Trust Ladder
The progressive autonomy system that governs how much freedom AI agents earn based on their track record. Agents start restricted and earn more autonomy through consistent, high-quality decisions.
Why the Trust Ladder?
Giving an AI agent full autonomy on day one is risky. Requiring manual approval for every action forever is unsustainable. The Trust Ladder solves both problems with progressive autonomy — agents earn freedom through performance.
New agents prove themselves before getting autonomy. Bad judgment is caught early while stakes are low.
As agents earn trust, they handle more decisions autonomously. Your approval queue shrinks over time.
You can override tiers at any time. Set minimum tiers, cap maximum tiers, or disable auto-upgrade entirely.
The 4 Tiers
Probation
All actions need approval
Every agent starts at Probation. All decisions — regardless of value or confidence — require explicit merchant approval. This lets you observe the agent's judgment before granting any autonomy. The agent proposes actions and waits for you to approve or reject each one.
Requirements
0 (starting tier)
Success Rate
N/A
Min Age
N/A
Auto-Approve Below
$0
Force Approval Above
$0
Supervised
Low-value auto-approved up to $50
The agent has demonstrated consistent judgment on initial decisions. Low-value actions (under $50) are auto-approved — things like small discount adjustments, inventory reorder suggestions, and routine follow-up messages. High-value actions still require your approval.
Requirements
20+ decisions at 80%+ success rate
Success Rate
80%
Min Age
3+ days
Auto-Approve Below
$50
Force Approval Above
$200
Trusted
Most auto-approved up to $500
The agent has built a strong track record over two weeks. Most actions up to $500 are auto-approved — discount offers, deal adjustments, inventory orders, customer outreach. Only exceptions and high-value decisions above $500 require manual approval. This is where most well-performing agents settle.
Requirements
100+ decisions at 90%+ success rate
Success Rate
90%
Min Age
14+ days
Auto-Approve Below
$500
Force Approval Above
$2,000
Autonomous
Full autonomy up to $5,000 within guardrails
Elite tier for agents with exceptional track records. The agent operates with full autonomy for actions up to $5,000 — only extraordinary situations require approval. Guardrails still apply (max discount percentages, forbidden actions, escalation rules). This tier is earned, not granted.
Requirements
500+ decisions at 95%+ success rate
Success Rate
95%
Min Age
30+ days
Auto-Approve Below
$5,000
Force Approval Above
$10,000
How Upgrades Work
Tier upgrades are evaluated automatically every time an agent makes a decision (on every agent listing). The system checks three criteria against the next tier's requirements:
Decision Count
Total number of decisions the agent has made since activation. Includes both auto-approved and manually reviewed decisions.
Success Rate
Percentage of decisions that were successful — approved by merchant or achieved desired outcome. Rejected decisions count against the rate.
Age Since First Decision
Number of days since the agent made its first decision. Prevents rapid promotion — even a perfect agent must prove consistency over time.
// Tier upgrade evaluation (runs on every agent decision)
function evaluateTierUpgrade(agent) {
const stats = agent.getPerformanceStats();
const nextTier = getNextTier(agent.current_tier);
if (!nextTier) return; // Already at Autonomous
const req = nextTier.requirements;
const daysSinceFirst = daysBetween(stats.first_decision_at, now());
if (
stats.total_decisions >= req.min_decisions &&
stats.success_rate >= req.min_success_rate &&
daysSinceFirst >= req.min_days
) {
agent.upgradeTo(nextTier.name);
notifyMerchant(
`Agent "${agent.name}" promoted to ${nextTier.name}!
` +
`Decisions: ${stats.total_decisions} | ` +
`Success: ${(stats.success_rate * 100).toFixed(1)}% | ` +
`Days active: ${daysSinceFirst}`
);
}
}Approval Overrides by Tier
Each tier defines two dollar thresholds that control automatic approval behavior. Actions between the two thresholds follow the configured approval mode (auto, manual, or hybrid).
| Tier | auto_approve_below | force_approval_above | Behavior |
|---|---|---|---|
| Probation | $0 | $0 | All actions require approval |
| Supervised | $50 | $200 | Small actions auto-approved |
| Trusted | $500 | $2,000 | Most actions auto-approved |
| Autonomous | $5,000 | $10,000 | Full autonomy within guardrails |
How the thresholds interact: Actions belowauto_approve_below are always auto-approved. Actions above force_approval_above always require manual approval. Actions in between follow the agent's configured approval mode.
Manual Override
Merchants can override the automatic tier system at any time using the/mintier command or the API. This sets a minimum tier floor — the agent will never drop below this tier, even if performance metrics temporarily dip. You can also use it to fast-track a trusted agent.
# Set minimum tier via messaging command
/mintier supervised
# Response:
# Minimum trust tier set to: Supervised
# All agents at Probation tier have been upgraded.
# Affected agents: 1 (Outreach Agent)
# Set for a specific agent via API
PUT /api/v1/agents/{agent_id}/trust-tier
{
"min_tier": "trusted",
"auto_upgrade_enabled": true
}
# Response:
{
"agent_id": "agent_pricing_abc",
"previous_tier": "supervised",
"current_tier": "trusted",
"min_tier": "trusted",
"auto_upgrade_enabled": true,
"next_tier_requirements": {
"tier": "autonomous",
"decisions_needed": 358,
"success_rate_needed": "95%",
"days_remaining": 16
}
}Use with caution: Setting/mintier autonomous gives all agents full autonomy up to $5,000 regardless of their track record. This bypasses the safety of progressive trust building. Consider setting it per-agent via the API for more granular control.