AI at work: employees are moving faster than firms

In France, generative AI has taken hold in everyday work faster than organizations have managed to put a framework around it. According to Great Place To Work's Great Insights 2026 study, 59% of employees already use generative AI in their work, a share that rises to 71% among managers and 74% among those under 35. Yet this adoption remains largely informal: many employees experiment on their own, with their own tools and prompts, often out of sight of management. A second study, run by OpinionWay for HR software vendor Factorial among 1,028 French office workers, confirms and extends the picture: the scale of the shift has surprised employees themselves as much as their employers.

Adoption led by the youngest workers

The OpinionWay-Factorial survey, conducted from January 21 to 27, 2026, finds that for 42% of office workers, the extent of their current AI use would have been “unimaginable” just a year ago. The shift is especially sharp among younger workers: 60% of those under 35 did not anticipate such a heavy presence of these tools in their daily work, and 45% of them now delegate a major share of their work to the machine. For this generation, AI is no longer optional but a crutch: 45% of under-35s say they would be “much less effective” without it, compared with just 24% of all employees surveyed.

A governance gap that is raising alarm

This individual pace is running well ahead of companies' own. According to Great Place To Work, 43% of employees say they get no support at all with AI, and where a framework does exist, it remains minimal: only 25% have access to company-approved tools, 18% to training and 14% to ethics awareness sessions. The OpinionWay study goes further: 56% of office workers say their company leaves them to “figure it out on their own,” with no rules or clear framework — a feeling of abandonment that is even stronger in larger organizations, reaching 61% at mid-sized firms and 58% at large corporations. That vacuum exposes companies to real risks of data leaks and unchecked use, a phenomenon experts call “shadow AI.”

Companies are now running to catch up with shadow AI, trying to regain control of something that is already happening.

• 26% of employees — and 47% of under-35s — present AI-generated content largely as their own work
• 22% have already felt embarrassed or ashamed to be praised for work actually done by the machine, a figure that nearly doubles among younger workers (41%)
• 23% no longer know whether an idea truly came from them or from the AI
• 41% report a seesaw effect: time saved on tedious tasks is lost again to checking AI output or laborious prompting
• 24% admit spending too much time polishing AI answers instead of actually moving their work forward

A few companies are starting to build structure

Faced with this gap, some organizations are trying to regain control. At Deloitte, management has rolled out an internal tool, “Deloitte Chat,” built on top of market foundation models but fitted with specific guardrails, alongside specialized AI tools by practice area: legal, financial audit, interview transcription. A global partnership signed with Anthropic is set to roll out Claude progressively to the group's employees. The firm says it has trained more than 90% of its staff in France on a responsible-use charter, and makes managers the first line of oversight for their teams' AI practices. For Mick Levy, the real challenge is not technological but organizational: building governance, redesigning processes end to end, and deciding what gets automated — or risk seeing the time AI saves dissolve into a “new presenteeism” made of extra emails and filler content.

What this changes for companies worldwide

France is not a special case: the same gap between individual AI use and corporate governance is showing up wherever generative AI has reached the workplace, from the United States to the United Kingdom and beyond. For any company, this is no longer an HR footnote — it is an immediate operational risk. Without a clear internal policy — approved tools, training, guardrails around confidential data — shadow AI settles in by default, with concrete consequences: sensitive information leaking into consumer-grade services, decisions made on unverified content, or internal tension over who gets credit for genuinely human work. Companies that put a framework in place early — vetted tools, ongoing training, managers actively involved, as Deloitte has done — turn scattered individual use into a controlled, collective productivity gain. That applies to a five-person startup as much as a multinational, at a moment when structuring enterprise AI use is becoming a priority project, just as agentic AI moves from producing text to executing entire tasks on its own.