The Digital Complication
← The Screening Room About

The Machine Has No Operator

Companion essay to The Screening Room

The page you just used is theatre. There is no AI behind it. There is no role. There is no employer. Everything you experienced was a scripted sequence designed to approximate the feel of a process that is, at this moment, being applied to millions of real CVs by systems that operate in approximately the same way, for approximately the same reasons, with approximately the same result.

The conditions it simulates are not theatre. They are arithmetic.

Applying for a job used to mean writing a letter, posting it, and waiting. It now means submitting a document into a system that will read it before any human does, score it against criteria no one will show you, and discard it for reasons no one will explain.

The volume created this. When applications could be submitted in seconds, they were. When hundreds of applications arrived per role, screening them by hand stopped being possible. When screening was automated, the criteria became whatever the algorithm decided correlated with past success. When candidates learned the criteria, they optimised for it. When optimisation became universal, the system raised its thresholds. When thresholds rose, candidates used AI to optimise further. When AI-optimised CVs became standard, employers deployed AI to detect AI-optimised CVs.

Each step is rational. Each participant is acting in their own interest. Each adaptation makes the problem worse for everyone else. The platforms that host the process profit from the friction in both directions: they sell application tools to candidates and filtering tools to employers. They have no incentive to resolve the loop. They have every incentive to sustain it.

The number that gets reported is 286 per cent. That is the increase in applications per job posting in a single year. It is offered as evidence that AI has made applying too easy, that candidates are carpet-bombing roles they are not qualified for, that the filtering tools are therefore necessary and justified.

The number that does not get reported is the vacancy rate. Vacancies per hundred jobs nearly halved between 2022 and 2025. There are fewer roles. The denominator shrank while the numerator exploded, and the ratio was reported as though only one variable had changed.

If you have five hundred people applying for a job that exists, that is a filtering problem. If you have five hundred people applying for a job that used to be three jobs, that is a structural one. The first problem sells software. The second problem requires policy. One explanation is offered by the people who sell the software. The other is left for the applicant to discover in their spare time, between rejections.

The twentieth-century social contract was straightforward. Work for forty years. Contribute to a system. Retire at a predictable age. Live on a livable income for whatever time remains.

That contract is being rewritten. Not by announcement. Not by debate. By parameter adjustment.

On 6 April 2026, the UK state pension age began rising from 66 to 67. The increase was legislated years ago, debated briefly, and has now arrived with the administrative silence of a direct debit. Spain follows in 2027. Germany in 2031. Denmark will reach 68 by 2030, then 69 by 2035. France raised its minimum pensionable age to 64 in 2023 and the country burned for weeks.

The OECD projects the average effective retirement age across forty-five nations will rise by approximately two years by the mid-2060s. Governments describe this as necessary. Demographers agree the arithmetic supports it. Neither group spends much time discussing what happens when the people who were planning to stop working are instead required to compete for employment in a market that has fewer vacancies, more applicants, and automated screening systems designed to optimise for characteristics that correlate with youth.

Nobody will ever stand at a podium and say: we are cancelling the retirement you were promised. It will simply stop being available, one cohort at a time, while the language used to describe it remains unchanged.

The framing is familiar. Young workers are told that older workers are occupying positions they should have vacated. Older workers are told that younger workers are entitled, unserious, and unwilling to start at the bottom. Both narratives are useful to institutions that would prefer attention directed at each other rather than at the arithmetic.

The arithmetic is simple. There are fewer vacancies. There are more applicants of every age. The screening systems serve neither group. The generations are not each other's problem. They are each other's company in the same waiting room, mediated by a system that treats all of them as data points and none of them as applicants.

The EU AI Act introduces a right to human review in high-risk automated decision-making, including employment. It exists in European law and nowhere else of significance. Minimum standards for algorithmic rejection are technically straightforward and politically untouched. A requirement to explain why a candidate was rejected, to whom the decision was made, on what basis and by what system, would cost almost nothing to implement and would change almost everything about how these tools are deployed. It has not been implemented.

Honest renegotiation of the pension contract is available as an option and declined. The parameters are adjusted; the promise is not updated. The language of retirement persists while the reality of retirement recedes.

Investment in the infrastructure that is visibly failing is deferred in every budget cycle in favour of whatever costs less this year. Healthcare, housing, transport, education. These are choices. They are being made. By nobody in particular.

None of this is being done to you by a coherent villain. There is no meeting where the plan is agreed. There is no memo. There is no conspiracy. There is only a system composed of millions of rational local decisions: by politicians responding to fiscal pressure, by employers managing cost per hire, by HR departments deploying the tools they were sold, by candidates using every advantage available to them. Each optimising their own position. Producing a collective outcome that nobody chose and almost nobody wants.

The machine has no operator. That is the hardest part to accept, and the reason it keeps running.

There is nothing here that works. No data is sent anywhere. No account is created. You may leave at any time. Most people do not.