Latest in AI
Glean Outlook Email Triggers: How to Set Up and Test a Real-Time Agent
Outlook can start a Glean agent sooner, but a useful inbox workflow still depends on who activates it and which messages match.
Glean’s November 19 release switches Outlook email triggers from periodic checks to webhook notifications. New Outlook-triggered agents use the new delivery path by default, and eligible existing agents move automatically. If you already use one to triage a busy inbox, the first thing to check is which messages make it run.
Set up the Outlook trigger in Glean
Start with an Outlook connector that can see the mail you want to use. The Outlook connector guide describes separate setup paths for indexing and live retrieval. Check the connector with your administrator if the messages are missing. Finding email in Glean search is a useful access check, but it does not prove that an agent trigger is active.
In Agent Builder, add an Outlook New email content trigger. In Workflow mode, content is the trigger type on the initial step. In Auto mode, look for Based on content updates under Triggers. After publishing, open the agent in the library and activate it for the user who should receive the runs. A colleague with access to the same published agent does not automatically run it on their inbox. Activation and source permissions are separate for each user.
Start with one kind of email
Suppose an operations team wants to triage delivery exceptions. Start with a known carrier’s sender address and one recurring subject, rather than every message to the operations team. The content trigger guide lists sender, subject, recipient and other Outlook fields you can use to narrow the match. Verify your exact subject rule against a reply as well as a new thread. A changed subject line is an easy way for relevant mail to slip past a narrow rule.
The agent’s first job can be small: extract the shipment reference and stated delay, then prepare a short internal summary with a link to the original email if available. If the sender has not included an ETA, leave it unknown. For the pilot, keep customer replies and ticket updates in human hands. A fast trigger reduces the time until someone can review an exception. It does not make the carrier’s email accurate or authorise the agent to make a promise.
Test the filter and the delivery separately
Glean’s trigger tester lets you inspect a small set of recent events and run the agent against an example. Use it on a saved agent without validation errors. Confirm that a matching message supplies the expected context, then check a similar message you want excluded. Where your workflow only needs a few fields, map those values explicitly so you can see what the next step depends on.
Next, use a fresh test message with the pilot user active and open View run history in the agent library. A good sample result tells you the agent can process the input. A new recorded run tells you that the configured subscription picked up a real event. If a message is missing from history, check the user’s activation, mailbox access and filter before rewriting the prompt.
Measure time to review
Glean describes delivery as real time, but the release notes do not give a numerical guarantee for end-to-end processing. Track when the message arrived, when the agent ran and when someone could act on the result. Also count irrelevant runs. An agent that surfaces an exception quickly is useful; one that wakes up for every routine update just moves the inbox noise somewhere else.
I’m Paras. I work at the intersection of AI, engineering and distribution.
I help AI teams turn technical features into clear guides that readers can use at work.