Global Observatory · AI & Labour
From experimental adoption to organisational restructuring: AI is transforming work
Editorial
In the two-month period of January-February 2026, the picture shifted from 'experimental' adoption to full-scale organisational restructuring. The clearest signals are three: rapid increase in corporate AI usage, emergence of new roles linked to AI agents, and the first highly explicit cases in which management linked workforce reductions to team redesign and automation. Institutions confirm an uncomfortable dynamic: AI is not merely replacing work, it is polarising it—with wage premiums for certain skills and mounting pressure on entry-level and more standardised roles.
Geographic area summary
🌐
GLOBALE
Large corporations
Block, Amazon, Klarna: the AI–cuts nexus becomes explicit
Block – 4,000 cuts and AI overhaul. At the end of February, Jack Dorsey announced over 4,000 redundancies—nearly half the workforce—as part of a restructuring explicitly linked to AI integration. It is one of the first cases in which the nexus between AI and structural workforce reduction has been declared frontally by top management: with 'intelligence tools', a much smaller team can do more and better.
Amazon – 16,000 corporate cuts. In January it completed a plan for roughly 30,000 exits begun in autumn. (Reuters) reports that Jassy attributed the latest round chiefly to bureaucracy and corporate culture, but coverage notes that the previous round had been linked to AI expansion. In many cases AI is not the sole cause, but it is now a central lever of restructuring alongside cost-cutting.
Klarna – the jobs that remain. In its February results it reiterated that early AI adoption helped reduce headcount, but specified that the positions destined to remain in an 'AI-powered' company will be increasingly concentrated on human relationships with customers and merchants. A concrete signal of the shift from transactional and repetitive roles to roles involving relationships, supervision and exception management.
🇺🇸
USA
AI Agents & New roles
Vacancies for AI agent skills +1,587%: the orchestration market is born
OpenAI launches enterprise services for AI agents. In February, OpenAI made available a service that allows companies to build and manage agents capable of performing specific tasks—for example, correcting software bugs. The implicit message: the 'next colleague' may be operational software integrated into workflows.
AI agent vacancies +1,587%. According to Randstad (cited by Reuters), job postings for 'AI agent' skills have exploded by 1,587% over the periods monitored. Whilst some roles are disappearing or being compressed, others are opening in integration, orchestration and control of agentic systems.
The 'AI intelligencer' is born. (Reuters) describes an emerging hybrid profile: someone capable of making AI agents work within corporate flows, debugging, translating business needs and governing deployment and reliability. The market is shifting from purely technical roles to bridging roles between model, process and decision.
Legal, tax, accounting professionals. (Thomson Reuters Institute) (February): 40% of professionals report that their organisation already uses GenAI (vs 22% the previous year); over 80% of current users use it weekly; 53% are planning or evaluating the use of agentic AI. It is a rewriting of workflows: less standard production, more review, validation and accountability.
🇪🇺
EU
Institutions & Research
OECD, IMF and ILO: unequal adoption, wage polarisation, managerial risks
OECD – AI adoption more than doubled in two years. At the end of January, the (OECD) reported that in 2025, 20.2% of firms in the countries surveyed used AI, against 14.2% in 2024 and 8.7% in 2023. Adoption, however, is highly unequal: large firms are ahead, small ones far behind. Productivity benefits and employment risks will not be distributed uniformly.
IMF – polarisation confirmed. Roughly 1 in 10 vacancies in advanced economies requires at least one 'new skill', often IT or AI. Where demand for AI skills grows, employment in the most vulnerable occupations is 3.6% lower after five years compared with less exposed areas. AI skills raise wages for some, but accentuate polarisation for others.
ILO – risk of algorithmic management. The (ILO) (February update) distinguishes between automation of tasks and automation of managerial functions: AI does not only replace those who execute, but can change how tasks are assigned, performance measured, work monitored and who stays or leaves decided. This raises issues of transparency, bias, organisational power and quality of work.
Anthropic vs Pentagon—ethics as an industrial variable. The clash between Anthropic and the US government over mass surveillance and autonomous weapons signals that the ethical choices of an AI supplier can affect contracts, employment and supply chains. Ethics is no longer ancillary: it has become an industrial and geopolitical variable.
Monthly deep-dive
Focus
The voices of the 'fathers' of AI: Hinton, Amodei, Hassabis
Three central figures in the AI landscape share convergent positions during the two-month period on employment risks and the urgency of governance
Geoffrey Hinton in January raised the tone further. In an interview reported by (Business Insider), he said he was 'very sad' about what AI has become, called it 'extremely dangerous' and warned of the risk of heavy job losses. His message remains clear: AI can bring benefits, but we are running faster than our ability to govern it.
The real mistake would be not to invest in research on how to coexist with systems more intelligent than ourselves. — Geoffrey Hinton, January 2026
Dario Amodei has maintained a hard line. At Davos he warned of possible job losses as model capabilities accelerate; a few days later he declared that AI will exceed human cognitive abilities in most tasks within a 'small number of years'. If this forecast were even only partially correct, the consequence for clerical work would be enormous.
Challenger Survey (Reuters): AI was linked to 7% of planned layoffs in the US in January 2026. In sectors most exposed to AI, wages are still rising for those bringing skills difficult to replace—but positions that disappear are increasing. The risk is not 'uniform collapse': it is harsh selection in the labour market.
Demis Hassabis added a complementary piece: in February he told (Bloomberg) that the risks of AI require urgent attention and international cooperation, distinguishing between malicious use by hostile actors and technical risks internal to increasingly autonomous systems. It is a less apocalyptic position than Hinton's, but it leads to the same operational point: more system autonomy means more need for governance, standards and clear human accountability.
Sources & references
01
02
03
04
05
06
07
08
09
10
11
12
Reuters — Block layoffs
Block: 4.000 tagli e overhaul IA — dichiarazione frontale di Jack Dorsey
https://www.reuters.com
Reuters — Amazon layoffs
Amazon: completamento piano da ~30.000 uscite e ruolo dell'IA nella ristrutturazione
https://www.reuters.com
Reuters — AI agent vacancies
Randstad: vacancy per competenze AI agent +1.587% nei job posting monitorati
https://www.reuters.com
Reuters — Amodei a Davos
Dario Amodei: perdite di posti di lavoro e superamento capacità cognitive umane
https://www.reuters.com
Reuters — Anthropic vs Pentagono
Scontro Anthropic–governo USA su sorveglianza di massa e armi autonome
https://www.reuters.com
OECD
Adozione IA nelle imprese: 8,7% (2023) → 14,2% (2024) → 20,2% (2025)
https://www.oecd.org
IMF
Polarizzazione occupazionale: skill IA, nuove vacancy e -3,6% occupazione in aree vulnerabili
https://www.imf.org
ILO
Algorithmic management: automazione delle funzioni manageriali e rischi di trasparenza e bias
https://www.ilo.org
Thomson Reuters Institute
GenAI nei settori legale, tax, accounting: 40% adozione, 53% valuta agentic AI
https://thomsonreuters.com/institute
Business Insider — Hinton
Geoffrey Hinton: 'extremely dangerous', perdite di posti di lavoro e urgenza di governance
https://www.businessinsider.com
Bloomberg — Hassabis
Demis Hassabis: rischi IA e necessità di cooperazione internazionale
https://www.bloomberg.com
OpenAI — Enterprise AI agents
Lancio del servizio enterprise per costruzione e gestione di AI agent aziendali
https://openai.com
Methodological note: Methodological note: Analysis based on public sources, sector reports and information materials available at the time of writing, with the support of a generative artificial intelligence system (Claude/ChatGPT), used as a tool to assist analysis and synthesis. Factual or interpretative inaccuracies are possible, particularly where information is incomplete, contested or under investigation. Interpretations, assessments and conclusions remain the sole responsibility of the author.