You Stay in Control

The Control Plane exists because AI without governance is a liability. We invented this pattern because the alternative - autonomous AI touching production systems - is unacceptable risk.

---

The Three-Part Contract

Every interaction with Azi follows the same pattern:

1. Azi Proposes

Azi generates:

• The specific action (query, configuration, etc.)
• Reasoning for why this approach
• Impact analysis (what will change)
• Rollback information (how to reverse)

2. You Review

Before anything executes, you see:

Element	What You Get
The action	Exact KQL, configuration change, or system modification
The reasoning	Why Azi chose this approach
The impact	What will change in your system
Test option	Run Query to preview results

3. You Decide

Three options with clear visual buttons:

Action	Button	What It Does
Approve	✓ checkmark	Execute the proposed action
Reject	🗑️ trash	Cancel and remove the proposal
Edit	✏️ pencil	Modify the proposal before approving

Visual cue: Proposed nodes appear with a gold/tan background and dashed edges - visually distinct from approved nodes. This makes it impossible to mistake a proposal for an executed action.

---

Autonomous vs Governed

Autonomous AI	Governed AI (Control Plane)
AI decides, system executes	AI proposes, human decides
"Trust the model"	"Verify then approve"
Black box reasoning	Transparent reasoning
Scattered logs	Complete audit trail
Rollback = hope	Rollback = built-in
Vendor controls AI	You control AI

---

The Audit Trail

Every interaction is permanently logged:

What	Recorded
Your request	The natural language or UI action that started it
Azi's proposal	The full action, reasoning, and impact analysis
Your decision	Approve, reject, or edit - with identity
Timestamp	When each step happened
Outcome	What changed in the system

Example Audit Entry

{
  "requestId": "req-20260118-001",
  "timestamp": "2026-01-18T10:30:00Z",
  "user": "operator@brewery.com",
  "request": "Show me temperature spikes over 80 this week",
  "proposal": {
    "type": "warm_query",
    "kql": "Telemetry | where temperature > 80 | where timestamp > ago(7d)",
    "reasoning": "Filtering for temperature values above 80°F in the last 7 days"
  },
  "decision": "approved",
  "decisionBy": "operator@brewery.com",
  "decisionAt": "2026-01-18T10:30:45Z"
}

---

Why This Matters

Compliance

When auditors ask "who approved this AI decision?" - you have an answer.

Accountability

When something goes wrong, you know exactly:

• What was proposed
• Who approved it
• When it happened
• How to reverse it

Trust

Your team can use AI assistance knowing that humans remain in control of production-impacting decisions.

---

Validation Before Deployment

The Control Plane catches errors before they hit production:

When you attempt to deploy invalid configurations:

1. Validation runs automatically before deployment
2. Errors are grouped by node - you see exactly which component has issues
3. Specific error messages tell you what to fix (e.g., "Description is required")
4. Deployment blocks until all validation errors are resolved

This isn't just error handling - it's governance in action. The system prevents invalid states from reaching production.

---

Reversibility

Every governed action includes a rollback path:

Action Type	Reversibility
Queries	Delete or modify the warm query
Configurations	Restore previous state from audit log
Connections	Disconnect, reconnect with prior config
Surfaces	Archive and restore from snapshot

Nothing is permanent. Nothing is one-way. You can always return to a known state.

---

The Bottom Line

You stay in control because:

1. Every proposal requires explicit approval
2. Every approval is logged with attribution
3. Every action is reversible
4. Your data stays in your Azure tenant
5. You can fire Azi and keep running

This is the Control Plane. We invented this pattern because industrial AI without governance isn't safe. Now it's essential.

---

Back to Start

← Back to Meet Azi

Start the Quickstart →