Attempting to fully automate L1 support with agentic AI

Automating repetitive L1 requests has been a perennial journey for ITSM leaders. The latest wave of AI technology—particularly agentic AI—takes a significant leap further, offering the potential to completely offload predictable work from technicians.

Agentic AI refers to a system that allows you to orchestrate multiple AI agents to carry out work on your behalf. In this system, an AI agent is essentially an LLM-powered tool capable of understanding natural language instructions and executing the appropriate pre-approved actions.

In such an agentic AI setup, the service desk administrator takes on the role of a orchestra conductor. As the conductor, the primary responsibilities are:

• Clearly visualizing the desired outcome for the service.
• Meticulously assembling the components required to deliver it.
• Continuously training and refining the agents until they produce the exact results you expect.

In the ideal future, a routine L1 issue would barely impact the productivity of the user. Take a familiar situation: an employee reports, "I can't access the Finance folder on the shared drive, and I have a deadline in an hour."

With an agentic AI system, this request quietly triggers a workflow driven by multiple AI agents working in tandem. Behind the scenes, the workflow follows a simple, pre-approved logic:

• An information retrieval agent begins the process:
  • This agent checks the knowledge base and policy documents to confirm that the Finance folder access requires Level 2 clearance and membership in the Finance-Users Active Directory group.
  • It fetches the online status of the file server from the CMDB API.
  • It fetches the user’s role from the HRMS API and privileges from the identity management system API.
• A reasoning agent sequences the checks:
  • This agent confirms the file server is online from the information retrieval agent.
  • It validates the user’s permissions against policy and determines whether access can be auto-granted or needs approval.
  • It confirms the user’s role from the information retrieval agent.
  • It understands that user privileges are missing from the Finance group in Active Directory.
  • It validates the user’s senior role from the previous agent, confirming the Level 2 requirement, and decides it can be auto-granted.
• A tool-using agent executes the actions:
  • This agent adds the user to the required Active Directory group and grants folder access via a PowerShell script.
  • It updates the ITSM tool, resolves the ticket, and sends a success notification on Slack.

To ensure multiple AI agents work together seamlessly to deliver services, IT administrators must lay the right foundation. This groundwork defines what agents can access, what actions they can take, and how they operate safely within enterprise boundaries. Key areas include:

• Access privileges: Define exactly what each agent can access (knowledge articles, systems, enterprise applications, etc.) using agent-based permissions, and enforce least-privilege control through an integrated PAM solution.
• Repository of tools and scripts: Maintain an approved library of scripts, APIs, and actions so agents execute only trusted, sanctioned operations.
• Guardrails and controls: Keep AI agents safe, ethical, and predictable by using layered safeguards such as prompt-level guidance, model-level filters for sensitive data and hallucinations, and tool-level restrictions on system and API access.

With a strong foundation in place, a repetitive service can be automated. But the job doesn’t end there. Administrators must constantly watch for edge cases the system cannot handle and fold those exceptions back into the workflow. For instance, AI agents shouldn’t freely create or deploy their own fixes as it can cause major disruption.

A safer approach relies on deterministic scripts, strong guardrails, and AI agents mainly interpreting intent and triggering pre-approved actions. With ongoing training and oversight, however, fully automated L1 support is no longer far off.

Curious about how AI agents could automate service delivery workflows? Our industry research report explores their current role and future impact in ITSM. Get your copy here.