Owner: CoE Lead. Input: Risk appetite published. Sub-steps:
- Pick the CoE operating model (
framework.md§6.1): Centralized / Federated / Hybrid / Platform-led / Domain-focused. For a first AI program, start Centralized and plan to evolve to Hybrid. - Identify and recruit Department Champions in the departments most likely to need agents first (Sales, Finance, Ops, Support, etc.). Champions are part-time CoE liaisons inside their own department — they bring use cases in, support adoption, and own the agent's KPI.
- Identify the Platform Team counterpart (IT / DevOps / infrastructure) responsible for identity, environments, runtime enforcement.
- Document the responsibility split (
framework.md§7) in a one-page RACI: CoE / Platform / Workload / End User / Security-Legal-Compliance. Output / gate criteria: Names against every role on the RACI. Champions in at least 2 departments. Decision branches:
- No department wants to provide a Champion → start with the one department that does. Don't try to launch everywhere at once. Skip-this-step risk: The CoE becomes a black box that no department trusts; or every department invents its own AI process.