How Fast, Not How Many
The rising tide of automation displacement
The current level of AI models is already driving worker displacement. Not just because the AI is getting smarter. Because human ingenuity, fuelled by AI assistance, is constantly improving the tooling built around it. If AI development stopped tomorrow, this disruption would continue. PlayStation games got dramatically better over the life of each console, not because the hardware changed but because developers learned to exploit what was already there. The same dynamic is playing out with AI, except the thing being optimised away is headcount.
The debate about automation and employment usually asks the wrong question. It asks whether machines will replace human workers. Optimists point to two centuries of evidence: the loom displaced weavers but created loom operators, the tractor displaced farm hands but created mechanics and logistics workers, the computer displaced typists but created an entire digital economy. Technology has always created more jobs than it destroyed. Why would this time be different?
The answer is not that all jobs disappear. The rate of displacement is outpacing the rate of new opportunity creation, and the gap is widening. It is not about how many jobs go. It is about how fast.
The mechanism
Consider a team of ten employees, A through J, doing similar knowledge work. Employee A starts using existing AI tools to automate part of their workflow. Not a breakthrough model. The same large language model that has been available for months. A builds a prompt template, integrates it into their daily process, and discovers it handles a chunk of work that used to take hours. A is now noticeably more productive. Employee J, the least productive member of the team, is made redundant. The company has the same output from nine people.
A continues to refine the process. A uses the AI itself to improve the tooling. The AI helps A write better prompts, build more tailored workflows, automate the integration between systems. The model has not improved. A’s ability to exploit it has, assisted by the model itself. A shares the refined process with B. Now both A and B are significantly more productive. H and I are redundant. The company has the same output from seven people.
A and B keep refining. The tooling gets more sophisticated, still built on the same underlying model. G goes. F goes. E goes. The team is now A, B, C, and D doing the work of ten. The AI model is identical to the one A started with. What changed is the accumulated human ingenuity applied to exploiting it, with the AI assisting at every step.
This is not speculative. This is the PlayStation dynamic. Early PlayStation 3 games looked primitive compared to the titles released in the console’s final years. The hardware was the same silicon throughout. Developers learned to extract more from a fixed capability. The same feedback loop (human creativity improving the use of a static platform) is now playing out across every knowledge-work industry simultaneously. The difference is that when game developers got better at using the PS3, nobody lost their job. When knowledge workers get better at using AI, their colleagues do.
What this looks like in practice
I am an infrastructure and automation engineer. I use AI tooling daily. I no longer write code directly. I no longer build infrastructure by hand. I argue with AI to refine a position and then direct its implementation, whether that is how to attack Kubernetes as a red team operator, how to build a new feature into existing tooling, or how to optimise context efficiency in an automation framework. I bring the inspiration and intuitive spark. The AI brings method, compensates for my ADHD executive function gaps, and does the grunt work of writing actual code and documentation. This saves me roughly 60 hours a week. I am at least three times more productive than I was before I adopted this workflow.
The model has not changed. I have changed. My ability to direct it, to structure problems for it, to build tooling around it, has improved through practice and through using the AI itself to refine the process. This is one person’s experience. Here is the data.
The rising tide
The Office for National Statistics reported in February 2026 that 891,000 people aged 18 to 24 in the UK were not in education, employment or training (NEET). That is 15.2% of the age group. The figure is rising quarter on quarter.
Goldman Sachs Research published in August 2025 that unemployment among 20- to 30-year-olds in tech-exposed occupations had risen by almost 3 percentage points since the start of 2025. Tech employment as a share of overall US employment has fallen below its pre-pandemic trend since November 2022. The occupations at highest risk of displacement: computer programmers, accountants and auditors, legal and administrative assistants, customer service representatives. These are not manual labourers. They are knowledge workers.
BT has announced plans to shed up to 55,000 workers, citing AI advances. Not future AI. Current AI.
The OECD reported in 2024 that generative AI exposure is greatest for high-skilled workers and women. This reverses the pattern of previous automation waves, which primarily displaced low-skilled workers and men. The displacement is moving up the income ladder, not staying at the bottom.
None of these sources claim that all jobs will disappear. That is not the argument. The argument is about rates. Displacement is happening at a pace that exceeds the creation of new roles. The tide is rising faster than new ground is appearing. You do not need to prove total submersion to recognise a flood.
Invisible in the averages
The displacement hides. Not from the fiscal system (HMRC will eventually see falling income tax receipts, falling National Insurance contributions, softening VAT, rising benefit claims) but in the aggregate statistics that policymakers watch.
When Employee A does the work of Employees A through J, the company’s headcount drops. But its revenue stays the same or increases. Employee A is more valuable and commands higher pay. Average earnings rise. Corporate profits increase (lower wage bill, same output). GDP holds up. The headline numbers look healthy.
What the aggregates conceal is a widening distribution. Fewer workers earning more. More people earning nothing or much less. The Gini coefficient widens while the Treasury sees stable receipts. Average earnings increase while the median hollows out. Corporate profits rise while benefit claims grow. Every headline indicator says the economy is performing while the lived reality for a growing number of people says the opposite.
The fiscal crisis does not announce itself. It arrives the day the aggregate tips, when the displaced outnumber the remaining productive workers enough to make the averages turn. By then, the structural damage is years old. The time to build the fiscal infrastructure to handle this is before the averages turn, not after.
The fiscal gap
Income tax sees the symptom. Fewer taxpayers, eventually. But it sees it late, because average earnings rise and corporate profits mask the underlying erosion. By the time income tax receipts visibly fall, the displacement is well advanced.
The Sovereign Energy and Bandwidth Excise (SEBE) sees the cause. The energy consumed by the AI tools Employee A is using to do the work of A through J is measurable from day one. Every prompt, every API call, every model inference consumes electricity at a data centre. That consumption is physical, metered, and unavoidable. SEBE captures it at the point of load, in real time, regardless of whether HMRC has noticed the headcount reduction.
SEBE generates £34-46 billion at launch in 2030 (2026 prices), growing to £93 billion by 2040 as automation scales. The full revenue model is derived from first principles using DESNZ and Ofcom data. Revenue grows automatically with automation deployment: more AI tooling means more energy consumption means more SEBE revenue means more capacity to fund the transition for displaced workers.
Income tax shrinks as the problem grows. SEBE grows with it.
The rate, not the number
This is not an argument against automation. Automation makes us more productive. It reduces costs. It solves problems that human labour alone cannot. The infrastructure engineer using AI to do the work of three is producing better outcomes, faster, for less. That is good.
But the fiscal system built on taxing human labour cannot survive a transition where human labour is being optimised away at a rate faster than new opportunities emerge. The question is not whether automation will eventually create new categories of work. It probably will. The question is whether it will create them fast enough, and the evidence from NEET statistics, tech employment data, and corporate announcements says it is not keeping pace.
The tools are already sufficient. The displacement is already underway. The models do not need to improve. The humans using them will keep getting better, keep building better tooling, keep finding new ways to do more with less. That is what humans do. It is also what makes the current fiscal model unsustainable.
The choice is not between automation and employment. It is between a tax system that watches the tide rise and one that captures the energy powering it.
Jason Huxley is an infrastructure and automation engineer with 30 years’ experience in large-scale enterprise systems. He is a member of the Green Party of England and Wales. The full SEBE proposal is open-source at github.com/djarid/SEBE under CC-BY 4.0.