Raleigh Deploys Agentic AI to Support Lean IT Help Desk
Raleigh, N.C. has partnered with ServiceNow to launch Alli, an agentic AI assistant designed to resolve complex IT tickets and support its small municipal tech support team.
The city of Raleigh, North Carolina, has integrated an agentic AI specialist named Alli into its IT help desk to assist a lean team of four human technicians. This small unit is responsible for managing technical support for approximately 4,400 municipal employees and maintaining about 13,000 devices across the city. According to Raleigh's Chief Information Officer, Mark Wittenburg, a city of this size would typically require nine to twelve full-time IT technicians. Faced with persistent staffing constraints, the Raleigh Information Technology Department partnered with enterprise software company ServiceNow to co-develop and test this new artificial intelligence tool in a live municipal environment.
Alli, officially classified as an L1 IT Service Desk AI Specialist, represents a collaborative effort to build a scalable solution for public sector organizations. Rather than simply pointing users to instructional documents, the agentic AI is designed to resolve technical tickets directly. Because the tool was developed within a live production environment, it had to navigate real-world complexities that cannot be easily replicated in a laboratory setting, such as incomplete information, shifting priorities, and unexpected workflow exceptions.
From Routing to Resolution
The deployment of Alli marks an evolution from the city's previous AI implementation, known as "Ral-E." While the older Ral-E assistant was designed to guide employees through Raleigh's internal knowledge bases and route tickets to human staff, Alli is capable of completing entire workflows autonomously. Mike Hurt, ServiceNow's group vice president for the U.S. public sector, noted that the partnership aimed to create a product that could scale to help other municipalities and enterprises experiencing similar resource constraints.
To ensure the system continuously improves, ServiceNow implemented an "AI Orchestrator" that evaluates Alli's performance on each resolved ticket. When the agent receives a low score, IT administrators analyze the case to determine if the failure stemmed from a lack of data, insufficient system access, or limitations in the model's core capabilities. Wittenburg compared this process to training a human employee, emphasizing that the city did not immediately grant the AI admin-level permissions, but instead focused on gradual training, verification, and building trust.
Addressing Risks and the Need for Human Guardrails
The introduction of agentic AI into local government operations has also brought administrative and ethical challenges to the forefront. During a recent City Council retreat, councilmembers raised critical questions about the deployment of these automated systems. Councilmember Stormie D. Forte expressed concerns regarding the potential for autonomous systems to execute tasks without human approval, while Councilmember Megan Patton questioned the broader risks that these technologies introduce to city operations.
Wittenburg acknowledged these risks, explaining that Raleigh's early adoption of the technology allows the city to establish necessary controls before these systems exceed their intended boundaries. Currently, the primary guardrails keeping the AI within its programmed parameters are human administrators. The city's IT department is actively working with ServiceNow to develop technical safeguards that will eventually automate some of these governance functions. Wittenburg's long-term vision is to adapt this agentic technology for "Ask Raleigh," the city's public-facing portal that serves approximately 500,000 residents.
What it means for developers
The transition from simple retrieval-augmented generation (RAG) systems like "Ral-E" to fully agentic systems like "Alli" highlights a significant shift in how enterprise AI is built and deployed. For software engineers and system architects, this project demonstrates that the future of enterprise automation lies in multi-step workflow execution rather than basic information routing. Developers working on similar agentic systems must prioritize building robust evaluation frameworks, such as the "AI Orchestrator" used by ServiceNow, to audit and critique model outputs in real time.
Furthermore, the Raleigh deployment emphasizes the critical role of human-in-the-loop (HITL) architecture. Because unsupervised AI agents pose operational and security risks, developers must design granular permission structures and observational layers that allow human supervisors to easily intervene and adjust system boundaries.
To build and test these sophisticated multi-agent systems, developers often need to experiment with multiple foundation models to find the right balance of reasoning capability, speed, and cost. Through platforms like Apixo, developers can access top-tier AI models, including Claude, GPT, Gemini, and DeepSeek, through a single API key at https://apixoai.online. This streamlined access allows engineering teams to cheaply prototype, test, and deploy agentic workflows without managing separate integrations for every model provider.
Source: This New Raleigh, N.C., IT Help Desk Agent Is, Well, Agentic — GovTech AI. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key
