Adopted from Microsoft CAF. A clean three-way split that scales:
+----------------------+
| AI CoE | Standards, governance, registry, training,
| (central, small) | prioritization, audit, responsible-AI policy
+----------+-----------+
|
v
+----------------------+
| Platform Team | Identity, environments, networking, observability,
| (technical foundation)| shared infra, runtime enforcement, security baseline
+----------+-----------+
|
v
+----------------------+
| Workload Teams | Business requirements, domain data, prompt design,
| (in each department)| agent build, integration, adoption, KPIs
+----------+-----------+
|
v
+----------------------+
| End Users | Use agents inside the tools they already use
| (everyone) | (Teams, Slack, CRM, ticketing, email, portals)
+----------------------+
Responsibilities in one line each:
- AI CoE — Responsible for the framework; Accountable for the AI portfolio.
- Platform team — Accountable for the technical foundation, runtime enforcement, observability, identity.
- Workload team — Accountable for the agent's business value and KPI in their department.
- End user — Responsible for using agents correctly.
- Security / Legal / Compliance — Accountable for risk sign-off on Medium- and High-risk agents.