Scheduled Tasks

TL;DR

• Issues caught before users notice — proactive monitoring replaces reactive dashboards
• Correlated insights instead of raw metrics — your agent reasons across data sources
• Describe checks in natural language — no scripts to write or maintain
• Create, edit, and manage tasks from the portal or chat

The problem

Operational tasks repeat. Every morning someone checks resource health. Every Monday someone pulls cost data. Every hour someone scans for anomalies. These repetitive tasks consume your team's time with predictable, automatable work — time better spent investigating real issues.

Traditional monitoring compounds the problem. Alert rules fire after a threshold is breached — by the time you see it, users are already affected. Dashboards show raw data but don't explain what it means. Each alert is isolated: your CPU alert doesn't know about the deployment that happened ten minutes ago. You correlate manually, across tools, every single time.

How scheduled tasks work

Your agent runs tasks on a schedule you define. Describe what you want done in natural language, set the frequency, and your agent handles execution automatically. Each execution creates a conversation thread where the agent plans its approach, queries data sources, reasons about findings, and produces an actionable summary.

This isn't a cron job running a script. Your agent uses its connectors, tools, knowledge, and memory to understand context. It notices that error rates are trending up 15% day-over-day even though they haven't hit the alerting threshold. It catches that storage usage will hit quota in three days at the current growth rate. It connects yesterday's deployment to today's exceptions.

Select Scheduled tasks in the left sidebar to manage all your tasks.

What makes this different

	Alert rules	Dashboards	Cron jobs	Scheduled tasks
When	After threshold breached	When you look	On schedule	On your schedule, before thresholds
What it shows	Single metric	Raw data	Script output	Correlated findings with explanation
Context	None	Whatever you configured	What the script queries	Cross-source, compared to baseline
Action	You investigate	You investigate	Whatever the script does	Summary with recommended next steps
Adapts	Static rules	Static views	Static scripts	Memory captures patterns over time

Unlike cron jobs, your agent understands natural language. You don't write scripts — you describe what needs to happen. Unlike runbooks, scheduled tasks execute automatically with the autonomy level you choose.

Before and after

Before	After
Check dashboards manually every morning	Agent checks proactively and posts summary
Correlate alerts across tools by hand	Agent correlates across all connected sources
Issues discovered after users report them	Trends caught before they become incidents
Write and maintain monitoring scripts	Describe checks in natural language
Each team member checks differently	Consistent automated checks every time
Need to change a task? Delete and recreate	Edit any task in place — execution history preserved

Task dashboard

The dashboard displays three key metrics at the top:

Metric	Description
Active tasks	Tasks currently enabled and running on schedule
Total tasks	All tasks including paused and completed
Total runs	Completed executions across all tasks

The task list shows each task with sortable columns:

Column	Description
Name	Task identifier — click to view execution history
Task status	On, Off, Ended, or Failed
Schedule	Human-readable format (e.g., "Daily at 8:00 AM")
Created by	Who created the task
Last run	Most recent execution time
Next run	Upcoming scheduled execution
Completed runs	Total successful executions

Editing a task

Modify any scheduled task directly — change the schedule, update instructions, reassign the subagent, or adjust run parameters. Your task's execution history is preserved.

Three ways to edit

Method	Steps
Toolbar	Select a task checkbox → click Edit task in the toolbar
Row menu	Click ⋯ on any task row → select Edit task
Execution view	Click a task name to open execution history → click Edit task

The edit dialog opens with all current values pre-populated. Change any combination of fields:

• Task name and instructions — update what the agent does
• Schedule — change frequency, time, or switch to a custom cron expression
• Response subagent — reassign to a different subagent
• Date range — adjust start date or set a new end date
• Message grouping for updates — switch between same thread or new threads per run
• Model tier — change the AI model tier based on task complexity
• Set a run limit — add, change, or remove the maximum execution count
• Agent autonomy level — switch between Autonomous and Review mode. When you select Autonomous, an info icon (ℹ️) appears — click it to review the Autonomous mode acknowledgement, which explains agent boundaries, AI model limitations, your responsibilities, and liability terms.

Click Save to apply your changes.

note

Save is disabled until you modify at least one field, preventing accidental no-op updates.

Model tiers

Each task can use a different AI model tier, letting you balance quality and cost per task. Select a tier when creating or editing a task.

Tier	Best for	Trade-off
Reasoning	Complex, multi-step analysis — deep investigations, root cause analysis	Highest quality, higher cost and latency
Fast Reasoning	Most tasks — health checks, reports, monitoring	Good reasoning with low latency
General Purpose (default)	Standard tasks — status checks, data retrieval	Balanced accuracy, cost, and speed
Fast	Simple polling, heartbeat checks, quick lookups	Lowest cost, fastest response, reduced accuracy

Cost optimization

Use Fast for simple recurring checks (resource health polling, metric queries) and reserve Reasoning for complex investigations. This can significantly reduce model costs for high-frequency tasks.

The model tier override applies only to the task — it does not change the agent's default model. Existing tasks without a tier set use the default (General Purpose).

Execution artifacts

When a task execution produces file outputs — CSV exports, PDF reports, chart images — those files appear as downloadable chips directly on the automation card. No thread diving required.

Artifacts are extracted from two sources:

• Markdown file links in the agent's response (preferred)
• Python execution outputs as a fallback when no links are present

Each chip is color-coded by file type: documents (cranberry), data files (seafoam), images (marigold), code (blue), and archives (neutral). Click any chip to download the file with one click.

Limit	Value
Artifacts per execution	20
Visible chips (collapsed card)	3 (with "+N more")
Visible chips (expanded / run history)	5 (with "+N more")

Expand the card or view the run history for per-execution artifact lists.

tip

If your tasks don't produce artifacts, update their instructions to include file generation — for example, add "export results as a CSV file" to the task description.

Example use cases

Use case	What the agent does
Daily health check	Reviews resource health, checks for degraded services, reports findings
Cost anomaly detection	Compares spend to baselines, flags unexpected increases
Security posture review	Checks for misconfigurations, expired certificates, open ports
Deployment verification	Verifies recent deployments are healthy after rollout
SLA reporting	Generates weekly availability and performance summaries

Example task prompts

Daily health check:

Review the health of all container apps in resource group prod-apps. Report any with restarts in the last 24 hours, memory usage above 80%, or error rates above 1%. Compare current error rates to last week's average.

Cost anomaly detection:

Analyze Azure cost data for my subscription. Compare today's spending rate to the 7-day average. Flag any resource group where spending increased more than 20%.

Get started

Resource	What you'll learn
Create a Scheduled Task →	Step-by-step tutorial

Related capabilities

Capability	What it adds
Execute Mitigations →	Take action when monitoring detects issues
Workflow Automation →	Chain tasks with triggers, custom agents, and notifications
Send Notifications →	How the agent delivers findings to your team
Run Modes →	Control agent autonomy per task
Connectors →	Access third-party observability tools