Agents now have access to new artificial intelligence tools tested by the General Services Bureau for the past eight months.
Through the USAI platform, agents can take advantage of features that include chat-based AI, code generation and document summaries.
“We are pleased to announce that we are committed to providing a range of services and services to providing services that are important to us,” said David Shive, GSA's Chief Information Officer. “GSA's business portfolio is very broad and we started applying this to these broad business domains. They started writing code that met the needs of code development that was completed in hours rather than days, weeks or months.
Shybe said employees were really beginning to show the impact of the USAI tool in a demonstration last spring Friday, led by then acting manager Stephen Ehikian.
Shive said 60% or 70% of these demonstrations are employees using USAI. “By reducing the drudge in daily work, you can either automate or augment the work they were doing, you don't have to circumvent five systems, or write code snippets to write the work,” he said.
GSA offers USAI as a No-Cost shared service.
Agents must sign memorandums with the GSA and decide which database to integrate with the large language model.
Shive said that agency customers can access USAI through a single-tenant cloud environment with GSA hosts.
“We don't have to do business operating technology for other agencies, not because it's bad, but because each agency's business mission is very different. The risk profiles for each agency are very different. “Each tenant, the underlying infrastructure, is all the same, but each tenant has developed an independent multi-tenant architecture. Each agency evaluates user provisioning and its behavior patterns, evaluates the behavior patterns of people and technology as a whole, and has complete control over the entire stack, including those based on the risk management framework.”
Access to 6 commercial AI models
Shive said some agencies were already interested in using the platform, but he wasn't ready to say who or how many.
Through USAI, GSA offers access to three different options and six different commercial AI leading language models, including Amazon, Microsoft, Google, Meta, Open AI, and Anthropic.
One of the services is a basic chatbot.
The second is an API layer that allows easy integration of datasets into chatbots. “The connections and mechanisms etc are the same for all tenants, but how they constitute them is up to the agency to decide,” Schieve said.
The third service is the console. This is an administrative dashboard that allows each agency to create tools and creation requirements based on risk tolerance, and more.
Shive said Usai has received authority to operate at a medium level of Fisma.
“There's the idea that you built something into something called a model garden. The idea of being there is the idea that not all models are the same. They all have different advantages and disadvantages, and we want to have access to the necessary models that you think are suitable for the job they're trying to do,” he said. “Knowing that these advantages and disadvantages are jumping over themselves at a very fast pace, it is a very dynamic industrial environment, which gives institutions the ability to try before purchasing on a large scale. In that rapidly changing, highly dynamic space, we can refer to the work we are trying to do with the best value model. We can fully visualize that telemetry and make decisions based on their internal standards, because it is the area of the Chief Information Officer to make those decisions to provide the tools that can do it.
GSA began construction of the USAI about 20 months ago, but its AI journey began more than a decade ago.
The agency has taken several important steps to reach the point where the USAI could emerge.
Use Case Monitoring Process
Zach Whitman, GSA's Chief Data Scientist and Chief AI Officer, said at the recent 930GOV conference sponsored by the Digital Government Institute, once generative AI became a repetition of its next capabilities. He said it gave employees the ability to start thinking about how to start using Genai tools as part of their daily workflow.
“One of the main things we understood was what people were afraid of. They were afraid of how they would perceive the use of tools as part of their work. They were afraid of the issue of data peeling. Am I putting the wrong thing in the tools? He said, “Not only did it technically understand this, it's also a clear mission for us culturally. There was a huge advantage in the policies we received from the White House.”
Whitman said that before GSA was widely offered across the agency, it set up several gates to jump over AI tools.
The GSA has created its own assessment factors based on its mission area and its own safety team consisting of technical and subject experts.
Whitman said the executive team took all this feedback and made a risk-based decision on which tools they could use.
Through this process, Whitman said that GSA believes the need for a popular AI platform has become USAI.
“What we did was build the platform using a lot of open source tools. Again, there was nothing so original. We ended up getting a chatbot interface and building an API layer that interferes with multiple hyperscalers, allowing us to access a variety of different models,” he said. “There was a way to measure the behavior of users on that system. The interesting part of putting together many models provides a really good opportunity to compare models based on use cases. I realized that certain models are better than others. Different models.”
Whitman added that GSA pilots and data show that Chatbot solves most of the use cases in which employees are interested in using AI.
He said the chatbot met the basic standards of user experience and did not require customization.
“What we do is to have interoperability between different models and ensure that we maintain the latest models as quickly as possible. So our agents have access to these latest models. Then we run the evaluation set to understand which models are best for which purpose. I said.
Copyright©2025 Federal News Network. Unauthorized reproduction is prohibited. This website is not intended for users within the European Economic Area.
