Design partner scope starts with control evidence, not autonomy promises.
AI2COE treats an Agentic AI pilot as a controlled evaluation path. The current public boundary is R0-R2: retrieve approved information, analyse evidence, and recommend for human review. No ERP, EAM, CMMS, procurement, inventory, or maintenance write-back is included.
Bounded use case
Define the industrial decision and prove an agent is needed before any pilot scope is accepted.
Approved data scope
Use synthetic, sample, or tenant-approved exported data only; no private records in the inquiry form.
Tenant isolation
Separate tenant context, credentials, logs, and evidence artifacts before customer-controlled testing.
Identity and authority
Assign a unique agent identity, scoped permissions, tool allowlist, and R0-R2 authority ceiling.
Named human owner
Require owner review for consequential findings, exceptions, approvals, and exit decisions.
Evaluation baseline
Use a reproducible baseline, source references, abstention checks, and prompt-injection review.