Agent Integration Checklist to Close Identity, Runtime, and Governance Gaps

Vendor overlap exposes the integration seams

Broadcom, Citrix, CrowdStrike, ServiceNow, and Genesys each released nearly identical AI-agent infrastructure stacks during a two-week period from late August to early September 2026.

The overlap shows the same control points, but identity, runtime, and governance can still sit in different products. An agent can act while no owner can trace, limit, or stop it. The fix is a checklist, not another vendor stack.

In 2024, 60% of generative AI pilots were dropped, mainly because of a governance gap.

Enterprise AI platform operators should ask whether a stack can be wired into existing identity, runtime, and governance controls before choosing a vendor.

Identity decides who or what is acting. Runtime control sets what the agent can do and how fast. Governance determines whether anyone can see it, stop it, and explain it later.

The Model Context Protocol (MCP), launched by Anthropic and now under the Linux Foundation, has passed 97 million SDK downloads.

At VMware Explore 2026, Broadcom introduced AgentMinder inside its Private AI Cloud as a control point for agent identity and runtime behavior.

Citrix added an AI Gateway to NetScaler in April and an MCP Gateway in July, and included centralized authentication plus tool-based rate limiting in its existing platform license without a separate SKU.

At Fal.Con 2026, CrowdStrike launched Falcon Guardian, with a claimed 99% detection efficacy against prompt attacks and 100ms latency.

ServiceNow made its AI Control Tower generally available, emphasizing platform-native governance for finding, monitoring, and terminating rogue agents in real time; Genesys presented a four-product stack at Xperience 2026.

Read the vendor moves as a control inventory. Each product targets a different seam. When identity, runtime, and governance live in separate products, the integration work is the real project. A vendor stack can cover one seam well, but it rarely covers all three.

The operator's job is to connect the seams into a control path before the agent touches production. Ask each vendor to show the control path in a live environment: where the agent is bound, where the call is limited, and where the termination record lands. If a vendor can only show a demo, the seam is still open.

Keep the evidence in the ticket. The owner should be able to point to the decision, the log, and the termination record. If the evidence lives in a different console, the operator will spend the rollout chasing exports. The ticket should survive an audit.

Run the three-layer checklist before rollout

The checklist is the pre-rollout control path. Each item maps to a seam where an uncoordinated stack fails.

• Identity: Confirm each agent has a named owner and a scoped credential.
• Identity: Verify the credential cannot reach tools or data outside the agent's approved scope.
• Runtime: Confirm every agent call passes through a gateway that can log and block it.
• Runtime: Verify rate limiting and kill switches work against a live sandbox.
• Governance: Confirm a dashboard lists active agents, owners, and recent actions.
• Governance: Verify a termination test can stop a rogue agent and record the result.

Identity items protect the audit trail before the agent can act, runtime items keep the gateway in the call path, and governance items make the rogue process visible to the operator.

Run the identity checks before any tool access, test the runtime controls before any external call, and verify the governance controls before any agent runs unattended. Record the result in the ticket.

Treat missing controls as deployment blockers

If a vendor stack cannot answer every check, stop the rollout. The missing piece is often a control point that no one assigned. Start with the owner, then the gateway, then the dashboard. When all three are in place, the agent is ready for production.