Established Ecosystems

Introducing a governance framework into an established environment—where agents are already deployed in diversely managed ways—requires a strategy that prioritizes immediate value and zero disruption.

The Mandrel Project recommends a four-phase “Governance Overlay” approach to bridge the gap between “Shadow AI” and a fully certified, high-assurance mesh.

Phase 1: Passive Discovery

Goal: Map the mesh without changing any code.

Start by drafting Mandrel Specs for your existing agents. Use the Mandrel CLI to validate these specs against your current operational knowledge.

• Action: Integrate mandrel-cli validate into your CI/CD pipelines as a non-blocking lint step.
• Value: Creates the first-ever unified registry of agent capabilities, model usage, and risk surfaces across the organization.

Phase 2: Advisory Observability

Goal: Quantify risk and spend using Layer 7 intelligence.

Deploy the Collet in Advisory Mode. In this mode, the Collet intercepts traffic but never blocks requests.

• Action: Update agent configuration to route tool calls or delegations through the Collet. Use risk-profile: balanced with all matrices set to advisory.
• Value: Real-time visibility into budget overruns, intent drift, and unauthorized tool calls via OpenTelemetry logs, without risking production downtime.

Phase 3: Targeted Enforcement

Goal: Migrate high-risk agents to full structural “Clamping”.

Once you have quantified the risk, pivot the most critical agents (e.g., those handling financial transactions or PII) to full enforcement.

• Action: Switch specific governance matrices from advisory to strict or challenge. Transition to the Sidecar Pattern for these workloads.
• Value: Deterministic risk mitigation where it matters most.

Phase 4: Unified Mesh Conformance

Goal: Standardize governance across all deployment platforms.

Leverage the Gantry Operator and platform-native generators to unify your diverse agent deployments under a single governance control plane.

• Action: Activate mandatory Compound Identity (Double Badge) verification and automated Metrology Lab promotion.
• Value: Complete operational consistency and “Single Pane of Glass” governance for the entire AI estate.

Migration Best Practices