Human Review & Approvals

Overview

Not all work should be fully autonomous.

Supervity follows a Human-in-Command model, where AI agents operate independently until human judgment, approval, or input is required.

Human review is not an exception or workaround — it is a first-class governance mechanism designed for real-world, enterprise-grade operations.

Why Human Review Exists

Human review enables safe delegation of work to AI agents while preserving accountability.

Compliance & Governance

• Financial approvals
• Legal and contract review
• Policy enforcement
• Audit and regulatory requirements

Quality Control

• Content moderation
• Output validation
• Data accuracy checks

Critical Decisions

• External customer communication
• Infrastructure or system changes
• Access and permission changes

Exception Handling

• High-value transactions
• Unusual patterns
• Edge cases and anomalies

How Human Review Works

Execution Pausing & Resumption

When an agent reaches a review step:

1. Execution pauses automatically
2. Reviewers are notified
3. A review form is generated with full execution context
4. A human submits a decision
5. The agent resumes execution based on that decision

Agents never bypass review steps unless explicitly configured to do so.

Review Forms

Review forms are generated dynamically and can include:

• Approve / reject actions
• Comments or rationale fields
• File uploads and attachments
• Multiple-choice decisions
• Conditional logic based on responses

Forms can be accessed inside Supervity or embedded externally.

Creating Human Review Steps

Using AI Agents (Recommended)

You can request human oversight in natural language:

“Send the drafted email to the manager for approval before sending.”

The agent will:

• Insert a review step
• Design the approval form
• Configure notifications
• Define resume behavior

This is the fastest and safest way to add human review.

Manual Configuration

You can also add review steps explicitly:

1. Add a Human Review step
2. Configure the form:
   • Fields (text, choice, file, etc.)
   • Required inputs
   • Reviewer instructions
3. Assign reviewers:
   • Specific users
   • Roles
   • External email addresses
4. Define timeouts and escalation rules

Review Experience

For Reviewers

Accessing Reviews

Reviewers receive review requests via:

• Email notifications
• In-app review dashboard
• Embedded external systems (Enterprise)

Each review includes:

• Context from earlier steps
• Relevant data and attachments
• Clear instructions on what’s required

Reviewer Actions

Reviewers can:

• Approve or reject
• Add comments
• Upload supporting files
• Modify inputs (if allowed)
• Escalate or delegate (if enabled)

All actions are logged and auditable.

For Workflow Owners

Monitoring Review Status

Each review moves through clear states:

• Pending
• Approved
• Rejected
• Expired

Status is visible in real time.

Audit Trail & History

Supervity records:

• Who reviewed
• When the decision was made
• The decision outcome
• Comments and attachments
• Impact on downstream execution

This supports audits, compliance, and retrospectives.

Configuration Options

Reviewer Assignment

Reviewers can be assigned as:

• Single reviewer
• Multiple reviewers:
  • Any one
  • All required
  • Majority-based
• Role-based (e.g. Manager, Admin)
• Dynamically based on workflow data

Notifications & Escalation

Notify reviewers via:

• Email
• Slack / Microsoft Teams
• In-app notifications

Escalation options include:

• Reminder after X hours
• Backup reviewers
• Hierarchical escalation

Timeouts & Fallbacks

Define what happens if no action is taken:

• Auto-approve
• Auto-reject
• Escalate
• Pause indefinitely

Example:

“If not reviewed within 24 hours, escalate to the senior manager.”

Common Human-in-Command Patterns

Approval Gate

Conditional Review

Multi-Level Review

Best Practices

Design Principles

• Use review only where judgment is required
• Provide full context to reviewers
• Keep forms concise
• Avoid unnecessary approval layers

Performance Considerations

• Set reasonable timeouts
• Use escalation paths
• Monitor review bottlenecks

Security

• Restrict reviewer access
• Protect sensitive data
• Enable audit logging

External & API-Driven Reviews (Enterprise)

Human review can be embedded into:

• External dashboards
• Command centers
• Internal tools

Using APIs, external systems can:

• Fetch pending reviews
• Render review forms
• Submit approve or reject actions
• Resume agent execution programmatically

See API Reference.

Monitoring & Analytics

Track review performance using:

• Average review time
• Approval and rejection rates
• Bottlenecks
• SLA compliance

These insights help improve operational efficiency.

Troubleshooting

Review Not Received

• Check notification settings
• Verify reviewer permissions
• Resend review request

Workflow Stuck Waiting

• Check timeout configuration
• Escalate manually (Admin)
• Cancel or restart execution

Getting Started

• Tutorial: Approval Workflow
  

  text

• Example: Expense Approval
  

  text

• View Pending Reviews
  Available from the Human Review dashboard

Human Review ensures AI agents act with autonomy, but never without accountability.