Imagine this: You, a veteran, see your doctor at your local VA hospital. After the visit, the doctor’s handwritten notes are transcribed, analyzed, tagged with certain injury or disease codes, and compared with your benefits plan to identify which services are covered and to what degree. All of it happens in seconds. And all of it is done by AI agents who then pass their findings on to a human.

These AI agents – collectively known as agentic AI – represent the next wave of generative AI. They are purpose-built AI tools that leverage large language models (LLMs) and use non-LLM based tools like search, database queries, email, etc., to perform specific tasks and often work in concert to accelerate workflows in really transformative ways. In our example above, an optical character recognition (OCR) agent turned a photo of the doctor’s notes to text; a summary of care agent analyzed and tagged them; and a benefits suggestion agent compared those tags to the patient’s benefits plan to determine coverage. A previously entirely manual process is now mostly automated. What used to take several hours takes just a few seconds.

So how is this different from the first wave of generative AI? Unlike single function chatbots or text generation tools, agentic AI features multiple collaborative agents that plan, act, and reason toward achieving more complex end-to-end goals.

This is the transformative potential of agentic AI. Its use cases extend across agencies and across missions. Similarly, collaborative AI agents could be used to build and test code to automate certain policies and approvals around, for example, Social Security benefits. Agents could be used to build and deploy data architectures more quickly and securely. Or, to create data science and machine learning models 10-times faster than we can today.

Where to Begin

Because agentic AI is widely seen as generative AI 2.0, it’s important for agencies to begin experimenting with it and investigating where it fits into their workflows.

1. Identify a Use Case
Like with any new technology or innovation, start by identifying a use case that makes sense for your organization and that is realistically achievable.

2. Develop a Proof of Concept
Next, design a proof of concept or proof of value that demonstrates the anticipated savings or efficiencies for the organization.

3. Review Data and Tool Access Policies
With a use case and proof of concept in place, be sure to review and enable your internal policies to ensure that the right data and tools can be accessed by the agents – securely, appropriately, and based on the specific needs of each use case.

4. Review Access Protocols by Use Case
Of course, as you review your access policies, be sure this happens on a use case by use case basis. Each use case will require access to different data sets in different ways.

5. Remove Internal Roadblocks
Be sure to also identify your decision makers, stakeholders, users, data, and tools, and articulate how they will work together across organizational boundaries.

6. Evaluate Your Infrastructure
It will also be important to review the infrastructure you will need for your chosen use case, because in all likelihood, your agentic AI use case will require GPUs and additional compute infrastructure.

7. Consider Security
As you’re building your new agentic workflows, ensure the right security protocols are in place. Here is an area where a zero trust architecture can help facilitate this type of experimentation by assuming no agent is trustworthy and validating access at every step.

8. Be Bold but Pragmatic
Finally, be adventurous in your use case search but know that not everything is an agentic AI use case. Agentic AI has generative capabilities beyond chat, and it's not just a look-up or an inference capability. Choosing the right use case, demonstrating success and building from there should be the goal.

Agentic AI’s benefit is speed to decisions. It brings efficiency at scale and enables teams to do things they didn’t think they could do before. By accelerating workflows and getting information to humans faster, agencies can make better, more informed decisions. In our VA benefits example at the start of this piece, agentic AI offered the possibility for the ever more efficient and effective use of data, leading to faster, better care. That’s delivering on the missions in entirely new ways for stakeholders. And that’s the kind of innovation that transforms how agencies serve their communities.