Elodie Collins
A top official from General Dynamics Information Technology said agentic artificial intelligence marks the second wave of generative AI. To prepare for the adoption of agentic AI, Dave Vennergrund, vice president of GDIT’s AI Center of Excellence, published a blog post that lays out eight steps federal agencies can take to incorporate the new technology into their workflow.
What is Agentic AI?
An agentic AI is a purpose-built, intelligent tool that utilizes large language models and non-LLM capabilities to perform a specific task. Different AI agents, according to Vennergrund, can work together to accelerate workflows.
He gave the example of a veteran consulting a doctor at a Department of Veterans Affairs hospital. Following a checkup, an AI agent will turn a photo of handwritten notes into digital text while another AI agent summarizes the document and tags relevant information for another agent to compare with the patient’s benefits plan. A previously manual process that takes a long time can be completed by a team of AI agents within seconds.
Vennergrund said the benefits of agentic AI extend beyond hospital settings. Its use cases can extend into different agencies and missions.
How to Adopt Agentic AI?
The journey of employing agentic AI in government begins with the identification of a use case. The GDIT official advised agencies to look for use cases that make sense within the organization and should be “realistically achievable.”
He also recommended the development of a proof of concept or value to validate that the introduction of the tool will lead to savings and efficiencies.
Next, the executive said organizations must review internal policies to enable agents to access the right data and tools securely. Moreover, Vennergrund noted that agencies must also review access policies on a use case by use case basis. He added that each use case will need a different access to individual data sets.
Once access policies are reviewed, the next step would be to remove any obstacles that would prevent the AI agents from performing their roles. This would include communicating with decision makers, stakeholders and users on how the new technology will have an impact across organizational boundaries.
Agencies must also have in place the computing infrastructure necessary to enable the agentic AI.
Vennergrund reminded agencies to consider security and recommended the implementation of a zero trust architecture at every step.
Finally, the executive encouraged organizations that wish to be adventurous and look beyond chat when searching for use cases.
