Tech • AI • Robotics • Game
Should AI be allowed to decide for us? A former McKinsey consultant responds
7/10TL;DR
French companies broadly want to adopt AI, but the decisive issue is no longer whether to use it: it is how to deploy it without losing skills, widening the SME gap, or undermining democratic control over key decisions.
KEY POINTS
Digital transformation was already slow and heavy
Large companies have spent roughly 15 to 25 years adapting to digital technology, often through major programs lasting 18 to 24 months. In sectors such as banking, dedicated digital teams can reach 600 to 700 people, reflecting how difficult it is to move a large organization. The core challenge has been constant adaptation because technology keeps shifting.
AI adds urgency, fatigue and fascination
For many executives, AI is not a sudden novelty but the latest wave after earlier digital investments, cloud migration and process redesign. Since 2022, generative AI has accelerated pressure because it works directly with language and knowledge, making its impact more visible across all functions. That has produced both excitement and a sense of exhaustion inside big organizations already deep into transformation.
Adoption is now a near-universal goal
Across large, mid-sized and smaller businesses, there is strong demand to integrate AI. The main obstacle is not willingness but execution: identifying useful cases, securing confidential data, building governance, and managing cybersecurity. In research-heavy sectors such as pharma or molecular design, deployment can require specialized models and controlled access to sensitive information.
France risks a widening divide between large firms and SMEs
Large French companies are adopting AI at a pace described as close to that of the United States, but SMEs are moving more slowly. That creates a growing digital divide inside the economy, with another gap between northern and southern Europe. Smaller firms often lack capital, in-house expertise and time to organize serious workforce retraining.
No job apocalypse, but a deep disruption of work
The likely effect of AI is not mass unemployment overnight but a reconfiguration of tasks over 10 to 15 years. Estimates cited put annual productivity gains at 0.7 to 1 point for a decade, while roughly 30% of hours worked could be automated over time. Most jobs will not disappear entirely, but many will have 10% to 15% of tasks automated, forcing employees to work differently.
Training is the central economic choice
Companies effectively face two paths: use AI mainly to cut payroll, or use it to raise the capability of existing staff. The stronger long-term model is described as pro-worker AI, where tools augment employees rather than simply replace them. That approach preserves hard-won operational knowledge and can produce more growth than a narrow cost-cutting strategy.
AI may democratize expertise
A major shift could come from widening access to high-level capability. With AI copilots, workers with less formal expertise may handle parts of more advanced roles, such as a nurse taking on tasks closer to those of a doctor. That does not eliminate senior professions, but it does require organizations to redesign roles, workflows and responsibility.
Training systems are not ready
The current training market largely teaches yesterday’s skills, while AI tools and practices change within months. Proposed remedies include public incentives for SME training, including training vouchers and tax credits, plus stronger regional coordination and involvement from vocational schools. Large firms may increasingly build their own internal academies, as major tech groups already retrain staff on a recurring cycle.
Human judgment still has red lines
A framework emerges around three zones: decisions AI should largely handle, decisions shared between humans and AI, and decisions that should remain human. Logistics optimization, fraud detection and parts of financial trading fit the first category. Recruitment and strategy are more hybrid. Criminal justice belongs in the protected zone because democratic societies still require one human to judge another.
Performance alone cannot settle legitimacy
Evidence from medicine illustrates the problem. Top doctors may diagnose correctly in about 74% of cases, AI alone in around 90%, yet doctor-AI combinations can still underperform if experts ignore the system. That makes the question political as much as technical: institutions need explicit rules on when AI informs decisions, when it leads, and when it must be excluded.
Citizen debate is becoming a governance necessity
The emerging risk is cognitive resignation: either delegating too much thought to machines or rejecting AI entirely. To avoid both extremes, AI deployment in sensitive areas such as education, work and public services needs public deliberation, not just management decisions or vendor promises. Without that, resistance is likely to harden even where AI could bring clear gains.
CONCLUSION
The central issue is no longer whether AI will enter companies and institutions, but whether leaders can combine productivity, retraining and legitimacy fast enough to keep control of the transition. The winners are likely to be those that treat AI as a tool for capability building rather than a shortcut to headcount reduction.
Ask a question