Automation
Automation basics¶
Automation runs work from a trigger, such as a schedule, a new file, or a button. An AI agent can make decisions and use tools within that work. A scheduled job may use an agent, a fixed set of steps, or both.
A weekly digest is a useful example: the schedule starts the job, the agent reads new notes and drafts a summary, and a check catches missing files before you review it.
From a task to a recurring workflow¶
We suggest starting with one output you can inspect. Compare several runs with the original material, including a run with a missing or awkward input. That will show which parts need a rule, a better source, or your judgment.
A playbook can hold the recurring instructions. The scheduler, tool permissions, and checks still need to be set up in the app that runs it.
What's actually worth automating¶
Look for work that repeats and has a clear result:
- Turn new meeting notes into an action list with links to the originals.
- Gather figures from a known set of reports into a draft summary.
- Label incoming files according to a naming rule.
- Prepare replies from approved customer information.
Ask whether you can tell a good result from a bad one without repeating the whole job yourself. If you can't, keep the work assisted until you have a workable check.
For exact tasks such as adding totals or copying fields, use a fixed rule or ordinary software where possible. Let the AI handle the parts that need interpretation.
Plan for a failed run¶
Decide what happens if a file is missing, a tool loses access, or a result cannot be verified. The job might stop, save an incomplete draft, or put the item in a review queue.
Make that state visible. A missing output should not look like a successful run.
Go deeper¶
- Plan a workflow: define the inputs, permissions, and result.
- Make it reliable: handle duplicates, partial work, and failures.
- Ideas to try: choose a first job with a clear review point.