Agentic AI is a type of artificial intelligence system that can accomplish specific goals with limited supervision through the autonomous orchestration of agents that mimic human decision-making. Virginia Governor Glen Youngkin recently issued Executive Order 51, which will launch the nation’s first use of agentic AI to empower statewide regulatory review. As Virginia leads the country in business attraction, regulatory modernization, and AI implementation, the governor’s latest executive actions will have ripple effects across the state and around the country. 

More on Agentic AI

While traditional AI operates within constraints controlled by humans, agentic AI employs autonomous, goal-driven behavior to perform specific tasks. Generative AI chatbots like ChatGPT can produce text, images, or code. Agentic AI systems take it a step further, using generated content to complete complex tasks autonomously. For instance, a generative AI chatbot can recommend the best times to take your car in for maintenance, but agentic AI can also book recurring appointments for you and add them to your calendar!

Some of the obvious advantages of agentic AI are its adaptability, intuitiveness, and ability to specialize in tasks in a human-like fashion. Virginia is capitalizing on this rapidly evolving capability to speed up its own processes, but how does it work?

First, agentic AI collects data from its environment in a process called perception. Then, using a variety of AI capabilities, including Natural Language Processing (NLP) and computer vision, it reasons with the data to extract meaningful insights and interpret the broader context of the data it is working with. Next, the AI sets various goals based on user inputs or predefined objectives. It also uses decision trees, reinforcement learning, and other planning algorithms to develop strategies that will help it achieve these goals. Once the strategies are developed, the AI evaluates each possible action and chooses the most optimal path based on accuracy, efficiency, and predicted outcomes -- stemming from “probabilistic models, utility functions, and machine learning based reasoning.” Finally, the agentic AI will execute on its choice, learning from outcomes and adapting its decision-making over time. 

Agentic AI’s autonomy is not only its greatest strength but also a considerable weakness. The nature of autonomous decision-making makes the AI more susceptible to ‘going off the rails’. Additionally, agentic AI systems make use of “reinforcement learning”: the pursuit of maximizing rewards. When a reward system for an AI is poorly designed, the AI might exploit loopholes to achieve ‘high scores’ in unintended ways. This issue is exacerbated when an AI attempts to optimize too aggressively without proper safeguards in place. 

In the context of Virginia’s attempt to implement agentic AI in its regulatory review procedures, a poorly designed AI system may be harmful. For instance, if the reward system compels AI to focus on streamlining regulations as much as possible -- targeting a specific percentage or number -- the AI may unintentionally remove or impact regulations just to maximize its rewards. A well-designed agentic AI system would not only help reduce regulatory burdens in Virginia but also prevent necessary regulations from being impacted.  

Regulatory Review

Virginia Office of Regulatory Management Director Reeve Bull claimed that the “Virginia model” for regulatory modernization is the new gold standard nationwide, and there is substantial evidence to support his claim. When Executive Order 51 was announced, state agencies had eliminated “47.9% of words in their guidance documents” and “streamlined 26.8% of regulatory requirements on the books.” These statistics demonstrate that most of Virginia’s agencies are on track to either meet or exceed the 25% streamlining goal, while the rest may potentially benefit from the assistance of AI. 

Many of these regulatory review efforts stem from Executive Order 19 (2022), which authorized a government-wide audit of regulations, permitting practices, and guidance documents. The order affected nearly 89,000 individual requirements. Notably, the Virginia Department of Environmental Quality reported a 65% drop in its regulatory timeline, while the Department of Transportation cut 20,000 regulatory items. Many environmental regulations had been drafted by agencies without any external review, and following the Order, stormwater regulations were reduced from 23 inches of printed material to just five. 

Regulatory review has tangible and direct impacts on citizens and small businesses in the state. For instance, updated building code rules — instated as a result of Order 19 — reduced the cost of new home construction by nearly 6% and saved homebuyers roughly $24,000. Brian Horn, the Director of the Department of Housing and Community Development, said more than 48,000 housing-related regulations were streamlined to achieve this outcome. 

Governor Youngkin says Virginia is now looking at a new 35% reduction target, and he hopes that AI will lead the charge. Agentic AI will power the regulatory reduction process by scanning all regulations and guidance documents, identifying ways they can be streamlined, flagging areas in which regulation contradicts statutes, identifying redundancies, and highlighting areas where language can be streamlined. By continuing to refine statewide regulations, Virginia hopes to continue being the model for regulatory modernization in the country. 

Other AI-Use Cases

Aside from Order 51, Virginia is modernizing with AI in many ways. Executive Order 30 (2024) established AI safety standards in various branches of the state.

The Order established policy standards for the ethical and transparent use of AI in government and business. Through mandatory measures such as approval processes for AI capabilities, disclaimers for AI-generated products, and measures to ensure data protection, Virginia is not only using AI to boost productivity and innovation but also firmly controlling it to make sure it does not go off the rails. 

The state also implemented information technology standards for AI and considerations for law enforcement. Furthermore, its education guidelines recognize the dual nature of AI, promoting integrity and partnership to use AI to empower students and augment their success. 

Finally, Order 30 established an Artificial Intelligence Task Force, consisting of the most prominent AI experts in the state -- drawn from educational institutions, non-profits, and private corporations. The Task Force will give recommendations to policymakers to strengthen its use of AI while responsibly guarding against risks. 

Replication and Repercussions

Virginia’s first-in-the-nation use of agentic AI for regulatory review is just one example of AI implementation across the country. Executive Order 51 will likely set a precedent for other states and encourage the careful use of this new technology. 

States have stepped up to fill in the gap the federal government has left through its, arguably, slow handling of AI. Consequently, new rules and uses of AI will continue to be seen soon. This new era of artificial intelligence governance poses exciting new possibilities for innovation and development, and states like Virginia are at the helm, spearheading change and charging full steam ahead.