems of scope, liability and international coordination. The conditional language is essential. These are live policy debates, not settled conclusions about what a tax would do.
Why the construction example is significant
Gates’s example is specific because it makes the transition problem hard to dismiss. “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,” he wrote.
This is not an argument that construction and care are equivalent jobs separated only by training. It is an argument that work carries accumulated experience, physical capability, professional identity, location, pay expectations and social relationships. A person can learn new skills while still losing the value built over many years in a trade.
Construction is also useful as an example because the work contains a mix of physical activity, coordination, judgement, planning, management and documentation. A change to one task does not automatically mean a whole occupation has been replaced. The more meaningful question is whether the tasks that remain provide a viable pathway for people to develop and use their expertise.
The International Labour Organization makes a similar distinction at a global level. Its 2025 update found that one in four workers worldwide is in an occupation with some degree of exposure to generative AI. The ILO does not treat exposure as a prediction of redundancy. It says that, because human input remains necessary, most jobs are more likely to be transformed than made redundant.
The methodology also matters. The ILO’s updated work combines task-level data, expert input and AI predictions across nearly 30,000 tasks. It is a careful attempt to measure potential occupational exposure, not a list of jobs that will certainly vanish. Its conclusion is that transitions should be managed through social dialogue, with attention to both working conditions and productivity.
“Human Reserved” is a choice about what work is for
Gates also gives a health-care example in which a machine could technically deliver news of an incurable disease but, in his view, should not do so. The point is that technical capability and social acceptability are not the same test.
The phrase “Human Reserved” borrows the language of protected places. Gates’s argument is that some activities might be reserved for economic reasons, because automation would displace people who cannot readily move into another role. In other cases, the reason could be dignity, care, trust or the need for human accountability.
The idea should not be mistaken for a claim that machines cannot do particular tasks. It asks whether capability should be the only test for deployment. That is a social and political question, not a purely technical one.
There are difficult choices embedded within it.
• The basis for protection: Would a task be Human Reserved because it supports entry into a profession, because of the consequences of a mistake, because of the relationship with the public, or because of large-scale displacement?
• The decision-maker: Employers, professional bodies, trade unions, regulators and the public may have different answers. A workable policy would need a legitimate way to reconcile them.
• The boundary: It is easier to name a sensitive interaction than to define an entire occupation. Most roles involve a mix of tasks, and that mix changes as technology changes.
• The review process: Gates suggests that a protected area could evolve over years or decades. Any rule would need regular review because the technology, labour market and public expectations will all change.
The OECD provides helpful context for why the discussion cannot focus only on unemployment. It identifies potential benefits from AI, including productivity, job quality and occupational safety and health. It also identifies risks around automation, loss of agency, bias and discrimination, privacy breaches and lack of transparency.
Its surveys of workers and employers in manufacturing and finance do not provide a verdict for every sector. They do show that training and worker consultation are associated with better outcomes for workers. That finding supports a straightforward conclusion: the implementation process matters, not just the tool.
Skills policy cannot be reduced to technical training
The OECD’s 2026 brief on AI and skills makes another useful distinction. It says fewer than 1% of workers will need advanced AI-specific skills such as programming or model development. For most workers, the relevant capabilities are digital skills and the ability to use, analyse and interpret data. Managerial skills and human skills, including problem-solving, creativity and innovation, also remain important.
That is relevant to Gates’s construction example. A transition plan cannot assume every affected worker needs to become an AI specialist. It has to consider how existing professional knowledge can be combined with new tools, and whether training is available in a form that people can realistically use.
The same OECD brief says skills shortages are a significant barrier to AI adoption. Around 40% of non-adopting employers in manufacturing and finance cited skills as the main barrier, as did more than half of SMEs not using generative AI. It also reports that workers who receive training are more likely to report positive outcomes from AI adoption. The underlying evidence has limits, and the OECD notes data gaps, but the pattern challenges the idea that technology can simply be introduced and left to settle itself.
Gates’s proposal is therefore larger than a tax question. It raises questions about who is consulted, who is trained, who is compensated, and which forms of human work a society chooses to value even when alternatives are available.
Takeaway
The most useful reading of Gates’s proposal is not as a forecast that robots will replace construction workers or that a robot tax is imminent. It is a prompt to look at automation decisions through the worker’s experience as well as the organisation’s efficiency case.
Three points should remain clear:
• Exposure is not the same as redundancy. The ILO’s evidence points towards transformation in many roles, so claims about inevitable job loss should be treated carefully.
• A tax would be difficult to design. Definitions, scope, incentives and international coordination would all shape its effects.
• Transition is more than retraining. Training, worker consultation, agency and the value of accumulated expertise all affect whether a change is workable for people.
The phrase “Human Reserved” is provocative because it asks whether work has value beyond the output it produces. That is the debate Gates is inviting. It will not be settled by the technical capability of an AI system alone.
The Project Flux newsletter follows the AI questions with consequences for real jobs, firms and professions. Join it for the next issue.
Links and Stuff
All content reflects our personal views and is not intended as professional advice or to represent any organisation.
1