Section 35 of 102
34. Agent Activation and Commissioning
Stable section ID: S05-CON-010-SECTION-35 · 28 content blocks
The Building Operations AI Agent shall not become operational merely because its software has been installed or connected.
Activation shall occur through a controlled commissioning process that verifies:
- Agent identity and Manifest.
- Software and model version.
- Building assignment.
- Building BIOS compatibility.
- Digital Twin connectivity.
- Sensor and equipment connectivity.
- Tool registration.
- Permission assignments.
- Cybersecurity configuration.
- Data-handling requirements.
- Human oversight responsibilities.
- Alert and escalation routes.
- Manual override capability.
- Degraded and offline behavior.
- Audit and event recording.
- Update and rollback procedures.
Commissioning shall test the agent’s ability to distinguish between normal, abnormal, unknown, and unavailable conditions. It shall also verify that the agent reduces its authority appropriately when required information or services are missing.
Control actions should initially be tested through simulation, shadow operation, or recommendation-only mode. Autonomous action shall be enabled only for functions that have passed the applicable verification and authorization process.
Activation records shall identify:
- Who authorized activation.
- What capabilities were enabled.
- What limitations remain.
- Which systems were tested.
- What baseline information was used.
- When revalidation is required.
Material changes to the agent, building configuration, control interfaces, or safety environment may require partial or complete recommissioning.