Shadow AI

Menlo Ventures' third annual report on generative AI in the enterprise, dated 2025-12-09, puts Shadow AI at close to 40% of application AI spending: employees paying for AI tools with their personal cards, inside an application layer worth $19 billion.

Ethan Mollick gave the behavioural reason at Valence's July 2025 summit: "People are turning to AI all the time as a coach and to help with work, but they're just not telling management about it." Employees stay quiet because they fear what revealing their productivity gains would do to their jobs. His Leader-Lab-Crowd framework treats that silent majority as the Crowd, and makes executive AI fluency the motivational lever: it legitimizes usage and brings people out of the shadows.

Olivier Rafal had already registered the governance side in February 2024, writing on CIO-Online that the democratization of GenAI produced "tant du shadow AI que de fortes attentes vis-à-vis de la DSI". Shadow AI arrives as a by-product of access, not as a policy failure anyone decided.

Agentic production run outside any frame moves the stakes: it fragments the SI and leaves reconciliation impossible. Matthieu Grymonprez and Romain Taillade carry a precedent already paid for, the self-service BI that produced more than 150 000 dashboards before the return to certified data products, and Taillade's agentic version of the same curve, the thirtieth Stripe integration: « ton SI peut s'écrouler sur son propre poids ».