Human approval
AWAITING HUMANWorkflow: Launch announcement · DE + JA
Requested by
Iris Meridian
Agentic operator
Approver
Amy L.
Brand Lead
DE locale claim triggers legal review per policy P-2 before CMS writeback.
Sovereign AI · Brand Operator Agent
Iris Meridian.
Iris Meridian is the agentic operator on top of GW Slate™. She interprets business requests, coordinates governed workflows, explains routing and policy, surfaces human approval decisions, and returns graded outcomes with receipts.
Iris Meridian · operator
session 9af2…Request
Localize next week's launch announcement for DE and JA and queue it for CMS.
Iris plan
- Retrieve approved EN source from CMS draft.
- Route DE + JA through local model (policy: local-only).
- Gate DE through legal review (policy P-2).
- Queue to CMS in dry-run for Brand Lead approval.
I will not write to CMS until Amy approves the diff.
Answer
The operator layer for governed agentic work.
Iris Meridian is GlobalizeWe's enterprise-facing agentic operator. She interprets business requests, coordinates governed workflows through GW Slate, explains routing and outcome grades, and gives humans a clear decision surface before AI-assisted work moves forward.
What Iris does
- Interprets the request.
- Plans the governed run.
- Explains routing and policy.
- Surfaces the human decision.
- Reports the outcome grade.
- Hands back a receipt.
Why an operator
A model alone is not an enterprise operator.
Models don't plan governed runs.
They generate. Iris plans — including which policy applies, which route is allowed, and which human must approve.
Models don't explain themselves to a CISO.
Iris narrates the routing, the policy bind, and the grade.
Models don't hold a decision surface.
Iris pauses, asks the right approver, and resumes only after a clean decision.
Models don't issue receipts.
Iris closes every run with a hash-anchored receipt routed to the receipt rail.
Operator loop
Interpret → plan → route → approve → execute → grade → receipt.
- 01STEPInterpretIris parses the request into a governed run.
- 02PASSPlanSteps, systems, approvers, policies.
- 03PASSBind policyGW Slate locks the run to a policy pack.
- 04PASSRoute modelLocal, hybrid, or cloud — per policy.
- 05GATEDecideIris asks the right human at the right time.
- 06PASSExecuteDry-run, diff, then promote.
- 07PASSGradeOutcome grade across dimensions.
- 08PASSReceiptSealed and routed to the receipt rail.
Outcome grade
Every run closes with a grade and a reason.
Outcome grade
A−
Workflow passed all gates with one minor brand revision.
- Brand
- 92/100
- Locale
- 88/100
- Safety
- 96/100
- Provenance
- 100/100
Voice + claims aligned.
DE, JA approved.
Outcome grade
F
Run blocked: unverified product claim. No CMS writeback occurred.
- Brand
- 70/100
- Locale
- 80/100
- Safety
- 50/100
- Provenance
- 100/100
Unsupported claim detected.
Policy P-1 violated.
What makes Iris different
A control-room operator, not a chat toy.
Plans, doesn't just generate
Iris writes the run plan first, binds policy second, executes third.
Explains routing
Iris tells you why she chose local, hybrid, or cloud — and what policy forced the choice.
Asks the right human
Iris reads the policy pack, picks the approver, and waits.
Reads back the grade
Iris reports the outcome grade in plain language with the receipt link.
Stops when policy says stop
Iris does not retry around a block. She returns a reason and a receipt.
Backed by GW Slate
Iris cannot act without policy, identity, and approval bound at runtime.
Buyer-specific examples
VP / Director
How does Iris show up in our day?
She runs governed workflows you point her at. She narrates her plan, surfaces approvals on a clean surface, and hands back a graded receipt. Your team uses AI; you keep authority.
- Operator surface for governed work
- Plain-language run plans
- One-click approval flow
- Graded receipts for review
Run one workflow with Iris.
Pick the workflow. Iris runs it. You see the plan, the approval, the grade, and the receipt.