AI agent growth is outpacing enterprise governance

Fortune 500 companies are expected to operate more than 150,000 AI agents by 2028, according to Gartner analysis cited by OutSystems' CIO. That would mark a 10,000-fold increase from 2025, when enterprises averaged just 15 agents — and it helps explain why governance is climbing to the top of the enterprise AI agenda.

The problem, the piece argues, is that adoption is moving faster than control. While nearly every global enterprise is already using AI agents, only 12% have introduced a centralized governance approach. For most organizations, governance remains fragmented, creating the same disorder it is supposed to prevent.

As companies roll out a mix of customer-built and pre-built agents, the risk of AI sprawl rises quickly. Employees are increasingly building their own agents, which can leave AI usage siloed and inconsistent across teams. According to the article, 94% of businesses are already seeing more complexity, technical debt and security risk as a result.

Centralized governance versus piecemeal controls

Many organizations are trying to respond with new compliance processes and safeguards. One widely used method is human-in-the-loop review for key decisions, now used by 52% of businesses. But the article warns that as AI scales, that approach can introduce its own inconsistencies when different teams, regions or departments apply different rules for access, security and usage.

The proposed fix is to build governance into the system from the start rather than bolt it on later. That means giving enterprises centralized visibility into:

Recommended reading

Europe’s chip push won’t end US cloud reliance

• which agents are running
• how they are connected
• where their dependencies sit

The article argues that manual governance cannot keep up with the speed at which agents interact with data and with one another. Instead, businesses need a central neutral system layer that orchestrates agents against a real-time view of enterprise architecture and operational constraints.

That, in turn, would let all agents follow the same rules, comply with the same standards and pull from shared knowledge — without slowing adoption. The core challenge, the author argues, is no longer model capability, but connecting AI to where real work happens inside the enterprise.