Agent Workflows — AI Playbooks

Agent Workflows — AI Playbooks

1. What it is

AI Playbooks let you build reusable, multi-step action sequences that span multiple apps. A playbook chains tool calls across domains — scheduling, HR, forms, service desk, and others — into a single runnable procedure. Users trigger playbooks through Ask AI by matching a trigger phrase, admins run them manually from the admin surface, or automation triggers fire them on a schedule or in response to a platform event.

• Enablement: This is a platform admin surface, not a separately licensed app. Any admin can access it at Admin → AI Playbooks (the direct URL is /admin/agent_workflows). When the Loop app is enabled for your business, these pages move into the Loop app tabs instead.
• What it is not: Playbooks are not the same as the AI Operations kill switches. AI Operations (Admin → AI Operations) governs whether agents can act at all; playbooks define what a sequence of actions does and when it runs.

2. Standing it up

1. Go to Admin → AI Playbooks.
2. Click New Playbook. Give it a name (required, max 100 characters), pick a primary domain, and optionally add trigger phrases so users can invoke it from Ask AI.
3. Build the steps. You have two paths:
   • Build with AI: type a plain-English description of what the playbook should do. AI drafts the name, description, steps, and declared inputs for you. Review the result in the visual step builder and edit before saving.
   • Manual: click Add step and pick an app (domain) and action (tool) for each step. Pass values between steps using $inputs.<name> or $steps.<step id>.<key>.
4. Mark any step that writes data as Require confirmation — it will pause for admin approval before executing.
5. Save. The playbook is active by default. Users whose Ask AI query matches a trigger phrase will see it offered; admins can also run it manually from the playbook detail page.

To set up unattended execution, create an Automation Trigger from the Automation Triggers tab. A playbook that pauses part-way through to ask the user for a value cannot be automated — the trigger will not save, because nobody is at the keyboard to answer.

3. How it fits together

Playbook — A named, versioned sequence of steps, laid out in the Steps designer. Each playbook belongs to one primary domain (for grouping and routing) but its steps can call tools from any domain. A playbook with steps is “executable” and can be run; one without steps is a “prompt-hint” that only guides the AI’s response.

System playbooks vs tenant playbooks — MangoApps provides global system playbooks (visible in the “System playbooks” section at the bottom of the index page). These are available to your users through Ask AI by default. You can:

• Disable a system playbook for your business so it no longer appears in Ask AI.
• Clone a system playbook into an editable tenant copy (the copy tracks its parent but is fully independent).
  Tenant-owned playbooks are managed with the normal activate/deactivate/delete controls.

Trigger — Binds a playbook to a schedule or a platform event so it runs unattended. Scheduled triggers use cron expressions interpreted in your business’s time zone. Event triggers fire when a matching platform event occurs (leave request created, shift cancelled, form submitted, etc.). A trigger runs as a specific user (set at creation) and has built-in duplicate-run protection: if a previous run from the same trigger hasn’t settled within 24 hours, the next occurrence is skipped.

Run — One execution of a playbook. Each run pins the execution plan at start so mid-run edits to the playbook do not change a run already in progress. The run tracks per-step results (ok, failed, skipped), an overall status, and wall-clock duration. Statuses: Pending, Running, Waiting, Awaiting confirmation, Confirmed, Completed, Failed, Cancelled.

Confirmation gate — Steps marked Require confirmation before running (or flagged as risky by their agent) pause the run and send an approval request to the run owner’s Inbox. The approval window is 7 days; after that the confirmation expires. Automation triggers can pre-approve specific tools via the auto-approve setting.

4. Running it

Creating a playbook with AI

1. Go to Admin → AI Playbooks → New Playbook.
2. In the Build with AI panel at the top, describe what the playbook should do.
3. Click Draft playbook. AI generates the name, description, steps, inputs, and trigger phrases.
4. Review every step in the visual step builder. AI can only use tools and domains that actually exist in your platform — unknown domains or tools are stripped server-side — but parameter values and step ordering should be verified.
5. Save when satisfied.

Running a playbook manually

1. Go to Admin → AI Playbooks and click a playbook name.
2. Click Run (the playbook must be active and have at least one step).
3. Fill in any declared inputs and confirm.
4. You are redirected to the run detail page. If a step requires confirmation, an approval request appears in your Inbox.

Setting up an automation trigger

1. Go to the Automation Triggers tab.
2. Click New Trigger, pick a playbook, and choose scheduled or event.
   • Scheduled: enter a cron expression. The first run time is shown before you save.
   • Event: pick a platform event name from the suggestions. Optionally add conditions (“Only if” rules) to filter which events fire the trigger.
3. Fill in the playbook’s required inputs. The trigger will not activate with missing required inputs.
4. Save. The trigger is active by default.

Reviewing run history

1. Go to the Run history tab.
2. Filter by status, playbook, or date range. Each row shows the playbook name, who ran it, status, step counts, and duration.
3. Click a run to see per-step results, dispatched parameters, and any error messages.
4. Use Export CSV to download the filtered history.
5. A failed run can be retried; a stuck run (Pending, Running, Waiting or Awaiting confirmation) can be cancelled.

Managing system playbooks

1. Scroll to the System playbooks section at the bottom of the AI Playbooks index.
2. To hide a system playbook from your users’ Ask AI, click Disable for this business.
3. To re-enable it, click Enable for this business.
4. To customize a system playbook, click Clone to customize — this creates an editable tenant copy.

5. Settings

Setting	Where	Default	What it changes
Active	Playbook detail page (toggle)	On	Inactive playbooks are not matched by Ask AI and cannot be run. Automation triggers bound to an inactive playbook will not fire.
Require confirmation (per step)	Step builder, per-step checkbox	Off	When on, the run pauses at this step and sends an approval request to the runner’s Inbox.
Auto-approve (per trigger)	Automation trigger form	Off	When on, confirmation-gated steps are approved automatically for this trigger’s unattended runs. Can be scoped to specific tools.
Trigger active	Automation trigger list (toggle)	On	Paused triggers do not fire on their next cron tick or event.

6. More help

• AI Operations — pause agents, review pending approvals, and monitor agent health.
• AI Consoles — per-solution control surface for autonomy and approval settings.
• Ask AI — the assistant answers questions about playbooks from these articles.