Rediger

Del via

Workflow automation in Azure SRE Agent

Tip

  • Handle incidents end-to-end without waking anyone.
  • Run scheduled tasks automatically with human oversight when needed.
  • Apply knowledge from past incidents consistently every time.

The problem: manual handoffs slow everything down

Operational workflows span multiple tools and require someone to remember what comes next. You check status in one system, make a decision, execute in another, and notify your team in a third. Each handoff adds latency and risk.

How workflow automation works

Screenshot of the Agent Canvas showing a configured workflow.

By using workflow automation, you can achieve the following goals:

  • Create automated workflows that run on schedule or in response to incidents.
  • Use custom agents with specific tools for specialized tasks.
  • Build end-to-end flows that trigger, investigate, act, and notify.

When a trigger fires (scheduled time or incident), your agent follows these steps:

  1. Receives the trigger: A scheduled task runs or an incident matches a response plan.
  2. Invokes the custom agent: The configured custom agent starts with its tools and instructions.
  3. Executes the workflow: The custom agent investigates, takes actions, and coordinates with other custom agents if needed.
  4. Notifies your team: Results are posted to Teams, email, or your incident platform.

Each custom agent has access to specific tools (from connectors) and follows its instructions autonomously or with approval, depending on the run mode.

What makes this approach different

This section describes how workflow automation compares to other approaches.

Unlike scripts, your agent adapts when patterns change. Scripts break when inputs vary. Your agent reasons about what to do based on what it finds.

Unlike runbooks, your agent executes the workflow, not just documents it. Runbooks tell humans what to do. Your agent does it.

Unlike IFTTT-style automation, your agent investigates before acting. It doesn't blindly execute when a trigger fires. It assesses the situation and decides the appropriate response.

Before and after

The following table shows how workflow automation changes common operational tasks.

Before After
Check status in monitoring tool Agent queries automatically
Decide what to do based on data Agent reasons and proposes action
Execute fix in another system Agent executes via connected tools
Notify team in Slack or Teams Agent sends contextual notification
Log what happened Agent records actions in thread

Build a workflow

Workflows combine three building blocks.

Building block What it does Where to configure
Connectors Provide tools from external systems (Outlook, Teams, GitHub, PagerDuty) Builder > Connectors
Custom agents Specialized workers with specific tool access and autonomy settings Builder > Agent Canvas
Triggers Start workflows on schedule or in response to incidents Builder > Scheduled tasks / Incident response plans

For step-by-step setup instructions, see Step 5: Automate actions in the getting started guide.

Example: daily health report with email

This workflow checks Azure resource health and emails a summary.

  1. Connector: Add Send email (Office 365 Outlook).

  2. Custom agent: Create health-reporter with the SendOutlookEmail tool.

  3. Scheduled task: Attach to custom agent with this prompt:

    Check the health of Azure resources in prod-rg:
1. Query Azure Resource Health for any degraded resources
2. Check Application Insights for error rate trends
3. Summarize findings
4. Email the report using SendOutlookEmail

The agent runs this process daily, investigates, and sends the email with no manual steps.

Custom agent delegation

Use multiple custom agents when a workflow needs different expertise at different steps.

Step Custom agent Reason
Database diagnostics @DatabaseExpert Specialized KQL queries
Send notifications @Notifier Email and Teams tools
Create incidents @IncidentCreator PagerDuty/ServiceNow integration

The orchestrator delegates tasks to custom agents as needed. For more information, see Custom agents.

Best practices

The following table summarizes recommended practices for workflow automation.

Practice Why it matters
Test in the Playground first Verify your custom agent's behavior before attaching to a trigger
Start in Review mode Verify the agent's judgment before full automation
Test with "Run task now" Validate scheduled workflows before production
One tool per custom agent Easier to audit, debug, and update
Use descriptive names email-health-report vs custom-agent-1

Tip

Test custom agents in the Playground

Before attaching a scheduled task, test your custom agent:

  1. Go to Builder > Agent Canvas.
  2. Select Test playground view.
  3. Pick a custom agent from the dropdown and select Apply.
  4. Type your planned instructions in the Test panel and verify the agent executes them correctly.

Once you're confident in the behavior, attach the trigger.

Get started

Resource What you learn
Automate actions Build an automated health check with email notifications
Create a scheduled task Step-by-step tutorial for scheduled automations

Related content

Capability What it adds
Scheduled tasks Proactive monitoring and recurring task patterns
Execute mitigations Actions your workflows can take
Send notifications Notification patterns and channels
Incident response Response plan triggers
Custom agents Detailed custom agent configuration
Connectors Available tool integrations
  • Last updated on