The debate around Artificial Intelligence is no longer solely focused on whether it “helps” or “replaces,” but on the speed at which it is improving and how this acceleration becomes a real pressure within companies. In recent days, Geoffrey Hinton — one of the most influential figures in deep learning and popularly known as the godfather of this revolution — has once again put numbers to the dizzying pace: according to his vision, progress is so rapid that approximately every seven months systems can perform tasks that, just a short time before, took twice as long.
This isn’t a phrase meant to scare on its own. It functions as a simple mental rule to explain why 2026 could bring a wave of difficult business decisions: if technology improves in short cycles, the gap between “this is experimental” and “this is profitable” also shortens. And when profitability becomes accessible, changes stop being hypothetical.
The metric that sums up the pace: “every seven months, double”
In practice, talking about “doubling capabilities” doesn’t mean a tool does magic. It means that reduces time, cost, and friction when performing repetitive or structured tasks: drafting standard responses, classifying incidents, summarizing documents, proposing code drafts, generating reports, creating marketing materials, analyzing tickets, or guiding users through administrative processes.
The important nuance is that many of these tasks don’t disappear: they are compressed. Where previously multiple people were needed, now a single person supervising and correcting might suffice. And that’s the quiet blow to employment: it’s not about a role “disappearing” overnight, but about the workload per employee growing enough that some positions become redundant.
Hinton has specifically emphasized the impact on “routine intellectual jobs,” a term that encompasses customer service, back office, administrative tasks, and document management. These are essential functions for thousands of companies day-to-day, but they are also the most susceptible to automation as technology becomes good — and affordable — at understanding context, following instructions, and maintaining conversations.
Why 2026 and not “some day”?
Because 2026 arrives with a factor that didn’t exist a few years ago: adoption is already normalized. Companies aren’t just experimenting anymore; they’re integrating assistants into workflows, connecting tools to CRMs, automating internal responses, generating drafts, and using models for support, sales, or operations.
And when the market moves, an incentive becomes hard to ignore: if your competitor reduces times and costs through automation, you also must do so to maintain margins. This dynamic turns Artificial Intelligence into a race for efficiency, and in such a race, technology doesn’t need to be perfect; it just needs to be “good enough” at a reasonable cost.
The likely result, according to various international analyses, is uneven adjustment: some roles will face cuts, others will transform, and some will grow. This isn’t the first time a technological disruption has occurred, but it could be among the first where potential substitution affects office and service tasks on a massive scale, not just physical or industrial work.
The other side of the message: reinventing oneself is not optional
In contrast to headlines framing the debate as “Artificial Intelligence destroys jobs,” data and history paint a more complex picture. For instance, the World Economic Forum projects an intense reconfiguration of the labor market in coming years: creation of new roles, disappearance of others, and especially, a rapid change in the skills considered “basic” in most jobs. Its latest report on the future of employment highlights that a significant part of current skills will need updating to adapt to the new scenario. (World Economic Forum)
Meanwhile, organizations like the IMF have warned that AI will impact a large proportion of jobs, especially in advanced economies, where automation could more intensely affect cognitive and administrative tasks. (Hugging Face)
Translated into real life: reinventing oneself doesn’t mean “learning to code” or becoming a data engineer overnight. It means understanding which parts of your work are automatable, which require human judgment, and how to reorient your profile toward higher-value tasks.
Which profiles tend to resist better (and why)
Without promising magic formulas, clear patterns emerge:
- Work with responsibility and risk: compliance, safety, decisions with legal or reputational impact. Automation can assist, but the final sign-off typically remains human.
- Complex human relationships: negotiation, leadership, crisis management, consultative sales, social or healthcare work. The tool helps, but trust cannot be replaced.
- Deep domain knowledge: specialists combining sector experience with the ability to validate results and detect errors.
- Supervision and control: profiles skilled in designing processes and auditing results (quality, traceability, biases, privacy).
Conversely, the most exposed roles are not typically “office jobs” in general, but those based on repetitive procedures, simple metrics, and high volumes of similar cases.
A disruptive twist: training lags behind the schedule
The social problem isn’t just task replacement: it’s the gap between technological speed and the rate at which people can reorient themselves. For workers, changing roles requires time, resources, and opportunities. For companies, redesigning processes and training teams costs money and needs leadership. For governments, updating educational frameworks and active employment policies is rarely quick.
This explains many experts’ concerns: if improvement is perceived as “doubling every few months,” organizational adjustments could be abrupt. Not because the entire economy automates simultaneously, but because certain departments (support, operations, administration, content marketing, preliminary analysis) could undergo rapid transitions.
The emerging scenario: fewer “traditional” jobs, more hybrid roles
In the short term, the most realistic change isn’t a world without employment, but a market with more hybrid jobs: people doing the same work but with new tools, and different expectations.
- The support agent who previously handled 40 tickets a day now reviews 120 generated by an assistant.
- The administrative worker who previously drafted documents from scratch now validates drafts, checks data, and manages exceptions.
- The analyst who previously produced manual reports now focuses on interpreting, explaining, and making decisions.
In this scenario, the advantage won’t just be “using Artificial Intelligence,” but knowing how to use it thoughtfully: asking for the right things, detecting errors, safeguarding data, and understanding limits.
Frequently Asked Questions
What does it really mean that AI “doubles capabilities” every seven months?
It’s usually interpreted as rapid productivity improvement: less time to complete similar tasks, more automation, and less supervision needed for repetitive processes.
Which jobs are most vulnerable to automation cuts in 2026?
The most vulnerable tend to combine routine tasks, large volumes of similar cases, and clear rules: first-line customer service, repetitive administrative tasks, and parts of back-office work.
How can a worker reinvent themselves without switching sectors completely?
By specializing in processes that require supervision: quality control, review, compliance, security, “complex” customer service, or roles combining business knowledge with digital skills.
Does automation necessarily mean more unemployment, or can it also create jobs?
It can do both simultaneously: replacing tasks in some roles and creating new ones elsewhere. The key usually depends on the pace of change and the ability to train and resettle workers.