Human-AI Interaction Standards
In the 10ˣ system, every interaction between humans and AI follows a defined protocol. No implicit assumptions. No autonomy without authorization. No black boxes.
Protocol: A defined interaction pattern between agents (human or AI) that governs handoffs, decisions, and execution. Protocols make collaboration explicit and auditable.
Suggest → Verify → Execute
The default protocol for high-stakes or novel interactions. AI generates a recommendation or proposed action. A human reviews and confirms before execution proceeds.
When: Use when: the action has external impact, stakes are high, or the pattern is new.
Steps
AI analyzes context and produces a recommendation with full reasoning.
Human examines the suggestion in context. All reasoning is visible — no black box.
Human confirms the recommendation is correct and authorized.
The action executes with human authorization. The audit trail records who authorized what and when.
In copy (human)
“AI suggests next steps; engineers review and confirm.”
In metadata (machine)
{"protocol": "AI → Human verification before execution"}Autonomous with Audit Trail
Used for high-confidence, routine actions where manual review adds friction without adding safety. AI acts autonomously — but every action is recorded.
When: Use when: the pattern is well-established, confidence is high, and the action is reversible or low-risk.
Steps
AI assesses its confidence level against a pre-defined threshold.
Confidence exceeds the autonomous threshold. Execution proceeds without human review.
The action is performed. No human delay in the loop.
Every detail — the action, the confidence score, the data that drove it — is logged immutably.
In copy (human)
“AI executes routine tasks; full audit trail available for review.”
In metadata (machine)
{"protocol": "AI executes after threshold confidence; immutable audit trail recorded"}Human Escalation
When AI encounters ambiguity, risk, or conditions outside its operating parameters, work is routed immediately to a human agent — with full context already assembled.
When: Use when: edge cases arise, risk flags are detected, or the situation falls outside trained parameters.
Steps
AI identifies ambiguity, a risk flag, or an out-of-bounds condition.
AI assembles all relevant context — history, options, constraints — into a structured brief.
The right human is notified immediately, with the full brief already assembled. No re-gathering.
Human makes the decision. Control returns to the AI-assisted flow with the resolution recorded.
In copy (human)
“AI flags uncertainty; human resolves with full context provided.”
In metadata (machine)
{"protocol": "AI detects escalation → Human resolves with full context"}Agents in the 10ˣ system
Human Roles
- Engineer
Implementation, architecture, technical decisions
- Manager
Delivery coordination, escalations, priorities
- Operator
Process execution, exceptions, quality
- Executive
Strategic decisions, resource allocation
AI Roles
- AI Assistant
Task automation, synthesis, first drafts
- Predictive Advisor
Pattern recognition, risk, predictions
- Flow Controller
Workflow orchestration, routing, handoffs