Reading time: 8 minutes
What the 2025 MIT and Yale studies reveal about the real impact of artificial intelligence on work and businesses.
Sources of analysis inspiration Reports Yale Lab Budget and MIT Project NANDA
For the past three years, artificial generative intelligence has been at the centre of the global economic debate. Saluted as the biggest technological break since the Internet, it arouses both enthusiasm and anxiety.
Between promise of unprecedented productivity and fear of machine substitution of man, discourses often clash faster than the facts confirm.
Two major studies published in 2025: MIT Project NANDA on the adoption of the GenAI in company and the Yale Lab Budget on its impact on employment, paint a much more nuanced picture. Far from the cataclysm announced, they reveal a contrasting reality: an economy in the learning phase, where the AI slowly transforms organizations without further disrupting the overall structure of work.
Three years after ChatGPT: the mirage of the instant revolution
Since November 2022, the planet has been packed around artificial generative intelligence. The promises of productivity, increased creativity and transformation of the professions have suggested a change comparable to that of the Internet.
But reality in 2025 is much more nuanced.
Two major studies: Yale Lab Budget and MIT Project NANDA come to restore a healthy balance between the fear of massive job replacement and the sometimes disproportionate enthusiasm of the « GenAI economy ».
Their common observation: The revolution of the AI is well under way, but it remains slow, selective and unevenly distributed.
The labour market: continuity more than rupture
Yale in the United States (from 2022 to 2025).
Outcome: no sign of structural shock.
| Indicator | Observed result | Interpretation |
|---|---|---|
| Change of professional mix | +1 % compared to the Internet period (1996-2002) | No significant rupture |
| Average AI exposure | 29% stable (low), 46% (average), 18% (strong) | No massive job displacement |
| Unemployment in exposed occupations | Constant (25-35 % automated tasks) | No visible effect on unemployment |
| Most dynamic sectors | Information, finance, professional services | Changes initiated before AI |
Clear The distribution of AI has not yet radically changed the structure of employment.
It acts as a bottom blade, slow, but continues.
« IA does not destroy jobs, it redraws their contours. » The Budget Lab, Yale, 2025
The illusion of massive replacement
Alarmist projections predicted the disappearance of 300 million jobs.
However, the Yale show that automated tasks represent only 11% of actual observed activities, against 70 % of tasks increased by AI.
In other words: human remains at the heart of the productive process, but with new tools.
MIT: a disturbing divide in companies
For its part, the MIT Project NANDA analysis of the adoption of GenAI in more than 300 international companies.
His observation is without appeal: the majority fails to transform the test.
The « GenAI Divide » : the new digital divide
Only 5 % companies get a measurable return on their investment in AI,
while 95 % remain blocked in the pilot phase or leave no tangible ROI.
| Observed factor | Result | Consequences |
|---|---|---|
| IA project success rate | 5 % | Concentration of value on a minority of actors |
| Functions targeted by IA budgets | 53% on marketing and sales | Poor targeting of productivity deposits |
| Highest ROI | Finance, support, internal operations | Efficient companies invest where AI automates and learns |
| Real use of AI by employees | 90% via personal tools (« shadow AI ») | Fracture between individual initiatives and corporate strategies |
Thus a Double fracture :
- Between companies that industrialise their AI solutions and those that stagnate.
- Between employees already using AI and their still hesitant organisations.
Two parallel worlds: macro stability, micro turbulence
The results of the two reports, while distinct, are:
| Analysis scale | Yale Outcome | MIT outcome | Synthetic reading |
|---|---|---|---|
| Macroeconomy (overall employment) | Stability, continuity | – | IIA diffuses slowly, without shock of employment |
| Microeconomy (enterprise) | – | Partial closure, concentrated ROI | AIA transforms pre-employment organisation |
| Time | Development over several decades | Break in 12-18 months in the most agile companies | Double adoption speed |
| Critical factor | Training and mobility | Integration into workflows | Human competence remains the pivot |
Analysis: two speeds, one heading
- Yale A stable labour market shows that the AI wave has not yet reached its full macroeconomic effect.
- MIT identifies an elite of enterprises (the « GenAI Crossers ») capable of fully exploiting technology and creating measurable gains.
These two temporalities coexist: The world of work remains on hold while pioneering companies reinvent their value chains.
The strategic turning point of the next 18 months
The authors of MIT The Commission considers that 18 months to come will be decisive.
Organisations able to integrate AI into their business processes (invoicing, auditing, document management, financial reporting, HR, etc.) will take action. an irreversible structural advance.
This is where the two perspectives come together:
- For Yale, the real impact of AI on employment will appear when companies have industrialised their cases of use.
- For MIT, this phase is precisely the one that 95 % companies have not yet crossed.
The "Crossers" of GenAI Divide
After MIT, companies that succeed in converting their AI initiatives into value share five characteristics:
- Native integration The AI is integrated into daily tools (CRM, ERP, accounting software, etc.).
- Data Pilot Each case of use is based on measurable economic indicators.
- Learning system approach : algorithms are enriched by feedback from users.
- Smart Targeting : priority to back office and support functions, real ROI deposits.
- Agile governance IA committees, rapid validation cycles, culture of experimentation.
For numbers and management professionals
The lessons of these studies are directly applicable to the world of consulting, accounting and auditing :
- IA does not replace expertise; it increases analysis and reporting capacity.
- Firms must position themselves as IA value integrators : validation of use cases, measurement of ROI, accompanying transformation.
- Training becomes a strategic lever: understanding models, biases, metrics, risks.
- IA tools (automated invoice analysis, bank reconciliation, predictive audit) must be connected to real processes not treated as isolated gadgets.
"The future belongs to those who can transform the promise of the AI into measurable organizational architecture. " — MIT Project NANDA, 2025.
The era of technological realism
Recent announcementsAmazon, eliminating 30,000 mainly qualified posts, mark a point of inflection in the trajectory that the ratios of MIT Project NANDA and Yale Lab Budget had so far described with relative optimism.
While these studies anticipated a phase of reallocation of powers and productive complementarity between humans and AI, industrial reality now seems to reveal a turning towards substitution.
What we observe, through these spectacular decisions, is the materialization of a new economic arbitration: Massive investment in artificial intelligence becomes the main lever of a cost-cutting strategy, even at the cost of a social shock.
Amazon, as Microsoft before him, not just experimenting with AI; it structurally integrates into its value chain, de facto redesigning the mapping of digital work.
This shift does not invalidate previous analyses; it rather highlights the time limit. The promise of a human-increasing AI already seems to enter a phase of brutal economic reality, where growth continues, but unemployed, and sometimes even against employment.
The tech companies thus inaugurate a algorithmic efficiency capitalism, where competitiveness is measured at the speed of automation.
In this sense, productivity becomes an end in itself, and the human question (training, conversion, inclusion) is relegated to the background, entrusted to the responsibility of States or to civil society.
In the face of this turning point, academic institutions such as MIT or Yale This will probably require a revision of their analytical framework to incorporate this new dialectic of progress : a progress that is now ambivalent, capable both of creating new opportunities and of Redefining the power relationship between labour and capital on a global scale.
What this announcement finally revealsAmazon, is that the revolution of the AI is no longer forward-looking; it has entered the phase of actual arbitrations, where economic models change faster than public policies, and where society must invent new forms of balance between innovation and social justice.
