Define the business job
Name the outcome, trigger, inputs, owner and acceptable output.
How It Works
Turn a defined business job into a controlled agent workflow.
Choose the job, select a relevant agent role, connect approved apps, set rules and review permissions. AI Robot Agents is designed to give customers one guided product experience without requiring them to operate the underlying automation runtime.
how AI agents work
guided AI agent setup
controlled AI agent workflows
Eight stages
From business problem to monitored workflow.
Select an agent role
Compare catalog roles and confirm current availability for the intended workflow.
Connect approved tools
Choose the accounts and data sources required for the job.
Configure rules
Set scope, required fields, output formats and exception paths.
Review permissions
Apply least privilege and identify sensitive actions that need approval.
Run the workflow
Start with bounded work and review the resulting output.
Monitor and approve
Inspect activity, resolve exceptions and approve sensitive steps when required.
Improve over time
Refine rules and access only after the workflow proves useful and observable.
What changes after deployment
Agents take on execution. People keep ownership.
Modern agentic workflows can handle multi-step work across business tools. The practical design goal is not maximum autonomy everywhere; it is useful autonomy inside a clear scope, with permissions, logs and human escalation where the risk is higher.
Good first workflows
✓High-frequency and repetitive
✓Clear inputs and outputs
✓Tool actions can be permission-scoped
✓Exceptions can go to a person
Controls by design
Decide what happens automatically and what requires a person.
A low-risk step such as classifying an inbound request may run within defined rules. A sensitive message, consequential record change or financial decision should pause for review or route to an accountable owner.
Inputs and boundaries
Specify allowed sources, required context, prohibited actions and the conditions that end a run.
Approval and escalation
Place human review before high-impact actions and route uncertain cases with the context intact.
Activity and iteration
Use execution history and exceptions to refine the workflow rather than expanding autonomy by default.
Ready to put AI to work?
Explore specialized agent roles and choose a practical starting workflow.