Automation and AI
Build and publish an AI agent
Use Agent Studio to configure versioned tenant agents, grounded knowledge, tools, specialist routing, validation and controlled deployment.
For AI administrators and automation owners9 min readUpdated
Define the agent boundary
Give an agent a clear business purpose, allowed facts, required questions, prohibited claims, escalation conditions and enabled tools. A narrow specialist is easier to validate than one prompt that tries to control every workflow.
Use versions and validation
- Edit a draft workflow or create from a suitable tenant-safe template.
- Connect only required knowledge and tools.
- Validate the workflow and resolve blocking findings.
- Test representative and adversarial requests in the playground.
- Publish the approved version and record the change.
- Pause or roll back when production results regress.
Treat tools as business actions
A tool that creates an order, sends a message, checks payment or opens a ticket must enforce the active tenant and validate its inputs. The prompt should never be the only authorization boundary.
Related guides
Ready to use this in your workspace?
Sign in to continue setup, or review the tenant go-live checklist before sending live traffic.