After massive adoption, record investments and high expectations, artificial intelligence is entering a strategic evaluation phase.
Artificial intelligence has established itself as the central driver of digital transformation. Prospective studies, strategic announcements and spectacular fundraising feed an ambitious story: increased productivity, accelerated innovation, decisive competitive advantage.
In certain sectors such as health, energy or marketing, the results are tangible. In others, feedback appears more measured. Although uses are growing rapidly, the quantifiable benefits do not always keep pace with expectations.
It is not a question of calling into question the current dynamic because AI structures strategies, mobilizes investments and has dominated economic debates for almost three years, but any profound transformation requires time to produce consolidated effects.
Two angles of analysis shed light on this discrepancy: that of developers, at the heart of software production, and that of managers, guarantors of investments and return on capital invested.
Focus Tech: AI in software development, massive adoption but limited gains
Several recent surveys, notably the Stack Overflow Developer Survey 2025 (49,000 respondents in 177 countries), as well as the work of the METR research laboratory, provide a contrasting observation.
- Real, but modest, productivity gains
According to a Bain & Company survey cited in these analyses, the improvement in developer productivity linked to AI would be between 10% and 15%. Measurable progress, but still very far from the expectations sometimes expressed around a “quantum leap” in performance.
- An AI that is sometimes counterproductive
The work of the METR laboratory shows that, in certain contexts – notably on open source projects involving experienced developers – the use of AI tools can lead to 19% more development time compared to work without AI assistance.
This result highlights that, in complex and demanding technical environments, AI does not guarantee an automatic gain in productivity.
- A constant need for verification
The developers interviewed report the need to correct errors or “hallucinations” produced by the models. The generated code must be reread, tested, adjusted. These additional steps mechanically reduce the expected acceleration effect.
AI assistance does not eliminate the control phase, but transforms it.
Surveys also indicate a decline in developers’ level of trust in AI tools compared to previous years, despite majority adoption. Satisfaction does not progress at the same rate as usage.
- Limits on complex projects
AI tools are less efficient for the production of large or technically complex programs. Their effectiveness appears more marked on limited tasks than on complete architectures.
- A marked contrast between use and trust
Despite these reservations, 84% of developers use or plan to use AI in their development process, and 51% of professional developers say they use it daily.
The paradox is clear: adoption is massive, but trust and satisfaction remain measured.
Focus Investments: financial benefits still limited
On the management side, the results published by PwC in its Global CEO Survey 2026 confirm this gap between expectations and reality.
- A majority of companies without measurable financial impact
In fact, 56% of executives say their AI investments have generated neither increased revenue nor reduced costs to date.
In France, 81% of companies indicate that they have not seen any impact on their turnover linked to AI.
- Tangible gains still in the minority
Only 12% of executives say they have simultaneously increased revenue and reduced costs thanks to their AI initiatives.
In this context, only 30% of global CEOs say they are confident about their revenue growth for the coming year — the lowest level observed in five years.
- A gap between investments and short-term benefits
The study highlights a significant differential between the scale of investments made in AI and measurable short-term financial results.
A phase of adjustment more than disillusionment
Data converges; AI is widely adopted, intensely tested and strategically funded, but its real benefits, whether operational or financial, today appear more progressive than disruptive.
This is neither disavowal nor immediate revolution. The available surveys rather describe a maturation phase which is lengthening with uses becoming structured, expectations being recalibrated and performance indicators now being confronted with the reality of business environments.
As the ecosystem stabilizes, the question no longer seems to be about the potential of AI, but about its concrete and operational conditions for effectiveness.
References:
- https://survey.stackoverflow.co/2025/
- https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/
- https://www.pwc.fr/fr/publications/series/global-ceo-survey.html
- https://trustmyscience.com/ia-developpement-logiciel-resultats-ne-suivent-pas-battage-mediatique-selon-enquete/
