Feature Deep Dive

Approval workflows for AI agents

Let agents prepare drafts and reports. Configure which actions need human review, check the sources and inspect activity logs before expanding what an agent can do.

DraftLet agents prepare the work.
ReviewStop sensitive actions for human approval.
CiteShow sources behind answers and decisions.
LogRecord what happened for later inspection.

Where approval fits in an agent workflow

An approval gate pauses an agent before an action covered by your review rules. Manor combines those rules with source citations and activity logs. What runs automatically and what needs review depends on your configured permissions, workflow and supported tools.

If you are comparing approval workflow software, this page covers review of AI agent work. Check your required routing, reviewer access and connected actions before choosing a setup.

What It Does

Human-in-the-loop control

Keep the owner in the loop when a workflow touches money, customers, commitments, legal wording, or uncertain context.

Sources and activity history

Use citations, summaries, and status logs so the reviewer can understand what the agent used and why it escalated.

Gradual autonomy

Start with drafts and reports, then widen narrow low-risk permissions only after the workflow has earned trust.

Set review rules before an agent acts

Example: a customer reply draft. Give the agent the customer's question and approved policy sources. Ask it to prepare a reply and flag missing information. Check the draft against those sources and correct any promises before sending. For the first test, keep sending manual; only allow a connected send after verifying its permissions and review behavior.

01

List sensitive actions

Name what must never run unseen: refunds, price changes, legal language, angry customers, payments, and external sends.

02

Define approval rules

Tell the agent when to draft, when to route, and when to stop for review.

03

Require citations and logs

Make customer-facing answers and recurring runs show source context and activity history.

04

Review misses weekly

Use the log to adjust sources, skills, permissions, and escalation rules.

Actions that should stay gated

  • Refunds, discounts, and billing changes.
  • Legal, compliance, or contract wording.
  • Angry customers or sensitive relationships.
  • New public claims or pricing promises.
  • Any action with missing source context.

Related Pages

FAQ

What is an approval gate?

An approval gate is a rule that makes an agent stop and ask a human before a sensitive action continues.

Why do activity logs matter?

Logs show what the agent inspected, prepared, skipped, escalated, and changed, which makes automation easier to supervise.

Does approval-first mean no automation?

No. It means automation starts with visible drafts, reports, and internal tasks before expanding to low-risk actions.

Use AI agents without hiding the decisions that matter

Create Manor workflows with approvals, citations, and logs from the first run.

Launch Manor