Written by Julien Ricciarelli-Bonnal

26 July 2026

Companies Are Already Cutting Their AI Teams

For the past two years, announcements about artificial intelligence have followed an almost predictable pattern. Companies created specialist units, recruited researchers, opened laboratories and presented each investment as a decisive step towards an unavoidable transformation. The question was no longer whether they should invest, but how quickly they needed to move in order to avoid falling behind.

Amazon has now sent a different signal. The company has eliminated several positions within its organisation dedicated to artificial general intelligence, without disclosing the total number of roles affected. Amazon continues to describe advanced models as one of its strategic priorities, but says it wants to focus resources on the initiatives delivering the greatest value to customers.

This decision does not mean that Amazon is abandoning artificial intelligence. The group continues to develop its Nova models, expand generative AI services within AWS and support companies deploying the technology. It does, however, reveal an important shift: even organisations at the centre of the AI race are beginning to select more rigorously which programmes, teams and ambitions they are still prepared to finance.

The phase of indiscriminate expansion may be coming to an end. After accumulating projects to secure a position in the market, companies are gradually entering a period in which artificial intelligence must justify its costs, headcount and actual contribution to the business.

Artificial Intelligence Is No Longer Exempt From Trade-Offs

Reducing an AI team may appear paradoxical at a time when investment in the technology remains extraordinarily high. The contradiction disappears once overall spending is separated from the way resources are allocated. A company may increase expenditure on infrastructure, data centres or commercial products while reducing selected research teams, exploratory programmes or functions that are no longer considered a priority.

That is what Amazon’s decision appears to suggest. The eliminated positions belong to an organisation working on advanced models, while the company says it intends to accelerate initiatives that matter most to customers. The technology remains strategic, but not every possible route for developing it is still being treated as equally valuable.

This development reflects a rule that the enthusiasm of recent years sometimes obscured: a strategic priority is never exempt from arbitration. The more capital, talent and attention a field attracts, the more carefully companies must decide which capabilities to retain, which products to support and which programmes to stop. Early abundance makes it possible to explore several directions; maturity eventually forces choices.

The movement extends beyond Amazon. Several large technology companies have reduced headcount while directing a growing share of their investment towards AI, infrastructure and automation. In many cases, restructuring has been accompanied by a reallocation of resources towards a smaller number of technological priorities.

It would nevertheless be too simplistic to interpret every job reduction as evidence that artificial intelligence has failed. Some cuts still correct the aggressive hiring carried out during the strongest years of digital growth. Others reflect organisational redesign, the removal of management layers or a concentration of spending around fewer products. The real novelty lies elsewhere: the label “AI” no longer protects a team automatically from restructuring.

Exploratory Projects Must Now Demonstrate Their Usefulness

During the first phase of adoption, many companies launched AI initiatives without demanding a clearly measurable return. That tolerance was partly rational. When a technology develops rapidly, waiting until every use case is known before beginning to explore it creates the risk of arriving too late. Internal laboratories, pilots and cross-functional teams allow organisations to build knowledge before commercial outcomes are fully visible.

That logic has limits. An experiment that remains permanently disconnected from operations eventually becomes a cost centre that is difficult to defend. Leaders must be able to explain how a research project, model or prototype improves a product, reduces an expense, accelerates a decision or creates a new source of revenue.

Amazon’s announced refocusing reflects that pressure. The company is not questioning the importance of advanced models, but it is moving its efforts closer to initiatives that matter to customers. The wording may sound conventional, yet it describes a profound shift: the value of an AI team will no longer be judged solely by the sophistication of its work, but also by its ability to turn that work into usable services.

This change is likely to affect teams whose mission remains too broad. “Working on AI”, “preparing for the future” or “exploring agents” is no longer a sufficient roadmap. An organisation must specify the problems it intends to solve, the users concerned, the resources required and the criteria that will determine whether an experiment deserves to continue.

Artificial intelligence is therefore ceasing to operate as a standalone programme and becoming a component of corporate strategy. That is a healthy development, provided it does not lead companies to demand immediate returns from every research effort. Some innovations require several years before they produce commercial results. The challenge is to distinguish patient exploration from a project that cannot articulate its own usefulness.

The Competition for AI Talent Is Becoming More Selective

The first years of generative AI triggered an extraordinary competition for researchers, engineers, data specialists and product leaders. Companies sometimes built teams before they had clearly defined where those teams would sit within the organisation. The scarcity of expertise was enough to justify recruitment.

That dynamic is evolving. The most sought-after profiles have not disappeared, but employers increasingly want teams capable of connecting technology with operational needs. Pure technical expertise remains essential, yet it must be combined with an understanding of business processes, regulatory constraints, available data and user expectations.

The departure of several senior figures from Amazon’s AGI organisation, followed by a restructuring under leadership covering other advanced technologies, suggests that the change is not limited to a handful of positions. It also concerns the way high-level expertise is grouped, managed and connected to the rest of the company.

This selectivity creates an apparently contradictory situation. Companies may reduce some teams while continuing to recruit aggressively in other areas. They eliminate functions that have become peripheral but still seek specialists capable of integrating models into products, organising data, securing deployments or supporting customers. A reduction in one unit therefore does not necessarily mean an overall decline in AI capability.

Not everyone will simply be moved into another role.

The market is entering a phase in which general familiarity with AI tools is no longer enough. Companies will need fewer people whose main role is to promote adoption internally and more professionals capable of turning a use case into a reliable, adopted and economically defensible system. Enthusiasm will remain useful, but it will no longer compensate for weak execution.

Artificial Intelligence Has Also Become a Convenient Restructuring Narrative

The growing number of announcements linking job cuts, automation and AI investment requires caution. Not every position described as affected by artificial intelligence has been directly replaced by a machine. The same language can also cover more conventional decisions involving cost reduction, correction of overstaffing or organisational simplification.

Companies now have access to a particularly powerful narrative. Describing an organisation as “leaner”, “more autonomous” or “AI-first” makes it possible to present a reduction in headcount as a forward-looking transformation rather than a defensive measure. The explanation may be sincere, but it can also simplify the causes of the decision.

The broader technology sector already presents a more complicated picture than a uniform replacement of human labour. Major groups continue to spend enormous sums on AI infrastructure while eliminating thousands of jobs. Some reductions correct the hiring conducted during the pandemic, while others reflect a transfer of capital towards data centres, chips and models.

Artificial intelligence is therefore reshaping organisations before it fully replaces all the tasks associated with it. It changes investment priorities, reduces tolerance for certain management layers, favours smaller teams and forces every function to explain its contribution. Its effect on employment comes as much from this reconfiguration as from the complete automation of a profession.

For employees and leaders alike, the distinction matters. A company that attributes every restructuring decision to AI risks hiding its own recruitment or strategic mistakes. Conversely, denying any technological influence would make it impossible to understand why some functions are being redesigned, merged or made smaller.

Companies Must Move From a Portfolio of Projects to a Strategy

The main lesson from this phase is not that companies invested too much in artificial intelligence. It is that they sometimes invested without prioritising their ambitions clearly enough. They launched assistants, prototypes, laboratories, training programmes and experiments without always deciding which initiatives should become products, which should remain exploratory and which should be discontinued.

A more mature strategy begins by distinguishing between three categories. Operational use cases should deliver a measurable benefit within a reasonable period. Structural investments, such as data governance or infrastructure, should strengthen several activities over time. Exploratory projects should remain limited, openly acknowledged as uncertain and reviewed at predefined stages.

This approach prevents two opposite mistakes. The first is continuing an initiative for too long simply because it carries the AI label. The second is abandoning all research as soon as it fails to generate immediate revenue. Companies need a balanced portfolio, but they also need rules that allow resources to move without turning every revision into a crisis.

That discipline also requires organisations to establish clear governance for their use of artificial intelligence⁠, including responsibilities, evaluation criteria, acceptable risks and the conditions for moving from experimentation to deployment. An AI team cannot remain permanently isolated from operational, financial, legal and commercial leadership.

Amazon’s job reductions therefore do not represent a broad retreat from artificial intelligence. They illustrate the transition from a period in which participation mattered almost as much as results to one in which every team will have to justify its role in a more precisely defined value chain.

Maturity Begins When No Initiative Is Considered Untouchable

The AI sector operated for a long time as though the speed of investment guaranteed future advantage. Recruiting more people, launching more models and multiplying demonstrations made it possible to display an ambition that was difficult to challenge. That strategy helped companies learn quickly, but it also created structures whose objectives sometimes became too broad or too detached from real needs.

The decisions now emerging do not mean that the earlier enthusiasm was entirely misplaced. They show that the technology is entering a more demanding phase. Companies can no longer simply announce that they are investing in artificial intelligence. They must explain what they are investing in, for which users and with what expectation of value.

Some teams will disappear. Others will be reduced, merged or moved closer to products. New skills will continue to be recruited at the same time. The movement will be neither a general retreat nor a uniform expansion, but a continuous reorganisation around initiatives capable of demonstrating their relevance.

Artificial intelligence is becoming a normal technology at the same moment that it remains exceptional in the scale of investment it attracts. It is entering budgets, organisational charts and strategic trade-offs like any other priority. That may be less spectacular than the promise of universal transformation, but it says far more about the technology’s true level of maturity.

Companies are not beginning to reduce their AI teams because they have stopped believing in artificial intelligence. They are doing so because they are finally beginning to decide what they actually want to achieve with it.

👉 Accumulating projects does not amount to a strategy. Ricciarelli Partners helps organisations structure their use cases, responsibilities and priorities through an effective AI Business Governance framework⁠.