How AI has turned job hunting into a Kafkaesque experience… and how to escape the trap

And that it doesn’t focus so much on their grades as on far more obscure factors: how many keywords match the job advertisement, whether the layout allows the programme to read the document properly, and other metrics that have nothing to do with merit or excellence.

The game of seduction – peppered with bluffs and trick questions – between employer and employee has turned into a war of algorithms. The result: Kafka has entered the office.

Putting together a decent application used to be quite a chore. You had to read the job advert carefully, tweak your CV and write a cover letter that was a bit more polished (and without any spelling mistakes!).Nowadays, all you need to do is paste the job advert into a chatbot. In a matter of seconds, the machine spits out a CV – or ten, or a hundred – so polished that if Jeff Bezos were to read it, he’d hire you on the spot. The problem is that neither Bezos nor, almost certainly, the employer the CV is addressed to will actually read it. At least not during the initial screening stage, which already feels a lot like a dialogue of the deaf between machines.

The figures are staggering. A graduate of the University of California, Davis, with a near-perfect academic record, sent out 200 letters last year for anything remotely related to his field, and in the vast majority of cases, he didn’t even receive a reply.

At Berkeley – which is hardly a local college – students estimate that it takes around 150 internship applications to secure just one or two interviews. Things aren’t any better here. The InfoJobs website closed 2025 with 2.46 million vacancies and 136 million registrations, equating to 56 candidates per vacancy. On the other hand, seven out of ten companies are unable to find the talent they need.

Hiring a stranger means relying on a few clues, such as grades, dress sense or a willingness to fit into a team. But these indicators are becoming less and less reliable.

Grades are inflated and even at Harvard, top marks are handed out like hot cakes (six out of ten in the 2024–2025 academic year). Fewer and fewer companies are screening CVs based on average grades.

Something similar is happening with cover letters. Two researchers set out to examine what was happening on Freelancer.com, a platform where thousands of professionals bid for assignments.

For years, a well-crafted cover letter was the best asset. Then ChatGPT arrived. “Basically, the whole system came crashing down,” sums up Jesse Silbert. The quality of cover letters improved across the board. Moreover, as they are all written using the same templates, they have become standardised and no longer reveal anything distinctive about the applicant.

Consultancy firms working with large companies estimate that half of their clients now consider the cover letter to be useless, and many have removed it from the recruitment process altogether.

Then there’s the interview, which is almost never face-to-face until the final stage.

The candidate sits down at their computer at nine o’clock in the morning, wearing a freshly ironed shirt. They have prepared their answers and set a digital background. If they’re one of the savvy ones, they’ve also got a programme open that listens to the questions and prompts the answer on screen, much like a newsreader’s.

At the top of the hour, a window pops up and an avatar greets them. There’s no one on the other side. The conversation is recorded; a programme transcribes it and measures parameters such as the speed at which they speak, their eye movements and the length of their pauses. It uses all this to calculate a score.

At the end of 2024, the ResumeBuilder website surveyed nearly 1,000 US executives. Half of their companies were already using artificial intelligence for recruitment: of these, eight in ten were using it to scan CVs; four in ten had a chatbot handling initial contact with candidates; and nearly one in four were letting the machine conduct the interview. In Spain, half of all companies use AI at some stage in the recruitment process.

Caught up in this game of cat and mouse, everyone loses. A study has given this disaster a name: algorithmic monoculture. And, as many companies buy similar tools, they end up making decisions based on the same biases.

How are companies responding? By looking at what is difficult to fake. Practical tests based on real-life scenarios are making a comeback, and elite universities, trial periods and previous employment (where applicable) are becoming increasingly important. There is also much greater emphasis on who recommends you.

But there’s another problem. The tasks assigned to new recruits (struggling with Excel, putting together a presentation…) used to be the training ground for the profession.

You learn by going through 200 tedious contracts. If work placements and junior roles are in short supply, careers get off to a later start, the talent pool dries up, and in a few years’ time many companies will find themselves short of middle managers and specialists.

That is why IBM announced in February that it was going to triple its recruitment of junior staff. Its head of human resources, Nickle LaMoreaux, summed it up as follows: "We need to rewrite every job description." What a novice used to do two or three years ago is now done by a machine, so new recruits are put in charge of dealing with clients and deciding when to call in someone with more seniority. Dropbox has increased its number of internship places by 25 per cent for similar reasons. But these are exceptions.

And it is becoming increasingly common for vacancies to remain unfilled. In the United States, the figures for the last three years paint a very unusual picture of the labour market, with unemployment at around 4 per cent – which looks good, until you realise that new hires have fallen.

There were already reasons not to recruit, or to do so with great caution, even before ChatGPT: the shock of the Great Resignation, when people were walking out just a few weeks after being hired.

Perhaps AI arrived just in time to provide a plausible justification for a decision that companies were already keen to make. Jay Caspian Kang explained it in ‘Executives’ jobs hang in the balance depending on how they handle the latest technology and that makes them extraordinarily compliant with whatever Silicon Valley dictates each season’.

On the other hand, there are also some unexpected… and ingenious reactions. When it comes to job interviews, Roy Lee, a student at Columbia University, made a splash in 2025 by launching a programme which, whilst you’re being interviewed via computer, listens to the question and whispers the answer onto a layer of the screen that the interviewer cannot see.

He used it during an interview with Amazon and posted the video online. His job offer was withdrawn and the university suspended him for a year, but he set up a company, Cluely – a virtual meeting assistant that provides instant answers without being detected – and is now a millionaire.

Daniel Susskind: “AI, when used properly, can greatly improve our lives. It shouldn’t be lumped together with social media.”

British economist Daniel Susskind (born Watford, 1987) is a leading voice in the debate surrounding technology and the future of work.

Professor of Business at Gresham College, London, and research associate at the Oxford Institute for Ethics in AI and the Stanford Institute for Human-Centered AI, Susskind recently published What Should My Children Do?, reflecting on how to prepare the next generation for an AI-driven labour market.

Susskind, a father of three children aged two, five, and eight, discusses educational reform, technological uncertainty, and the future of skilled professions.

Q. What should our children do in the face of what lies ahead?

A. The first thing is to admit that the traditional approach no longer works: predicting what jobs will be available and deducing from that what we should teach. That no longer works.

Q. Why not?

A. Let me give you an example. The UK began teaching programming in all schools in 2013. It seemed like a brilliant idea. Fast forward ten years. What do these AI systems do better than almost anything else? Write code. We promised a generation a skill for life, and it became obsolete before they’d even finished school.

Q. But is it reasonable to try to stay one step ahead of events…

A. Right. But we’re not doing it properly. Instead of trying to eliminate uncertainty, we need to accept it and reframe the question: how do we prepare the next generation to thrive in a future about which we know very little?

Q. Right. And what is the little we do know?

A. Two things. Firstly, it will be full of technologies that are far more capable than those we have today. Secondly, we don’t know anything else. I know that’s rather discouraging, but there’s a great deal we can do.

Q. Where shall we start?

A. The basics: reading, maths and critical thinking. Although it’s impossible to know what advanced skills will be required in the future, they all rely in some way on these abilities. And that’s the one thing we won’t regret having taught.

Q. Should the use of AI be restricted?

A. We need to teach people how to use it properly, and we also need to teach them how to manage without it. We’ve seen this before: in the 1970s, many people feared that the calculator would put an end to mental arithmetic, just as today there are fears that AI will put an end to reading.

Q. So what do we do about social media?

A. I wouldn’t lump them all together. I support restrictions, such as no mobile phones before the age of 14 and no social media before the age of 16. But I’m concerned that this could lead to a broad-brush approach to technological conservatism, and that we’ll end up banning AI in the classroom, at university or in the workplace as well. That would be a mistake. Social media takes power away from us and fragments our attention; artificial intelligence, when used properly, can greatly improve our lives.

Q. The Economist headlines that "the job apocalypse has been postponed"…

A. If we look at the labour market as a whole, there is no shortage of jobs. However, from the perspective of young people, the picture is rather bleak. Entry-level roles in companies are disappearing. And whilst this is a serious issue now, it will also have wider consequences. If these roles disappear, the training they provided will disappear with them.

Q. How does one learn a trade in the age of AI?

A. There are ways. An astronaut flies a space shuttle in a simulator before heading into space, and a power station manager operates the reactor in a simulator before sitting in front of the real one. Imagine a novice lawyer arguing a case before a simulated judge powered by artificial intelligence, with cases and materials generated by AI. We’ll give them, in one year, the complex work that used to take ten years to get to.

Q. Hang on, what about humans interacting with humans?

A. We should bring back mentoring – the apprentice learning whilst sitting alongside their master or tutor. That personal, guild-like relationship will have to make a comeback in many professions, because the kind of office work where people used to learn is going to disappear.

Q. And what does the future hold for doctors?

R. In 2017, a team of computer scientists built a system that could determine from a photograph whether a mole was cancerous with the same accuracy as the best dermatologists. One of the researchers was Sebastian Thrun, a pioneer in autonomous vehicles. Someone with hardly any medical knowledge solved a medical problem. That’s why I tell young people to choose the problem they want to solve, rather than the profession they see today. If what they want is for people to be healthier, let them study medicine – but they should know that in ten or twenty years’ time, they’ll be treating patients in ways they can’t even imagine now. They might never even wear a white coat or see a patient in a surgery.

Q. You say that AI can provide every child with a private tutor…

A. An average pupil with a private tutor usually ends up outperforming almost all their classmates in a standard classroom. But a private tutor is extremely expensive. Now, with AI, we can do this at a much lower cost, and tailor teaching to each student’s strengths and weaknesses. And I’ll be honest with you: AI offers a standard of teaching that neither I nor many others can match.

Q. Are accountants, consultants and middle managers the ones who are worst off?

A. That’s where routine office work is concentrated. And whilst these technologies are very good at cognitive tasks, they are considerably less effective when it comes to the manual dexterity required by the trades.

Q. Should we tell our children to become plumbers, physiotherapists...?

A. I try not to hand out ‘safe work’ passes, but if you’re asking me where the machine is at its most difficult, that’s where you’ll make the most progress.

THREE QUESTIONS TO FIND OUT IF YOU’LL KEEP YOUR JOB

1 | IS YOUR TASK SIMPLE OR COMPLEX?

The economist Luis Garicano explains it as follows: there are jobs in which tasks can be separated (they are ‘simple’ or a ‘weak set’) and others in which they all come as a single package (complex or a ‘strong set’). The radiologist reads X-rays – which is what the machine does well – but also talks to the patient, consults the doctor and explains the results. Without knowing the patient’s medical history, it is difficult to interpret an image correctly.

2 | IF WHAT YOU PRODUCE BECAME CHEAPER, HOW MUCH MORE WOULD PEOPLE WANT?

In the case of radiology, there has been another effect: as tests have become cheaper, doctors are ordering far more of them. Consequently, more radiologists are needed.

This is known as ‘Jevons’ paradox’ – the same phenomenon that led banks, following the introduction of the cash machine, to hire more counter staff for decades: as it was cheaper to run a branch, they opened many more, each with fewer staff but with more staff overall. You might say that the number of bank staff is falling, but this is down to an invention nobody saw coming – the mobile phone and its galaxy of banking apps – and comes after they had survived for half a century in the face of the cash machine.

The classic example is that of Henry Ford and the motor car. When Ford introduced the assembly line in 1913, which made it possible to produce far more cars with far fewer workers, it did not lead to a collapse in employment in the industry; quite the opposite happened.

As cars were cheaper, more were sold, and they had to take on many more workers to meet the increased demand.

3 | IS THE AI THE EXPERT, OR IS IT YOU?

Technology affects professions to varying degrees depending on whether it ‘complements’ or ‘replaces’ human expertise. Between 1980 and 2018, the report shows that automation had opposite effects in two sectors: stock clerks (who manage stock) saw their wages fall by 13 per cent as their tasks became simpler. As the work became easier, there was a much larger pool of potential workers, which drove down their wages. For bookkeeping assistants (who handle day-to-day accounts), the opposite occurred: their earnings rose by 40 per cent as they were freed from routine tasks to focus on complex analysis. MIT economists David Autor and Neil Thompson demonstrated that this pattern was repeated across more than 300 occupations over a forty-year period.

Now, with artificial intelligence, we are entering uncharted territory. In theory, and according to the ‘technoligarchs’ who ‘own’ it, AI does not seek to compete directly with humans, but rather to raise their technical standards, enabling workers with basic training to solve complex problems that previously required specialised professionals.

And there are examples to back this up, such as Schneider Electric’s ‘electrician’s assistant’ tool, which enables an ordinary electrician to solve complex problems that previously required engineers with a postgraduate qualification. “I think we’ll start to see more and more cases like this, where AI helps to raise the level of expertise,” says David Autor.