Policy & Practice | Fall 2026

autonomously—or escalate it with a plain-language explanation of what needs human attention. n Real-time decision support and documentation. Agents can surface relevant policy, case history, and risk indicators when workers need them, while generating case comments and documentation in real time. This can support more consistent determina tions, faster response times, and reduced audit exposure. Agentic AI amplifies risk because it operates at workflow scale, not task scale. A generative AI tool that produces a flawed summary affects one document. An agent with access to a case management system, a document repository, and a client communication channel can propagate errors across an entire process before anyone notices. Two Strategies, Running in Parallel: Aligning Technology With Workforce Readiness When human services agencies invest in agentic AI without investing in the people and processes around it, they risk building solutions that look efficient on paper but erode in practice. Oversight weakens as fewer people understand what the agent did, rework grows as errors are caught later rather than prevented earlier, and outcomes become harder to sustain as institu tional knowledge shifts from staff to systems no one is fully accountable for. Sustainable modernization requires investment in systems that continuously sense, learn, and adapt as conditions change on two fronts simultaneously. People-enabled technology: systems designed from the start for traceability, accountability, and mean ingful human oversight. As agentic

override it. Without that under standing, issues may surface later through appeals, federal reviews, or customer complaints—undermining trust and imposing costly rework. The lesson for leaders is clear: agentic AI is not simply another technology upgrade. It is an oper ating model change that requires agencies to redesign work, oversight, and workforce capabilities together. Modernization succeeds not when people step out of the loop, but when they are equipped to intervene at the moments that matter. What Agentic AI Makes Possible in Human Services The capability profile of AI agents is meaningfully different from the tools agencies have used before. Unlike rule based automation, agents can handle unstructured information, adapt to changing conditions, and make context-sensitive decisions. Unlike stand-alone language models, they do not just generate content; they execute actions across systems and workflows. For human services agencies, signifi cant capabilities include: n Workflow automation across the case life cycle. Agents can manage multistep processes from intake through eligibility determination, connecting eligibility systems, document repositories, and com munication platforms. In practice, this can include pre-interview data aggregation, real-time valida tion during client conversations, and automated post-interview consolidation. n Adaptive exception handling. Traditional automation flags exceptions and stops. Agents can interpret an exception, evaluate options, and in some cases resolve it

Like autopilot, these tools can make complex work faster, more consistent, and easier to manage—but only when the people overseeing them under stand both their capabilities and their limits. Human services workers need to know not only what an AI agent reviewed, but what it did not review, the assumptions behind its recom mendations, and the circumstances in which human judgment should

Michael J.Walsh is a principal in Deloitte Consulting LLP’s Government and Public Services practice.

Tiffany Dovey Fishman is a senior research leader with

Deloitte’s Center for Government Insights.

Lindsey Harbison is a manager with Deloitte Consulting LLP.

Policy & Practice Fall 2026 22