Outcomes
Examples
Use these patterns while learning this chapter.
Example workflow
New employee onboarding
HR approves a new hire and the company wants consistent access, tasks, calendar invites, and manager notifications.
Conditions
- Employee email exists
- Department equals Engineering
- Start date is within 14 days
Actions
- Create user invitation
- Assign position-based role
- Create onboarding tasks
- Send calendar invite
- Notify hiring manager
Safety checks
- Role assignment requires approval
- Invitation email is rate limited
- Tasks are created from a template
Example workflow
Project risk review
A project reaches 80 percent completion and leadership wants a risk summary plus a customer review meeting.
Conditions
- Progress is greater than or equal to 80
- Project status equals active
- Client contact exists
Actions
- Ask AI to summarize risks
- Email project manager
- Create client review meeting
- Post update to project group
Safety checks
- AI output is reviewed before client email
- Meeting invite includes only project participants
- No budget details are sent to members without permission
Step 1
Generate a draft workflow
AI Copilot can turn a natural language process into a workflow draft with trigger, conditions, actions, assumptions, and risks.
What to do
- Open AI Process Agent.
- Prompt: "Automate onboarding for engineering hires."
- Review the draft, questions, and sample payload.
Check before continuing
- The draft uses approved actions only.
- Write actions still require admin review.
Step 2
Run guardrail review
Guardrails identify risky actions like tenant suspension, role changes, bulk emails, invoice creation, and external webhooks.
What to do
- Run AI Guardrail Reviewer before activation.
- Read severity and recommendation.
- Add approval gates where needed.
Check before continuing
- High-risk actions are approved or moved behind human-in-the-loop nodes.
Step 3
Create a reusable template
Templates use variables such as manager, customer, projectBudget, or welcomeEmail so teams can reuse a workflow safely.
What to do
- Open Templates.
- Use guided setup variables.
- Save a marketplace-ready package when the pattern is proven.
Check before continuing
- Variables are documented.
- Default values are safe.
- The template starts in dry-run mode.
Pro tips
- Treat AI output as a draft, not an activated workflow.
- Keep successful workflows as templates after they run cleanly for a few weeks.
Avoid these mistakes
- Do not let AI bypass RBAC or write confirmations.
- Do not reuse a tenant-specific template without removing hardcoded IDs.