AI and the invisible tsunami in the labour market

While many people still regard artificial intelligence as an office novelty, some experts are already warning of a silent crisis.

Behind the scenes in Silicon Valley, AI investors and researchers are sounding the alarm: a new generation of systems is transforming not merely tools, but the underlying logic of work itself. Anyone who continues to dismiss the subject as a technology fad risks reliving the feeling of March 2020, when the world realised too late that the virus was not simply “a distant problem in China”.

The 2026 turning point: when AI began rebuilding itself

On 5 February 2026, two launches largely escaped public attention, yet insiders regarded them as a watershed: OpenAI’s GPT-5.3 Codex and Anthropic’s Opus 4.6. These were not just more “intelligent” models; they were systems able to intervene directly in their own development process.

AI is moving from being a tool that helps the programmer to becoming the chief engineer of its own evolution, closing a loop of continuous self-improvement.

According to technical documents released by the companies, preliminary versions of GPT-5.3 Codex were used to debug the code behind their own training, tune parameters and examine performance failures. Put simply, AI does not merely carry out tasks: it also helps build its own next, more sophisticated version.

This shift overturns the notion of linear progress. Where human teams once refined models year after year, the curve now becomes steeper. Every new generation of AI makes a more substantial contribution to creating the one that follows. Dario Amodei, Anthropic’s CEO, expects that within one or two years this cycle could function with almost complete autonomy, requiring minimal human intervention.

From programmer to observer: the new human role

For developers, the shock is already tangible. Entrepreneurs such as Matt Shumer say they have stopped “programming line by line”. He describes workflows in which he explains in natural language what he wants a system to do, steps away from the computer for a few hours, and comes back to a finished product that has been tested, refined and polished to a standard above that of a senior specialist.

In this setting, the technology professional ceases to be the craftsperson writing code and instead becomes something like a screenwriter, editor or product manager. In many instances, they are almost a qualified observer. Written prompts replace the keyboard. That shortens development time, but it also reduces the need for large human teams.

When one person equipped with advanced AI can produce the work of an entire team, the employment arithmetic simply does not add up.

The invisible tsunami in the labour market

The comforting illusion is that this wave affects only software engineers. Shumer and other experts warn that code was merely the first frontier, because AI needed to master programming in order to speed up its own development. Once that stage is passed, its target expands to almost anything involving structured reasoning.

Law, finance, medicine, accountancy, marketing, journalism, design, customer service and human resources: almost every activity based on text, numbers, images or standardisable decisions comes into view. The original promise of “automating repetitive tasks” is giving way to something broader: a general-purpose replacement for cognitive effort.

Dario Amodei forecasts the elimination of up to 50% of entry-level office roles over a period of one to five years. This is not limited to call centres or junior roles in banks. Junior analysts, legal assistants, beginner copywriters, consultancy trainees and even hospital residents risk seeing parts of their work absorbed by systems that are increasingly inexpensive and available around the clock.

No escape route: why this revolution differs

Earlier technological transitions offered safe havens. When machines removed factory jobs, many workers moved into offices. Now the office is under pressure as well. Any plan for professional “reinvention” must account for the fact that AI is already ahead in many areas.

The old strategy of “studying something more stable” loses force when even traditional careers are being rewritten by algorithms trained on billions of data points.

The shift reaches fields once considered naturally protected from automation, including journalism and content creation. Generative models produce text, scripts, images and videos in seconds, adjusting tone, style and depth according to the instruction given. Reporters, once responsible for every stage, now compete for relevance with machines able to cover financial results, sports scores and even preliminary legal analysis.

Who faces the most immediate risk?

There is no definitive list, but experts identify several functions as more exposed in the years ahead:

• Repetitive office work, including data entry, spreadsheets and standardised reports.
• Customer service via chat, email or telephone where scripts are predictable.
• Mass content production, such as product descriptions and simple press releases.
• Basic legal support, including reviewing standard contracts and researching case law.
• Back-office routines in banks, insurers and large companies.

At the same time, niches are emerging where people still make a visible difference: setting strategies, making ethical decisions, designing public policy, managing crises, leading mixed teams of people and AI, and, above all, critically overseeing the automated systems themselves.

How to prepare without panicking

The pandemic analogy appears frequently among analysts: before 2020, most people ignored technical reports about an expanding virus. Something similar is now happening with warnings about AI’s impact. The difference is that there is no visible lockdown, no hospital queues and no daily headlines pointing to the problem. The risk is growing quietly, inside IT departments and innovation teams.

A number of practical steps can help reduce individual vulnerability:

Action	Why it makes sense
Learn to use AI tools in everyday work	Professionals who understand these systems are more likely to be retained to orchestrate hybrid workflows.
Build critical-analysis and decision-making skills	Machines generate options, but there is still room for people to set direction and take responsibility.
Seek fields requiring direct human contact	Healthcare, education, complex negotiation and leadership still require empathy and presence.
Keep updating your skills continuously	Reinvention cycles are becoming shorter; those who stop learning quickly become obsolete.

Terms that are changing meaning in the AI era

Some concepts acquire new connotations in this environment. “Autonomy”, for instance, no longer means only operating without constant supervision; it also means a system’s ability to define intermediate steps, create internal tools and adapt to failures without detailed instructions.

Another key term is “cognitive substitute”. It refers to systems that do more than perform mechanical tasks: they take on entire portions of human reasoning, such as planning a project, selecting legal approaches or building a complete investment portfolio based on a client’s objectives and constraints.

Possible scenarios for the coming years

One likely path is an uncomfortable coexistence of productivity gains and staff cuts. Companies that adopt AI aggressively may produce more with fewer people, pushing competitors to follow the same route. In low-margin sectors, the pressure to cut costs is likely to be brutal.

At the same time, public policy could create buffers: regulation of AI use in certain sectors, professional reskilling programmes, tax incentives for companies that retain human teams in critical roles, and even discussions about a minimum income linked to automation.

In practical terms, anyone currently in an office job needs to consider personal scenarios: what would happen if half the tasks in their sector were automated within two years? What new responsibilities could justify their continued role? Which skills could they realistically develop during that period?

These questions may sound harsh, but they work as an early-warning radar. The difference between being overwhelmed by the wave and learning to ride it lies in seeing AI not as a distant curiosity, but as a central factor in career decisions from now on.