Agent tool calls in enterprise sessions have grown 14x since February. Every employee now runs a swarm of concurrent agents across browser, endpoint, and cloud, each with its own identity, tools, and data access. Most have no registry, no policy, and no audit trail. Security teams answer with a dozen point tools: MCP proxies, prompt firewalls, AI-DSPM, NHI vaults, none of which follow the agent past a single surface. SASE sees traffic. Identity sees logins. Both are blind to what the agent actually did. Island governs the workspace itself, brokering every credential, tool call, and result through one control plane.

Key Takeaways

• Agents, not employees, are now the unit of work, and most run with no registry, no policy, and no audit trail
• Point tools each see one slice of agent activity. None follow the agent across browser, endpoint, and cloud
• The controls enterprises built for humans apply one to one to agents, if something sits in the middle to enforce them