---
title: "How do I divide AI review responsibilities across my team?"
description: "Assign responsibility for tasks, skills, agent permissions, and the evidence used to improve them."
sidebar_title: "Assign review responsibilities"
order: 3
---

# How do I divide AI review responsibilities across my team?

Give each workflow a person responsible for its intended use and acceptance criteria. Use Orlo presets to let colleagues contribute without giving everyone administrative access.

| Responsibility | Orlo access or workflow |
| --- | --- |
| Define and release a task | An admin manages the standard, comparisons, and releases |
| Inspect task results | Task observer reads standards, comparisons, and live results |
| Write shared procedures | Skill contributor saves drafts; Skill manager reviews and publishes |
| Connect an agent | Agent operator sets it up; an admin supplies runtime credentials |
| Set tool permissions | Agent permissions manager maintains rules and limits |
| Decide a held action | Agent approver reviews the exact call |
| Review examples | Reviewer assesses corrections, labels, and saved steps |
| Inspect recorded history | Evidence auditor reads metadata; other presets can add relevant detail |

Presets can be combined. They apply across the organization within their area; a link to one task is not a resource-level restriction. Use [People and access](/console/settings/people-keys-and-notifications) to invite colleagues and set an expiry where needed.

## A working rhythm

1. Define the task’s intended use and representative examples.
2. Compare candidates and inspect disagreements before releasing.
3. Test application handling and agent permissions before real traffic.
4. Review signals, held actions, and proposed corrections regularly.
5. Add accepted examples, compare the next version, and record the release decision.

Business, security, and legal teams determine the requirements for their context. Orlo records the configured behavior and submitted work; a review process must still decide whether the evidence meets those requirements.

Start with [one production improvement](/console/tasks/improve-from-production), then apply the same review process to the next workflow.