Presented by Granicus in Partnership with Carahsoft
For someone seeking food assistance, health coverage, or a permit, government modernization becomes meaningful when it makes a task easier to complete. Can they find the right information? Understand what comes next? Get help when they need it? Those questions are central to how Michael Boyce, Senior AI Advisor for the State of Maryland, thinks about artificial intelligence and digital government.
In this episode of The Future Public Servant, Boyce describes an approach that connects the state’s public websites, internal workflows, and AI governance around a common purpose: improving outcomes for Marylanders. His focus extends from how residents discover government services to how public employees process applications and find the information needed to support them.
That perspective also reflects a change in where government’s digital front door begins. Residents may still visit an agency website directly, but they may also ask an AI tool to explain a benefit or help them navigate a process. As those interactions evolve, the quality of government’s underlying information becomes increasingly consequential.
Start With the People Using the Service
Boyce’s starting point is human-centered design: building technology with ongoing input from the people who use it.
Even a thoughtful planning process will miss something. A requirements document cannot anticipate every situation a resident encounters, and a system that appears straightforward to its designers may create obstacles for someone trying to use it.
“And the best way to improve that technology is to talk to humans, to talk to people, to talk to users about, hey, how is this working for you?” Boyce says.
That engagement can take different forms, including watching someone use a service, collecting structured feedback, or testing a proposed design. What matters is maintaining a feedback loop that reveals whether the technology helps people accomplish their goals.
For government, this changes how success is evaluated. Publishing information or launching a tool is a step in the process. Understanding whether someone can successfully navigate a service requires direct engagement with the people relying on it.
Boyce sees that principle as equally relevant as AI becomes part of the experience. The technology should remain grounded in what people need and how they actually interact with government.
AI Expands the Digital Front Door
Maryland’s websites provide access to services that touch daily life, including nutrition assistance, Medicaid, and permits. Improving those websites therefore remains a core part of improving government service delivery.
Boyce explains that AI adds another dimension to that work. People increasingly use existing AI tools to find and interpret information. Those tools may retrieve material from government websites and combine it into a response, placing a new interface between an agency’s published content and the person seeking help.
In the interview, Boyce reports that approximately 10% of Maryland’s website traffic in the preceding month came from AI systems retrieving information. He expects that activity to grow.
For agencies, that creates additional questions to investigate. Does an AI tool find the relevant government information? Does its response accurately explain the service? Does the resident receive the information the agency intended to communicate?
Boyce suggests examining realistic questions people might ask about a program and testing how different AI systems respond. That work extends the familiar responsibility of maintaining a useful website into understanding how its content appears through other channels.
The digital front door is becoming more distributed, and government needs to understand the experiences emerging around its information.
Website Fundamentals Support AI Access
Maryland’s preparation for that shift includes work already underway through the Maryland Digital Service team.
Boyce describes an effort to consolidate public-facing websites, improve accessibility, and ensure content remains relevant and current. Those improvements serve people visiting the sites directly while also helping systems that retrieve information from them.
Accessible websites often contain richer structural information and metadata. Boyce explains that these features can support search engines and AI tools as they locate and interpret content.
The connection gives website modernization a broader role in AI readiness. Clear information, accessible design, and consistent upkeep help establish a foundation for the interactions happening through emerging tools.
Boyce also encourages agencies to consider the capabilities residents already bring with them. In some situations, making government information work well with widely used AI tools may be an effective starting point.
That does not eliminate the need for purpose-built services. It does encourage agencies to begin with the public’s needs and the existing digital environment when deciding what to build or buy.
Look Beyond the Public-Facing Chatbot
Maryland is also developing direct applications of AI to support service delivery.
Boyce describes a chatbot that helps families navigate the process of applying for summer nutrition benefits. At the time of the interview, it had supported more than 30,000 conversations, with feedback he describes as positive.
But he emphasizes that AI’s contribution may also occur after a resident submits an application.
Benefits delivery can involve interviews, requests for supporting information, caseworker research, and multiple processing steps. For someone experiencing food insecurity, delays in those processes have immediate consequences.
AI could help improve parts of that work even when residents never interact with an AI interface themselves. The value would be experienced through a more timely or understandable service.
Boyce presents faster benefits delivery as an outcome agencies should work toward, rather than a result that follows automatically from adopting a tool. The question is where AI can make a meaningful difference within the actual process.
That requires examining the full service journey, including work that is less visible to the public.
Support Caseworkers With Reliable Information
One area of Maryland’s work involves helping caseworkers find information within large collections of documentation.
When employees work under time pressure, locating the right guidance can be difficult. AI search tools offer a way to retrieve relevant material more effectively, but their usefulness depends on whether the information they return is reliable.
Boyce says Maryland is pairing these tools with testing and evaluation. An answer that appears authoritative can still be inaccurate or fabricated, creating a need to assess performance carefully.
The state is also working on using AI to identify potential matches across program information so it can reach people who may qualify for summer nutrition benefits without realizing it.
Boyce describes privacy as part of that design challenge. Limiting the personal information used in matching also leaves less information available to establish whether records refer to the same person. AI can help identify possible matches, with a human reviewing them before the state proceeds.
Both examples illustrate the relationship between technology and public employees. AI can assist with finding information and identifying possibilities, while staff remain responsible for reviewing results within the service process.
Distinguish Productivity Tools From Specific Workflows
Boyce separates AI adoption into two broad categories.
The first is general workforce productivity. Like spreadsheets and other familiar workplace software, AI tools can support many everyday tasks. Maryland’s policies and governance framework establish expectations for their use as the state evaluates and expands access to productivity tools.
The second category involves AI applied to a particular workflow, such as assisting with benefits information or matching records. These applications require attention to the specific process, intended outcome, and consequences of an error.
That distinction helps clarify what an agency is evaluating. A tool used to summarize a routine document raises different questions from a system supporting decisions that affect access to benefits.
It also connects workforce adoption to institutional capability. Employees need useful tools, clear expectations, and support that reflects the work they are performing.
Match Oversight to Risk—and Support Implementation
Boyce sees risk-based governance as a strong foundation for responsible adoption.
Applying identical requirements to every AI application would be difficult given the range of technologies and uses involved. A risk-based approach allows routine activities to move through an appropriate level of review while directing greater scrutiny toward consequential applications.
He also wants agencies to consider potential public benefit alongside risk. A higher-risk proposal may warrant impact assessments, evaluation, continuous monitoring, and other safeguards. That designation should support a serious assessment of whether the work can proceed responsibly.
Maryland’s developing AI Innovation Lab provides another part of that approach. Boyce describes it as a source of implementation support that can help keep governance connected to practical experience.
Central IT needs to understand what agencies are trying to accomplish and help them work through the challenges of delivery. Lessons from implementation can then inform oversight.
Across the interview, Boyce returns to the people affected by these decisions: residents trying to access services and employees working to deliver them. Their experiences provide the measure of whether Maryland’s evolving digital front door is doing its job.
Three Key Takeaways
- Website modernization is part of AI readiness. Accessible, current, well-structured information supports residents and the AI tools they increasingly use to find government services.
- AI’s service value extends behind the scenes. Caseworker research, application processing, and outreach to potentially eligible families offer opportunities to improve the broader benefits experience.
- Governance and implementation need to inform each other. Risk-based oversight, testing, human review, and practical support help agencies evaluate opportunities and apply AI responsibly.
About The Future Public Servant
This interview is part of The Future Public Servant, a State Gov Today series presented by Granicus in partnership with Carahsoft. Through conversations with state government leaders, the series explores how artificial intelligence is changing public services, reshaping government work, and influencing the skills and capabilities public servants need for the future.
Explore more interviews at TheFuturePublicServant.com.