In one line
Self-improving AI agent that triages support tickets and auto-fixes known issues.
Context & problem
Recurring tickets in the internal support portal, with repetitive fixes: running a script, changing Teams policies, simple user support.
A ServiceNow incident or service request triggers the agent. The ticket is loaded into its context and assessed for triage or a fix. The agent makes tool calls to several external systems (e.g. Entra ID, Teams, phone API) to enrich the information. It suggests an action or fix, creates a test case for validation and sends an approval request (L2/L3) to the engineer. After execution, the test validates the result, and the results and actions taken are written back into ServiceNow. Tickets the agent can't solve yet are assigned to the engineer with the enriched information. Once a ticket is closed, the agent learns from the actions taken or from the cause and resolution of the incident.
What I did
- Agent reads tickets through the ServiceNow API and runs locally or in Azure AI Foundry
- Executes fixes via PowerShell (Teams policies, AD, Exchange), Power Automate and Graph API
- Safety level: auto-fix known patterns, human approval for new ones or with low confidence
- Learns from my own fixes – resolutions feed back into its knowledge base
- Also used as the use case for a study assignment (Modul B)
Result / impact
10% of tickets auto-fixed and growing every week; at least 1 hour saved per week. Stability and validation accuracy are still being improved.
Details
- Processes automated: ticket triage, Teams policy changes
- Compliance & data protection: human-in-the-loop for anything new, transcript & audit log for every action. Compliance & IT Security approval pending, since still in PoC phase