The U.S. Department of Government Efficiency, or DOGE, is reportedly working on a chatbot dubbed GSAi to make federal workers more productive.

Wired on Thursday cited sources as saying that the chatbot is being developed for the U.S General Services Administration. The GSA is an agency with about 12,000 employees that manages federal buildings and vehicles. Additionally, it helps other federal agencies run procurement programs.

Initially, the GSA reportedly held talks with Google LLC about using the company’s Gemini chatbot to support its artificial intelligence efforts. Gemini is a chatbot powered by an eponymous series of large language models. According to Wired, the GSA decided to develop its custom GSAi chatbot after finding that Google’s Gemini didn’t meet the project’s requirements. 

It’s unclear what LLM will power GSAi. GSA might seek to develop a custom LLM or fine-tune an existing, perhaps open-source model, which would be faster and more cost-efficient. A third possibility is that the agency will adopt a proprietary cloud-hosted LLM. 

According to Wired, one goal of the GSA’s custom chatbot is to make staffers more productive. Additionally, DOGE plans to use the assistant to analyze data about the agency’s procurement agreements and other contracts.

Thomas Shedd, a former Tesla Inc. employee who now leads the GSA’s technology group, recently discussed the project in an internal meeting. He described GSAi as a “centralized place for contracts so we can run analysis on them,” Wired reported. Shedd detailed that “this is not new at all — this is something that’s been in motion before we started. The thing that’s different is potentially building that whole system in-house and building it very quickly.”

Earlier this week, The Washington Post reported that DOGE is using AI to analyze the Education Department’s internal documents. Similarly to the GSA’s chatbot project, the initiative is said to focus on contracts and other documents that contain financial data. DOGE reportedly intends to “replicate this process across many departments and agencies.”

According to the Post, the AI software being used to analyze the Education Department’s data runs in Microsoft Corp.’s Azure platform. This raises the possibility that DOGE might decide to build GSAi on Azure as well. The cloud platform provides access to Microsoft-developed LLMs, as well models from the open-source ecosystem and partners such OpenAI..

Besides boosting employee productivity and analyzing procurement data, the GSA’s AI push reportedly has a third goal. It’s believed the agency is seeking to make its software development teams more efficient.

As part of the effort, GSA reportedly sought to deploy the popular Cursor code editor. The software is based on Microsoft’s popular VS Code programming tool. Cursor extends the tool’s feature set with a chatbot that can generate code based on natural language prompts.

MuGSA’s technology team reportedly approved Cursor for internal use, but later reversed course and launched a review of the tool. The reason for the decision is that Cursor’s developer doesn’t have plans to achieve FedRAMP authorization in the near future. FedRAMP is a program that requires software makers to undergo cybersecurity evaluations before selling their products to the government.

In lieu of Cursor, GSA now reportedly plans to adopt Microsoft’s GitHub Copilot programming assistant. The latter tool relied solely on OpenAI models to generate code when it launched in 2021. Last year, Microsoft added support for other models including the LLM series that powers Google’s Gemini chatbot.

Photo: GSA