Technology giants maintain an aggressive discourse about the potential of their systems to take over functions traditionally performed by people. While a considerable portion of professions are moving towards total automation, other careers are likely to undergo a profound transformation, requiring the constant use of these digital platforms on a daily basis.
Executives from large corporations around the planet direct billions of dollars to these technological innovations. The main driver behind this massive volume of investments is the real prospect of reducing payrolls and reducing operational costs with human resources.
The concept that keeping the workforce intact already represents a form of growth has gained strength on boards of directors. This mentality raises debates about the need to hire new real professionals, when highly trained virtual agents can already assume complex responsibilities autonomously.
If this corporate dynamic is consolidated, the effects will be felt in practically all areas of the global economy. The transition to this new production model is progressing at a speed much higher than initial projections, forcing governments and workers to seek immediate means of adaptation.
Expert alerts and companies’ race for technological qualifications
Nobel Prize winners in Economics recently published a manifesto demanding urgent action from global authorities. Academics argue that the adoption of these tools should serve to increase the population’s quality of life, avoiding mass layoffs. According to recent data from the World Economic Forum, it is estimated that automation will affect almost a quarter of jobs by the end of the decade, which reinforces the need for solid public policies to cushion the social impact of this transition.
On the same topic: OpenAI expands ad testing on ChatGPT to boost revenue
Over the past month, several companies based in London have reported a growing obstacle: a shortage of professionals who possess the new skills required by automated systems. The job market is undergoing a severe reconfiguration and, even though the scenario is still changing, current statistics already outline clear trends.
Recent surveys have mapped the percentage of corporate activities successfully completed by different virtual platforms. The criterion used for measurement was the time that a qualified human professional would spend to deliver exactly the same result.
Performance evolution of the main enterprise language models
Historical monitoring of the performance of these platforms reveals significant efficiency jumps in a period of just three years. The data details problem-solving capabilities according to complexity and required turnaround time:
Learn more: Change at Google: DeepMind creator takes on unprecedented role focused on general artificial intelligence
- GPT-4, made available in March 2023, was able to resolve 85% of demands that took less than five minutes, but delivered only 15% of tasks of medium duration (five to 59 minutes) and failed in all activities longer than one hour.
- GPT-4o, which hit the market in May 2024, maintained practically identical numbers, solving 83% of quick routines, 15% of intermediate ones and zeroing long-term tests.
- Claude 3.5 Sonnet (New) (Inspect), released in October 2024, marked the beginning of a change, completing 88% of short tasks, 42% of medium ones and achieving 3% success in operations exceeding sixty minutes.
- o3 (Inspect), introduced in April 2025, raised the productivity bar by completing 98% of quick demands, 85% of tasks up to an hour and 17% of longer jobs.
- Gemini 3.1 Pro, February 2026, achieved excellent levels by processing 99% of five-minute actions, 96% of intermediate actions and 46% of complex activities.
- The Claude Mythos Preview (initial version), also from April 2026, broke records by solving 99% of short tasks, 96% of medium ones and an impressive 67% of jobs that would require more than an hour of human dedication.
These indicators serve as the technology industry’s main compass to measure the evolution of machines over intellectual work. The focus of the assessments is mainly on the creation, review and maintenance of programming codes.
Reading these numbers proves that large language models (LLMs) are no longer useful tools only for shortcuts lasting a few seconds. Today, platforms process intricate logic that would keep an experienced employee in front of a computer for more than an hour.
The most modern versions already audit cryptocurrency contracts in search of vulnerabilities and can rewrite parts of the code itself to improve its performance. Such audits would cost entire software engineering teams days of work.
There is a strong expectation that the next updates, scheduled for next year, will allow systems to create new versions of themselves from scratch. This degree of technological independence marks a new era in computing.
Although programming concentrates the greatest advances, the same phenomenon begins to invade other corporate departments. Financial risk analysis, legal process screening and entry-level positions in advertising agencies are already feeling the pressure of automation.
Real impacts on hiring and the vulnerability of young professionals
The most complete reports on the topic come from the United States, crossing four years of information on hiring and firing by age group. The study separates highly exposed professions, such as programmers and support staff, from shielded careers, such as nurses, early childhood educators and beauticians.
More on this story: New professional network function allows you to hide publications created by artificial intelligence
Researchers at Stanford University, in California, identified a 2.7% drop in the employability of young people between 22 and 25 years old since the commercial explosion of ChatGPT. The impact rises dramatically to 12.8% when the analysis is restricted to financial, technology and creative economy departments.
Part of the economic community disputes the exclusive attribution of this drop to algorithms. Some analysts argue that the monetary tightening and the rise in American interest rates also cooled hiring in these specific sectors.
The profile of vacancies advertised on recruitment platforms underwent a visible mutation after the popularization of LLMs. Job descriptions and technical requirements have radically changed focus.
The Organization for Economic Co-operation and Development (OECD) published a recent document isolating traditional economic variables. The entity confirmed a huge disparity in the creation of vacancies between vulnerable areas, such as legal services and telephone sales, and immune sectors, such as civil construction and gastronomy.
The United Kingdom appears in the reports as one of the territories most affected by this decline in office vacancies. The decline in hiring occurred precisely at a time of stability and subsequent fall in British interest rates.
The reduction in job offers also happened before the adjustment in the local social security contribution, National Insurance, implemented last year. This rules out the hypothesis that taxes would be solely to blame for the downturn.
The British economy depends heavily on the services sector, which puts the country at the top of the list of nations susceptible to a wave of structural unemployment caused by the mass adoption of virtual agents.
The high cost of data processing and the rationing of virtual resources
The revenue of technology companies is based on the consumption of “tokens”. These digital fragments are the unit of measurement that systems use to read, interpret and return texts. In practice, a token represents approximately three-quarters of a word in the English language.
The year 2026 saw an unprecedented explosion in the volume of requests to servers. The amount of data processed grew at a much faster pace than the drop in prices charged by developers.
Learn more: Anthropic leader predicts structural unemployment by AI and suggests taxing the sector
To control spending, multinationals created internal consumption classification tables. The objective is to ensure that only employees who truly deliver significant productivity gains have unrestricted access to the most powerful and expensive versions of the algorithms.
The processing volume has reached quadrillions of tokens in recent months. The overwhelming majority of this traffic comes not from humans typing in questions, but from autonomous systems operating in the background to solve chain problems.
The result of this unbridled automation came in the form of million-dollar invoices for corporations. The financial shock forced technology directors to cut several teams’ access to cutting-edge servers.
This financial bottleneck reveals a practical ceiling for replacing humans with machines. Depending on the volume of data processed, keeping a virtual agent running 24 hours a day can cost more than signing up a traditional employee.
The viability of automation now depends strictly on the complexity of the task. To avoid exorbitant charges, Western companies began to migrate their operations to open source platforms, many of them developed in China and distributed without licensing costs. The global job market continues to navigate a sea of uncertainty, but the restructuring of corporate routines is already a path of no return.