Policy & Practice | Fall 2025

to increasing degrees of autonomy in task sequencing and execution. Multi-agent AI extends the possibili ties even further by allowing multiple AI agents to work collaboratively toward a single objective. A primary AI agent breaks down the steps of a task and delegates those steps to other specialist agents, bringing a human into the loop when necessary. This collaboration can enhance quality by narrowing each agent’s scope, enabling high levels of specialization, and improving access to information, as agents can work off the same or dif ferent language models and interact with a variety of tools. The ability of today’s AI agents to orchestrate across different automa tion tools is inherent to its value. Agentic AI shares the same LLM foundation as traditional AI, but incorporates other technologies, such as robotic process automation (RPA), intelligent automation, and LLM/LMM chatbots, collectively boosting the power and value of these investments (see Figure 1). AGENTIC AI USE CASES FOR HUMAN SERVICES AGENCIES Human services agencies are begin ning to experiment with AI agents and multi-agent systems to transform pro cesses, improve efficiency, and expand impact. Use cases are wide ranging— from using intelligent workforce development agents to help job seekers identify skill gaps, suggest learning paths, and generate tailored résumés and cover letters for recommended roles, to using system moderniza tion agents to modernize decades-old legacy systems to updated, open source architectures more quickly. Here we highlight two areas ripe for Agentic AI-driven transformation: benefits management and contact centers. Benefits management: The appli cation process for state-managed programs can be cumbersome and confusing for residents. AI agents can alleviate this frustration by assisting with state portal account setup, benefits research, and applying for benefits using known information. They can also help with filling in required infor mation, interview scheduling, ensuring

errors are detected and corrected, doc umentation requirements are met, and residents receive real-time updates and alerts. Once eligibility is determined, AI agents can assist with benefits plan selection, enrollment, and ongoing case maintenance, leading to improved customer experience and higher self service rates. For caseworkers, AI can expedite the pre-adjudication process by screening initial applications, detecting errors, collecting missing information, verifying documents, scheduling inter views, suggesting next steps based on the latest data, and providing decision support for eligibility determinations. This allows caseworkers to focus on more complex tasks, increasing the accuracy of determinations and helping with timely benefits disbursements. Contact centers: Traditional customer support systems often rely on scripted interactions, which can fail to resolve complex or unique inquiries—leading to customer frus tration and escalation. This includes many chatbots that were implemented during the pandemic, which focused on high-volume but low-complexity interactions. In contrast, multi-agent AI systems can understand plain-lan guage requests and generate relevant and natural responses without being limited to a curated set of questions or use cases. These agents can converse in the language of the client’s preference, consider the customer’s history, prefer ences, and real-time context to provide hyper-personalized responses. These advanced systems can handle many complex inquiries effectively, reducing the need for escalation to live agents while improving customer satisfaction. GETTING STARTED GenAI tools are advancing rapidly, with no signs of slowing down. Forward-looking human services agencies are beginning to put AI agents to use. To get started on this journey, agencies should explore initial use cases and lay the groundwork for future advancements by considering these actions: 1. Assess and prioritize use cases: Begin with a comprehensive assess ment of current operations to identify high-impact areas where AI agents

Jamia McDonald , JD, is a Principal with Deloitte Consulting LLP’s Government and Public Services practice

Hari Murthy is a Managing Director in Deloitte Consulting LLP’s Public Sector prac tice.

Naman Chaurasia is a Senior Manager with Deloitte’s AI and Engineering practice.

Reid Exley is a VP of Product Management in Deloitte's Public Sector practice.

Tiffany Dovey Fishman is a Senior Manager with

Deloitte’s Center for Government Insights.

Policy & Practice Fall 2025 24