Energy IT shop not interested in Grok, Perplexity as its AI portfolio expands

While the Department of Energy is working to expand AI model options for employees using its Joulix suite of tools, it isn’t looking to just add for the sake of more, per the agency’s  top technology official. 

“I could have an agreement with every vendor, every frontier model out there,” Dawn Zimmer, CIO at the Energy Department, said in an interview with FedScoop. “But which ones do people really want? Nobody’s going to use all of them.”

Some vendors aren’t making the cut. 

“There hasn’t been a huge demand for Perplexity,” Zimmer said. “We’re not probably going to add Grok at this point — I haven’t had a lot of demand for it.”

Perplexity was the first AI platform to sign a direct deal with the General Services Administration last November, enabling agency access through GSA’s Multiple Award Schedule rather than just via a reseller. Perplexity also received the FedRAMP Low authorization and is part of pilots at the Department of Justice and Labor. But the vendor’s pursuit for government customers hasn’t resulted in widespread use as of late. 

Grok, made by Elon Musk-led xAI, has similarly had a longer onramp to ubiquity in the federal government landscape compared to generative AI leaders OpenAI and Anthropic. The company has been met with concern by civic and public interest groups for its reported tendency to produce biased, inaccurate and concerning outputs. Even so, Grok has made some inroads, gaining the backing of the Department of Agriculture in its pursuit of a FedRAMP High authorization. 

DOE’s decision to expand AI model options grew out of the spat between the Department of Defense and Anthropic, which resulted in President Donald Trump directing agencies to phase out the AI startup’s technology. The series of events laid bare the dangers of resting AI efforts on the shoulders of just one vendor and underlined the need for redundancy and model diversity. 

While the clash continues to play out in court, federal agencies are in a holding position — especially those that were already using Anthropic’s technology, like DOE.  

“We’ll just wait to see how that plays out,” Zimmer said.  

The Department of Energy has tapped Anthropic for a number of use cases. Its Idaho National Laboratory, Lawrence Livermore National Laboratory and Pacific Northwest National Laboratory are piloting Claude for coding, according to DOE’s AI use case inventory. The agency is also using the AI tool to generate first drafts of documents and communication materials, as well as to search for agency information, per its consolidated inventory. 

“Claude … everybody wants it,” Zimmer said. “We were grandfathered in because we did purchase Claude in 2024 so we have some time to keep working with it.”

Initially, agencies that were using Claude at various levels had a six-month phase-out period, which would have wrapped up at the end of next month if upheld. However, a preliminary injunction has paused that process. 

In addition to Claude, Joulix offers DOE employees access to Google’s Gemini. The agency is actively working on adding an OpenAI integration, too. 

Employee demand for models is a large part of the Energy Department’s expansion playbook, but the agency is also using other feedback and criteria to help discern whether models are warranted. 

“Which one’s going to give them the best data to do the job they need to do here at Energy?” Zimmer said as an example of what she’s considering during the process.