AI governance isn't a €500K hire
TL;DR
74% of organizations want to deploy agents while only 21% have a mature governance framework (Deloitte). Bridging that gap with a €500K CAIO misses where governance happens: inside the team, with the shadow AI that already exists. The work is to make those uses visible, equip them, and lay down a lightweight framework so they scale.
On paper, everyone agrees: AI in the workplace needs a framework. The numbers tell another story: 74% of organizations plan to deploy agents within two years, while only 21% have a mature agent governance model (Deloitte, State of AI 2026). Bpifrance sees the same gap from another angle: the six success factors for SMBs adopting AI — in-house training (89%), leadership (83%), clear governance (57%) — are all organizational; technology does not even appear on the list. Adoption has been running ahead of governance everywhere for long enough to look like a steady state rather than a delay.
The €500K-hire reflex
When something does not keep up, the natural answer is to hire someone whose job it is. A LinkedIn post from this summer drew reactions for exactly that reason: it staged an organization discovering the gap between its appetite for agents and its governance, whose first move was to open a €500K Chief AI Officer position to close it. The hire has everything going for it: visible, datable, accountable. It turns a governance problem into a recruiting problem, and a recruiting problem can be solved with a budget: it goes on an org chart, it gets celebrated in committee.
What people on the ground say
The sharpest answers came from the post’s comments. The counter-argument comes up again and again: colleagues already using an agent on the sly exist (the shadow AI ), and giving those people the means to scale costs less than a €500K expert — and pays more. Others point out that work-mode transformations fail in the majority of cases, more often through top-down decree than through a lack of expertise. Bpifrance’s numbers point the same way: the factors that correlate with success are practices carried by people, with training (89%) and leadership (83%) first, governance (57%) following, and no trace of technology on the list.
Four developers, one agent, one ambassador
I lived this debate on a smaller scale when I rolled out an agent (Kilo Code) to 4 developers; the full write-up is already published on this site. Adoption held because of an internal ambassador, someone the team already knew, and a lightweight framework laid down as we went. The mirror concession holds: ambassadors without a framework do not scale. Everyone tinkers with their own shadow AI, practices diverge, and the risks (data, debt, cost) stay invisible until they become expensive. The framework’s job is to make sure initiatives compound instead of cannibalizing each other.
Meanwhile, the issue has moved beyond team maturity: Article 4 of the EU AI Act has required training for people handling these systems since February 2025. I detailed what applies to an SMB in another dispatch. Governance now has to fit the compliance calendar.
Governance gets built inside the team, or it gets delegated to a position. The first version means finding the people already practising it on the sly, then giving them means: licenses, time to train, and a lightweight framework that turns underground use into an owned practice. The second costs €500K and sits on the org chart. On paper, they look alike. A year in, only one of them will have trained anyone.