Artificial Intelligence (AI) is progressing so rapidly that new buzzwords are gaining popularity by the month. “Agentic AI” is the latest hot topic, especially in the public sector. While the terminology is new, it merely puts a label on how AI systems have been evolving for many years.
To understand what agentic AI means, let’s start with the basics.
AI Models: The Building Blocks for AI Agents
AI models are algorithms that analyze massive amounts of data, learn patterns and relationships to predict outcomes, make recommendations, and automate decisions faster and more accurately than human teams.
For example, the Southern States Energy Board (SSEB) uses AI to read, categorize, and summarize comments submitted by stakeholders and citizens. According to Ben Wernette, Principal Scientist and Strategic Partnerships Lead at SSEB, “We had a dataset of more than 40,000 comments about a federal action. We don’t have the horsepower to comb through these data and extract critical insights. We had a colleague start the process of manually attempting to comb through these comments. In the time it took him to get through about 5% of the data, we had output results from the AI models that summarized everything, and we were having active conversations about the data.”
AI Agents: Performing Repetitive Tasks on Behalf of a User
When AI models are used in an automated sequence with other models, rules, or advanced analytics to execute a workflow based on predefined rules and goals, the model is referred to as an “agent.”
For example, previously, nurses and medical coding specialists at a major U.S. healthcare provider had to manually sift through thousands of pages of patient records to determine coverage eligibility. Now, an AI agent processes over ten million pages per day and surfaces the most relevant data for human reviewers, accelerating decision-making and enhancing accuracy and consistency.
Importantly, AI agents operate on a spectrum of decision-making—from fully autonomous to human-guided—depending on the complexity of the task, the speed and accuracy needed, and regulatory requirements.
Col.Jon Lowe with the U.K. Ministry of Defense (MOD) uses AI to assist with job assignments of soldiers. He explained, “With HR information that’s highly sensitive, there’s particular care given about having the machine take over some tasks and having humans doing others.”
While the MOD has an AI agent read huge volumes of personnel information and weigh strengths, weaknesses, interests, and extenuating circumstances, a human oversees the output and final decision-making. AI gets assignment orders to soldiers faster.
AI agents that make real-time decisions without human intervention are described as “human out of the loop.” Agents that produce information needed for humans to act are described as “human in the loop.”
“The actual decision to appoint someone means you’re sending an individual and their family quite often halfway around the world,” explains Col.Lowe. “That’s a really important decision… so we have a human in the loop in that sort of decision.”
What Exactly Makes AI Agentic?
The terms “AI agents” and “agentic AI” are sometimes used interchangeably—but they have distinct meanings. AI agents are specific components designed to perform repetitive tasks on behalf of a user, while agentic AI is a broader framework that coordinates multiple AI agents to autonomously pursue and accomplish complex goals.
In other words, AI agents are the tools, and agentic AI is the system that uses those tools to think, decide, and act on its own.
Each time a person uses a credit card, agentic AI performs a series of complex tasks in near-real-time. It compares the characteristics of the sale to those of other common sales, and if the sale is consistent with normal buying, an agent approves the transaction and records it. If the sale is inconsistent, the AI agent declines the sale, initiates a text or email to the card holder, and records the declined transaction in its system.
With the many millions of transactions that occur at any given minute, it would be impossible for humans to perform this task manually. It would also be impossible for humans to use separate AI agents, manually passing information generated by one agent to another. Consequently, an agentic AI system is essential.
Using AI agents and agentic AI to tackle certain tasks enables humans to redirect their attention to more meaningful work. According to Dan Houston, Executive Manager, Data Science and Exploration at the U.S. Postal Service, “We’re trying to make employees more productive and put them on the tasks that we want humans to do. If your employees are uncomfortable about using agents, start having that conversation with them so they understand the opportunities they might be missing out on.”
AI Governance: Empowering Informed, Adaptable, and Trusted Outcomes
For AI to help organizations make better and faster decisions, people need to trust it. This is particularly true with AI agents and agentic AI which involve a greater degree of autonomy. Governance standards can facilitate this trust by helping ensure that AI systems deliver accurate outcomes, adhere to ethical standards, maintain data privacy, provide transparency, and align with an organization’s values and regulatory obligations.
“It becomes really important—how did the machine come up with that decision that fundamentally affects your life?” says Col. Lowe. “When AI is making a decision, we’ve got to be able to explain it fully to our people. You’ve got to be a lot more transparent about it. That’s where we are now with the use of AI.”
AI agents and agentic AI will continue to redefine how organizations, including governments, operate, make decisions, and interact with technology—but we must remember that it’s called artificial intelligence because it is the simulation of human intelligence in machines. These technologies are meant to augment the work we do, making it easier and faster to produce better outcomes for the people we serve.
“Don’t get hung up on the terminology of ‘AI agents’ and ‘agentic AI,’” says Houston, “Look around your environment and you’ll find that they’re probably already there.”
The author, Jennifer Robinson, is SAS’ Global Public Sector Strategic Advisor, working to help governments maximize the use of their data through data integration, data management, and analytics.