Action tiers
Map each task to suggest, draft, request approval, execute approved low-risk work, or block and escalate.
Loading AI Team page
A per-Agent action table that makes autonomy levels visible by action type, channel, system, approval owner, cost limit, and escalation trigger.
Buyer question
Autonomy becomes a reviewed operating choice instead of an unclear demo promise. UAE SME buyers are open to automation but cautious about liability. Clear authority tiers help owners keep control of customer messages, finance impact, access changes, and public commitments.
Package outcome
Autonomy becomes a reviewed operating choice instead of an unclear demo promise.
What buyers should see
Each section should be short enough for an SME owner, manager, or procurement reviewer to understand before approving access or go-live.
Map each task to suggest, draft, request approval, execute approved low-risk work, or block and escalate.
Record the client owner or AI Team reviewer for customer-facing, irreversible, high-cost, sensitive, or unclear actions.
Require reassessment when the Agent gets a new channel, tool, provider, system write, public-output path, or cost exposure.
Artifact template
Make the Agent's autonomy boundaries inspectable by action type, channel, system, approval owner, cost threshold, and escalation trigger.
List each meaningful action the Agent may suggest, draft, request approval for, execute, or block.
delegated_authority_tier
Use only the approved authority modes: suggest, draft, request approval, execute approved low-risk work, or block.
approval_policy
Name the client approver or AI Team reviewer for external, irreversible, costly, sensitive, or unclear actions.
human_review_decision
Map each action to the connected system, permission, write path, and revocation owner.
security_tool_boundary
Record warning threshold, monthly cap, stop rule, and escalation owner for provider or paid-tool use.
pass_through_cost_policy
Define the conditions that force refusal, approval, manual review, pause, or re-scope.
refusal_escalation
Before go-live and whenever tool access, channel, public-output path, cost exposure, or workflow scope changes.
Risk handling
A package only builds trust when it also explains when AI Team should pause, escalate, re-scope, or decline instead of pushing the Agent live.
Source records
These are the implementation records and public surfaces that should remain aligned as the package becomes a dashboard, PDF, or sales handoff.
Related paths
Use these links when a buyer needs supporting context before setup review.
FAQ
These answers are intentionally practical so owner-operators can decide whether the package is ready, needs review, or should pause setup.
Autonomy becomes a reviewed operating choice instead of an unclear demo promise.
The action affects money, legal exposure, HR, regulated work, or public claims. The Agent asks for a tool or record outside approved scope. The customer request is angry, urgent, ambiguous, or policy-sensitive. The cost would exceed the approved warning threshold or stop rule.
Use AI_TEAM_AGENT_CATALOG.md, config/agent-tool-governance.json, config/agent-evidence-pack.json, supabase/tests/agent_execution_approval_bridge_tests.sql as the starting source records for this package.