Before
Users would likely need to read each Slack request, determine whether it was a discussion, information request, or action, and decide whether it involved sensitive content. They would then check authorization, retrieve relevant thread history, compose a response when access was not permitted, or identify and hand off the request to the correct workflow.
How it works
- Capture a Slack bot mention, acknowledge the event, and retrieve the surrounding thread when conversation context is available.
- Classify the request by purpose and sensitivity, then apply the configured authorization decision for sensitive requests.
- Post an appropriate response for discussion or unauthorized requests, or route approved requests to the corresponding predefined subflow.
Workflow details
- Message intake: The workflow starts when a user mentions the bot in Slack, immediately acknowledges the webhook, and ignores retry events to help prevent duplicate processing.
- Conversation context: If the mention belongs to a thread, the workflow retrieves the full thread and adds prior messages to the request context so the analysis reflects the ongoing conversation.
- Request classification: The Gatekeeper AI Agent summarizes the request and classifies its purpose as discussion, information retrieval, or action. It also identifies whether the request involves sensitive information or activity.
- Access-controlled response: Discussion requests and requests that fail the configured authorization check are handled by an AI-generated message posted back into the Slack conversation, including an appropriate response for unauthorized submissions.
- Approved dispatch: Authorized or non-sensitive requests are interpreted by the Dispatcher AI Agent, mapped to a predefined subflow library, and forwarded with the relevant request context so the selected process can continue automatically.
What changed
Every eligible Slack mention is now analyzed and routed through the same automated gatekeeping process. Responses are posted directly in the relevant conversation, while approved requests are handed to a predefined subflow without requiring a person to interpret the request and select the next workflow.
Work removed
Reading Slack mentions one by one, retrieving thread history, classifying request intent and sensitivity, checking the configured authorization rule, composing access-denied or general replies, selecting the destination subflow, and manually posting the response or handoff.
Expected outcome
Each eligible mention receives one automated intent classification, sensitivity decision, and response-or-routing outcome, removing at least two recurring manual checks—request interpretation and destination selection—without claiming measured customer performance.
