Human Approval

Autonomy scales until it doesn't. Enterprises need a circuit breaker where consequential decisions require a qualified human to approve, reject, or edit before execution.

A surgeon must sign off before the robotic arm makes the incision — automation assists, but accountability stays human for irreversible steps.

Scenario

An agent drafts a $50K credit memo. Instead of posting to ERP, it opens an approval request with customer history and policy citation. A finance manager approves in Slack; only then does the agent call `create_credit_memo`.

Solution

In Enterprise AI, apply Human Approval to this scenario: An agent drafts a $50K credit memo. Identify the inputs, run the technique, validate the output, and note one thing you would monitor in production.

Do this before moving to the next module — reading alone is not enough.

Open the Code Walkthrough below and run it locally. Change one parameter related to Human Approval (e.g. model, temperature, top_k, or tool name), observe the difference in output, and write 2–3 sentences explaining what changed.

Highlighted lines show where Human Approval happens in the code.

1async function executeWithApproval(action, context) {  # key line for Human Approval2  if (action.riskTier >= RISK_HIGH) {3    const ticket = await approvals.create({  # core API call for Human Approval4      action: action.name,5      payload: context.payload,6      requester: context.userId,7      expiresIn: "4h",8    });9    const decision = await approvals.waitForDecision(ticket.id);  # key line for Human Approval10    if (decision.status !== "approved") throw new ApprovalRejected(decision.reason);  # key line for Human Approval11  }12  return action.handler(context);  # return the result13}

• Treating Human Approval as a black box without evaluation
• Ignoring cost and latency in production
• Skipping error handling for enterprise human approval

Quick recap — the most important points from this module.