Agentic AI
Enterprise AI agents are racing ahead while governance frameworks lag behind
Most large organizations are deploying AI agents inside their applications, but few understand the sovereign AI frameworks needed to maintain control over them.
The New Stack
Enterprise adoption of AI agents is accelerating rapidly, yet the governance structures and architectural principles needed to keep them under control remain poorly understood across organizations. A new survey reveals a stark mismatch: while most large enterprises already embed AI agents in their business applications, only a fraction grasp the sovereignty and guardrail requirements these systems demand.
The adoption-governance gap#
An IDC survey of over 500 IT and business leaders at organizations exceeding $1 billion in annual revenue found that 86% already deploy AI agents embedded in applications. Yet only 12% report that their organization widely understands the risks and controls associated with sovereign AI deployment.
The research, commissioned by Canadian AI company Cohere, reveals an even more troubling fact: just 13% of respondents felt sovereign AI concepts were broadly understood across their organizations, and roughly one in three struggled to articulate what sovereign AI actually means in their own words.
The gap widens further when looking ahead. While 5% currently develop agents in-house, 44% expect to do so within the next 12 months. Third-party agent adoption is projected to jump from 13% to 67% over the same period.
What sovereignty means in practice#
The IDC survey defines sovereign AI as the ability to maintain genuine choice and control over design, development, deployment, accessibility, operation, maintenance, and governance of AI systems—plus the underlying infrastructure they run on.
Joelle Pineau, Cohere's chief AI officer, frames the issue clearly: "Sovereignty is fundamentally an architectural question. If AI is tied to a single provider or governed by terms that can change, you don't have control; you have dependency."
For enterprises, this translates into concrete requirements: the ability to run AI models and agentic platforms entirely on private infrastructure, with no external dependencies or automatic updates pushed from vendors. Cohere's model allows customers to deploy its complete stack—including models, retrieval-augmented generation, and agent tools—in fully air-gapped environments. Software updates arrive as signed packages that customers validate and deploy on their own schedule, rather than being imposed by an external provider.
