Human-in-the-loop by design

Governed by your judgment.
Powered by their intelligence.

ETAM is a 4-stage progression that governs every action your AI agents take. No agent ever acts without earning the right first. AI that earns trust. Humans who keep control.

The problem with AI today

All-or-nothing is the only option.

Either you give the AI full control and hope it doesn't make mistakes, or you approve every action and it takes more time than doing it yourself. There is no middle ground.

The ETAM solution

A graduated trust system.

Agents start by watching. Then analyzing. Then recommending. Only after they prove themselves do they earn the right to act on their own.

The four stages.

How agents earn autonomy, one rung at a time.

OBSERVE

Monitor

Watches business data in real time. Tracks patterns, detects anomalies, learns what normal looks like. Never acts. Only observes.

CONNECT

Analyze

Connects dots across departments. Surfaces trends and insights you would not have noticed. Presents analysis but does not suggest actions.

PROPOSE

Recommend

Proposes specific actions with supporting data and confidence scores. You approve, reject, or modify each. Where most agents live day-to-day.

EXECUTE

Act

Executes approved action types autonomously. Logs everything. Reviewable and revocable. Earned through consistent performance, never given.

You approve every step. Until you don't have to.

Less trust · More oversight
More trust · Less oversight
How trust is measured

Two systems keep every agent honest.

Trust is calculated, not declared. Every approval and rejection feeds a confidence score that controls what the agent is allowed to do next.

01 The asymmetric trust engine

Trust is hard to earn, easy to lose.

Confidence is recalculated on every decision, but not symmetrically.

+1x

every approval

-5x

every rejection

One bad recommendation erases five good ones. Agents must consistently prove judgment before earning more autonomy.

02 The rolling decision window

Recent work decides, not lifetime stats.

A rough start can be recovered. A strong agent cannot coast.

You define the decision window for each agent based on their role. Some roles need 50 consistent decisions before earning autonomy. Others need 10. Each new decision rolls in, the oldest rolls out. Adjust the window anytime as your confidence grows.

Hit your threshold → eligible for next trust level
85%rolling window

03 Three confidence zones

What happens at each score.

You control every threshold. Defaults shown are conservative by design.

your agent · 78% (soft gate)
0%65%90%100%
Hard StopSoft GateAuto

Hard Stop

< 65% confidence

Requires explicit human approval. No timer. No auto-approve. The agent waits, no matter how long.

Soft Gate

65 - 89% confidence

Agent recommends; auto-approves after 24h if no response. You can override at any time.

Full Autonomy

90%+ confidence

Agent acts immediately without asking. Every action logged for review. Revoke any time.

HUMAN-IN-THE-LOOP

You are always in control.

Trust is graduated, not delegated. Every level of autonomy comes with guardrails you own and can revoke with one click.

No self-promotion

No agent ever promotes itself. You must explicitly approve every trust-level upgrade, even if it has earned the score.

One-click revoke

Override any agent at any time. Revoke autonomy with one click. Instant rollback to recommend-only mode.

Full audit trail

Every action is logged with the reasoning, the data it acted on, and the confidence score at the moment of decision.

Per-agent trust

Different agents, different trust levels. Trust your CMO agent to act; keep your CFO agent in approval mode. Your call.

See ETAM in action.

AI that earns trust. Humans who keep control. Book a 30-minute call and watch an agent earn trust in real time on your data.