Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minutePut approval immediately before the consequential external action the reviewer is meant to control—not merely after content is generated or somewhere earlier in the process. Match the gate to the action’s potential harm and reversibility, show the reviewer the exact action and the evidence behind it, and keep the item in draft until approval is recorded.
Choose the gate by consequence, not by habit
A single approval rule for every automated action creates two problems: it slows low-risk work while leaving dangerous actions with too little scrutiny. The Microsoft-published AI Agent Runbooks review-and-approval pattern distinguishes four levels of control. Treat them as a decision framework, not a universal standard:
| Gate | When it fits | What happens |
|---|---|---|
| Notify | Low-value actions that are easy to reverse | The workflow acts, then informs a person. |
| Confirm | Moderate-consequence actions a reviewer can assess | The workflow summarizes the proposed action and waits for confirmation before acting. |
| Draft | Content carrying an organization’s voice, commitments, or numbers | Automation prepares the content; a person commits it, such as by sending it. |
| Mandatory review | Regulated or safety-sensitive actions | A named, qualified role must sign off before execution. |
For outbound communication, separate creating a message from sending it. Automated systems may prepare or revise a customer-facing message, but the reviewer should approve the actual message that will leave the organization. A confirmation click is not meaningful if the person cannot evaluate whether the content is correct. As the Microsoft-published pattern puts it, “A gate where the human cannot realistically assess correctness is theatre.”
Make the review inspectable
Present a review unit small enough to examine, with the exact proposed action and its supporting context. Show evidence, confidence where available, and changes from the prior version. Avoid prompts that ask for approval without providing the material needed to judge the request. Bulk approval is also a poor fit when the volume or grouping makes individual inspection unrealistic.
Recommended Free Tools
Keep the destination status honest
Until the reviewer commits the action, preserve draft status in the destination system—the place where colleagues or customers may encounter it. A draft label in a chat transcript alone does not prevent someone downstream from treating content as final. If the action changes after approval, require review of the new version: the earlier decision did not cover content the reviewer never saw. This is a conservative design rule; the cited product documentation does not establish a universal technical mechanism for binding approval to an exact payload.
Decide authority and rejection before launch
For high-consequence work, name the role authorized to approve and decide whether the reviewer must be independent of the person or system that prepared the action. Define what rejection means: revise, reassign, return to the requester, or discard. A workflow should not leave declined work in an ambiguous state or silently send it later.
Place approval in the workflow as a real state
Approval should pause execution as an explicit state, not amount to an informal notification that may be missed. A dependable pattern prepares the proposed action, routes it to an authorized reviewer, pauses, records the decision, and resumes only according to that decision.
- Prepare: Assemble the final proposed action and the evidence needed to assess it.
- Route: Send the review request to someone with the relevant knowledge and authority.
- Pause: Prevent the external action while approval is pending.
- Record: Capture approval, rejection, requested changes, or timeout as distinct outcomes.
- Recheck: Confirm the action still matches the version the reviewer saw.
- Execute: Perform only the approved action, then record the outcome and any correction.
Microsoft Agent Framework documents approval-required tool calls that pause a workflow and emit a request event; a response resumes it. It distinguishes approve-or-reject tool approval from an interactive handoff, which is used when an agent needs free-form input. Pending requests can be retained in checkpoints and re-emitted when a workflow is restored. These are framework-specific mechanics, not capabilities to assume in other systems. See the Microsoft Agent Framework human-in-the-loop workflow documentation.
Plan for no response and failure
Specify what happens if the reviewer does not respond: expire, cancel, escalate, or remain paused. Do not let a timeout implicitly become approval unless that is an explicit, defensible policy for the action. Elastic describes pausing at a critical decision, presenting structured findings, waiting for a response, and resuming based on it. Its documentation identifies potentially impactful remediation, ambiguous classifications, escalation, and early trial runs as situations for human input. Timeout behavior varies by Elastic Stack version and step type, so set and verify the policy for the deployed version in the Elastic Workflows human-in-the-loop documentation.
Account for platform-specific approval mechanics
Workflow products can supply useful pause, routing, and resume mechanisms, but a feature list does not establish that a workflow is well governed. Check the implementation against the team’s actual action risks, identity rules, destination system, and audit requirements.
Rank #4
Microsoft Human in the Loop connector
Microsoft’s Human in the Loop connector documentation says approval timestamps are displayed in UTC and an approver can be specified by email, UPN, or Microsoft Entra ID. It also lists request throttling limits and limitations involving guest users and actionable approval email. Verify these constraints against the intended reviewer population and workload.
Zapier Request Approval
Zapier’s help page, marked updated October 1, 2026, describes a Request Approval step that pauses a Zap so reviewers can approve, decline, or change submitted content. The feature requires a paid account and sharing the Zap with reviewers. See Zapier’s Request Approval documentation.
Free tools Windows power users keep installed
One-click scans. No signup required.
Oracle Integration human tasks
Oracle Integration documentation describes assigned human tasks and forms for consequential actions, uncertainty, and tool errors. Check that the task assignment and form capture the decision and context your process needs.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Measure whether the gate is working
Track measures that reveal both the burden of review and the cost of getting a decision wrong. The Microsoft-published pattern recommends:
- Straight-through rate: how often work proceeds without a human gate.
- Reviewer time per item and time spent waiting in the review queue.
- Correction rate by field or action, plus rejection rate.
- Defects discovered after approval.
Use the results to determine whether the review surface gives people enough information and whether the gate is positioned appropriately. Start with review where correctness has not been demonstrated. Relax controls only for lower-risk categories after sustained evidence and an explicit business decision; keep high-consequence actions gated unless an accountable decision changes that policy.
Evaluate the design, not just the approval button
Before deployment, walk through a normal case, a rejection, a timeout, an escalation, and a change made after approval. Check these points:
- Is the gate directly before the action it is meant to control?
- Can the reviewer inspect the exact content, evidence, and relevant changes?
- Does the reviewer have the expertise and authority the action requires?
- Does the destination system clearly retain draft status until commitment?
- Are rejection, timeout, escalation, and correction recorded and handled deliberately?
- Can the workflow verify that the executed action is the one approved?
- Do identity, access, integration, version, throttling, and account limits fit the deployment?
- Are reviewer effort and queue delays measured alongside defects?
The Microsoft runbook is guidance, not a formal standard. The cited sources document design patterns and platform features, not comparative performance or a universal optimal placement for approval. Choose a gate policy for the specific action and make the decision traceable.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
