NEW YORK CITY — Agentic AI has quickly become one of education’s most talked-about technologies, but relatively few school districts and colleges have deployed AI agents in day-to-day operations.
At the Bridges conference Tuesday, however, education leaders shared examples of AI agents already running in production, from tutoring students and processing enrollment workflows to supporting staff with administrative tasks.
Their biggest divide did not appear to be whether K-12 districts and higher education institutions believe AI agents have value, but over how much autonomy and access to data they are willing to grant them.
DISTINGUISHING BETWEEN AN AGENT AND A CHATBOT
Early in a panel conversation, speakers drew an explicit distinction between AI agents and the chatbots many educators already know.
“Agents is one of the most abused terms in ed tech right now, and everybody seems to have a different definition of it,” Kris Hagel, CIO of Washington’s Peninsula School District, said.
Joe Siedlecki, chief impact officer at Amira Learning, an ed-tech company that builds AI-powered reading coaches and literacy screening tools, offered a distinction: A real AI agent makes decisions in service of a specific goal, whereas a chatbot just responds to prompts. In an educational setting, he said, an AI agent observes student outcomes and adjusts its instructional feedback based on those results.
“It’s making decisions pinned back to a specific goal, and the intent is to observe and change action,” Siedlecki said.
DIVERGING APPROACHES TO AGENTIC AI
Once panelists established a clearer understanding of agentic AI’s capabilities compared to chatbots, it became apparent they had differing strategies for how they implemented their agents.
For Milwaukee Area Technical College, protecting student data remains the guiding principle. Sadique Isahaku, the Wisconsin college’s vice provost and chief academic officer, said it has intentionally limited its use of AI agents while monitoring how the technology evolves.
“We are very, very careful about agents because they have potential to get into your systems, and maybe, we don’t know yet … collect data that probably shouldn’t go out to the public,” Isahaku said.
Rather than allowing agents broad access to institutional data, Isahaku said the college currently uses faculty-controlled agents from Gemini Notebook (previously Notebook LM), Google’s AI-powered research assistant, to provide tutoring outside of instructor office hours.
Johnston Community College in North Carolina has taken a different approach. Carrie Pitts-Densmore, the college’s vice president of enrollment and marketing, said AI agents are deeply integrated into administrative workflows to help scale enrollment operations amid growth and staffing constraints.
“We do use agents to look at student records, so that we can replicate what our employees do,” she said. “We couldn’t hire as quickly as we were growing, and so in order to scale that, we had to put these agents in motion. We … assign them jobs, and the goal is to move the student through the enrollment cycle from inquiry to completion.”
Similar pressures are shaping how K-12 districts are approaching AI agents. Some, Hagel said, are looking for ways to extend staff capacity amid ongoing financial and workforce constraints.
“I think every K-12 public school district right now is running low on money, and sometimes you have to try and accomplish some things and do that without either the systems or the people that are there,” he said.
Pitts-Densmore noted that, while at the Bridges conference, she currently has 12 AI agents performing tasks on her behalf.
Despite their different approaches to data access, Hagel and Pitts-Densmore described AI agents as tools for handling repetitive work while leaving human judgment intact.
His college’s monthly enrollment reporting process, Isahaku said, determines institutional funding based on student enrollment.
Because the work is repetitive, he said, an AI agent now completes most of the process, letting employees focus on more human tasks like relationship building and curriculum design.
“Eventually you get to a point where it can do all the majority of that work, and then leave the very last verification step for the human to do at the end,” he said. “When we’re talking $150 million a year, $60 million a year, you don’t want to screw up those numbers … you do want somebody just at the very end to verify those before we file them with the state.”
Hagel said his district began piloting AI agents last year and plans to make them available to every adult staffer this fall: “Right now, we’re working on building out the skills that we assign to the agents that users can use.”
REMAINING OPEN TO CHANGE
Panelists suggested the bigger challenge is no longer adopting AI agents — but preparing institutions, staff and students to use them effectively.
Leaders cannot simply deploy AI and expect it to be adopted successfully, Pitts-Densmore said: “It’s not ‘set-it-and-forget-it.’” Instead, they should identify internal champions who can encourage experimentation and iteration among the community.
“It’s also important … for you to have a culture where you can fail forward and where you are not going to be penalized for trying something and it not working, because it might not work, and that’s OK,” she said.
At Milwaukee Area Technical College, Isahaku said faculty adoption has been driven less by top-down directives than by instructors sharing successful classroom uses with one another.
“Faculty conversations are probably the most effective in getting the culture, getting everybody going,” he said. “Faculty members who have done [it] or have found innovative ways of augmenting the work they do, they come to showcase it … and how it is impacting students’ success and student learning in the classroom.”