The State Department wants to have artificial intelligence agents that can take action for employees, the department’s chief information officer Kelly Fletcher said Wednesday.
The department already has an enterprise generative-AI chatbot, dubbed StateChat, which it launched last year. That chatbot can help with translations or answer questions from the department’s foreign-affairs manual, Fletcher said at an ACT-IAC event Wednesday.
Now the department is looking at “AI agents that will take actions for humans,” said Fletcher. “I want it to not only tell me, ‘How much leave do I have’ … but then I want it to put in my leave slip, which is in a different system. We’re building to that.”
That action-taking is what distinguishes AI agents from generative AI. It’s something that the AI company Anthropic, behind large language model Claude, is zeroed in on.
Co-founder Jack Clark said Monday during a D.C forum that, by the end of 2026 or early 2027, Anthropic expects to build systems that “won’t just passively answer questions,” but can “be given tasks that take hours, days or weeks to complete and then go off and do them autonomously.”
Although agentic AI offers the potential to help automate operations and increase productivity, the technology also comes with risks, including oversight challenges, difficulties in testing and evaluation and the potential for job displacement.
The government’s chief information officer, Greg Barbaccia, has said that he wants to use AI to help make up for losses across the federal workforce as the Trump administration has shed thousands of workers. Among them is State’s own former chief data officer and AI officer, Matthew Graviss, who left the department in February after over four years working there.
Despite AI’s potential, Fletcher said that adoption hasn’t necessarily been easy at State.
The department initially rolled out its chatbot to 3,000 beta testers. Now it is being used by 45,000 to 50,000 of the department’s 80,000 workers, said Fletcher, who noted that “it has taken a huge amount of education and training” to get those users.
“One month ago, I answered the question, ‘Is it allowable for me to use it?’” she said. “Something I wildly underestimated with AI is the amount of training and education and conversation required to get folks who would benefit greatly from it to use it.”
The chatbot can help State employees navigate internal policies.
“If you need to know how to move your cat with you to Conakry,” she offered as an example, the chatbot will show you “all the locations [in State policies] that explain how to move a cat.”
“Then we’ll let you click on them and read the actual text or give you a summary,” she said. “The idea here is, in large part, to reduce administrative toil.”
For agents, Fletcher said that her goal is to put the department’s administrative functions behind one chatbot and consolidate other potential agents around certain mission sets.
“Looking forward, I think that AI is going to be embedded in just about everything,” said Fletcher, offering the potential for AI to prioritize cybersecurity alerts as an example. State is also testing a chatbot to help users navigate its electronic health record patient portal.
“I think the trick is going to be, ‘How do we embed it smartly, and how do we ensure that people know what to use it for?’” she said.
Read the full article here