Microsoft's 2026 Work Trend Index reports that active agents in Microsoft 365 grew 15 times year over year, and 18 times in large enterprises. A Lumenova roundup also cites an August 2025 Gartner prediction that 40% of enterprise applications would feature task-specific AI agents by 2026, up from under 5% in 2025. These are vendor and secondary-source figures, but the direction is not in doubt.
Why sprawl is a risk
- Nobody knows how many agents exist or what they can access.
- Agents outlive the people who created them.
- Permissions accumulate and are rarely reviewed.
- Duplicate agents give inconsistent answers.
Controls that work
- A registry: every agent has an owner, purpose, data access and expiry date.
- Standard patterns and templates for creating agents safely.
- Periodic review: unused agents are retired.
- Central logging, regardless of where the agent runs.
The first control is knowing what you have. Start with an inventory.