An IT service desk agent, introduced in three phases
How we would introduce an agent to a 15-person IT service desk handling about 1,200 tickets a month, starting read-only and earning the right to act.
This is a composite example built from common enterprise requirements. It is not a specific client engagement, and the figures are design targets for the scenario, not measured results.
Routine tickets crowd out real work
About 40% of tickets were routine: VPN problems, account unlocks, certificate renewals and access requests. Each took a few minutes of work but often hours of waiting, because engineers were busy with harder problems.
Tools and limits
| MCP server | Tools | Limit |
|---|---|---|
| Directory | Look up user, team, manager | Read only |
| Monitoring | Read VPN and device logs | Read only |
| VPN service | Renew device certificate | Allowed automatically, logged |
| Access management | Request and grant access | Manager approval required |
| Ticketing | Create, update and close tickets | Allowed, logged |
MFA settings, admin rights and deletions are not available to the agent at all. You can watch this design in the interactive demo.
Three phases
- Month 1, shadow: the agent diagnoses each routine ticket and suggests the fix. Engineers act and mark whether it was right.
- Months 2-3, low-risk actions: certificate renewals and unlocks run automatically. Everything else is drafted for an engineer.
- Month 4 onward, approvals: access requests are prepared and sent to the manager for one-click approval.
Targets and how they are checked
| Measure | Target | Checked by |
|---|---|---|
| Routine tickets resolved without an engineer | Most low-risk ones | Ticket data by category |
| Time to resolve routine tickets | Minutes instead of hours | Ticket timestamps |
| Harmful or wrong actions | None | Weekly review of the audit log |
What we would flag
- Engineers must review in month one. Without their feedback, there is no evidence to move to phase two.
- Managers must answer approvals quickly, or access requests simply wait somewhere new.
- Tools will need tuning. Expect to adjust tool descriptions and limits after the first weeks.
Thinking about AI agents?
Tell us the process you would like to automate and the systems it touches. We will come back with a plain view of what an agent could safely do, what should stay with people, and how to start small.