Most educators have now used an AI chatbot. You type a question, it writes an answer, and you decide what to do with it. That model of AI, where a person asks and a machine responds, is quickly becoming familiar in staffrooms, classrooms and admissions offices.
The next step is different. Agentic AI does not just answer; it takes a goal, works out the steps, uses tools and data to carry them out, and checks its own progress along the way. This post explains what AI agents for education are in plain terms, what it could look like in a school, what the early research says, and the questions school leaders should be asking now.
You don’t need a technical background to follow it. The aim is a working sense of the ideas, so the conversations now starting in schools make sense from the first minute.
Build your school’s AI confidenceExplore AI courses
Agentic AI is AI that pursues a goal across several steps, rather than producing a single reply to a single prompt. Writing in Inside Higher Ed, Ray Schroeder describes these tools as “capable of reasoned assessment of what is needed to accomplish a goal”, lining up a chain of linked tasks and completing them with little supervision, much like a human assistant.
A simple way to see the difference is to compare how each would handle the same request.
| Chatbot (generative AI) | Agent (agentic AI) | |
|---|---|---|
| How it starts | You ask a question | You set a goal |
| What it does | Writes one response | Plans steps, then carries them out |
| What it can reach | What you paste in | Connected systems and data you grant it access to |
| Who does the next step | You | The agent, checking in with you where needed |
| Example | “Draft a letter to parents about the trip” | “Get the Year/Grade 7 trip organised”: checks the calendar, drafts the letter, sends consent forms, chases missing replies, flags dietary needs to the trip lead |
Three ideas sit underneath every agent. It has a goal rather than a prompt. It has tools, meaning permission to read from or act in other systems. And it has a loop: it acts, checks the result, and adjusts. The more tools and autonomy an agent has, the more useful it can be, and the more carefully a school needs to decide where a human signs off.
An agentic AI school is not a school run by robots. It is a school where AI agents take on defined, multi-step pieces of work across teaching and operations, while staff set the goals, grant the permissions and make the judgement calls. The examples below are illustrative, but each is already possible with current technology.
In the classroom
A teacher asks an agent to build a three-week unit. It drafts lesson plans against the school’s curriculum, creates differentiated resources and lines up formative checks for the teacher to review.
After a quiz, an agent spots which students missed the same concept, groups them and suggests a short reteach activity for the next lesson.
A tutor agent guides a student through a problem with questions rather than answers, and flags to the teacher when the student is stuck.
In school operations
An admissions agent follows up with families who started but didn’t finish an application, answers routine questions and books tours into staff calendars.
A trip or events agent handles consent forms, payment reminders and dietary information end to end, escalating anything unusual.
A reporting agent pulls attendance and assessment data together for a termly report and highlights what has changed.
In pastoral care
An agent watches for patterns across attendance, grades and behaviour notes, and raises an early alert to a form tutor or head of year.
The human decides what happens next.
The common thread is that agents handle the repeatable steps, so staff spend more of their time on the parts of school life that need a person: relationships, judgement and care.
The evidence on agentic AI in schools specifically is still thin, because the technology is so new. But research on the AI tools that came before it points clearly in two directions: AI can give teachers real time back, and how it is designed matters a great deal for student learning.
Time saved for teachers. A June 2025 Gallup and Walton Family Foundation survey of US teachers found that those who use AI regularly save an average of 5.9 hours a week, roughly six weeks across a school year. In England, an independent trial run by NFER and funded by the EEF gave 464 primary teachers in 108 schools access (or not) to Oak National Academy’s AI lesson assistant, Aila. Teachers using it spent about 24% less time on lesson planning, around 49 minutes a week, and their lessons were judged no worse in quality.
Design shapes learning. A study published in PNAS in 2025 gave nearly a thousand high school maths students access to GPT-4 during practice. Students with an unrestricted chatbot did 48% better in practice, but scored 17% lower on exams once the AI was taken away. A version built as a tutor, with safeguards that guided rather than answered, largely removed that harm. For agents, which can do far more on a student’s behalf, this finding matters even more.
Guidance lags behind use. Google’s November 2026 AI in Education report cites research finding that only 18% of US teachers have had formal guidance from school leaders on AI. Agents raise the stakes on that gap, since they act inside school systems rather than just producing text.
The takeaway for leaders: the productivity case is building, but the learning case depends on schools deciding what AI should and shouldn’t do for students.
The first real examples of AI agents in education are coming from a small but growing number of schools, colleges and national systems moving beyond chatbots. Most are early, and few have published hard results yet, but they show the range of approaches.
| Who | What they’re doing | What to note |
|---|---|---|
| Peninsula School District, Washington, US | Piloted AI agents with staff and plans to offer them to every adult employee, building “skills” staff can assign to agents | A whole-staff rollout, not a single-department trial |
| Milwaukee Area Technical College, Wisconsin, US | Faculty-controlled tutoring agents outside office hours; an agent now does most of the monthly enrolment reporting tied to state funding | A person checks every report before it’s filed, and agent access to systems is deliberately limited |
| Johnston Community College, North Carolina, US | The head of enrolment runs 12 agents that review student records across the enrolment cycle | Leaders say it needs internal champions and isn’t hands-off |
| Alpha School, US (13 campuses) | Two hours a day of AI-led academic instruction; adult “guides” focus on mentoring and motivation | The most radical model; results are school-reported, fees are around $75,000 a year, and critics question screen time and human connection |
| Estonia’s AI Leap | National programme giving 20,000 upper secondary students and 3,000 teachers AI tools from September 2025, expanding in 2026 | A system-level bet built on teacher training first |
The pattern so far: operations and admin are moving fastest, because the tasks are repeatable and a human can check the output. Classroom use is more cautious, and fully AI-led models like Alpha remain outliers.
Because agents act rather than just suggest, the questions a school asks before switching one on are bigger than those for a chatbot. Every early adopter above draws the same line: agents handle repetitive steps, and people keep judgement and final sign-off.
01
Human in the loop
Decide in advance which steps an agent can complete on its own and which need a person to approve. Anything touching grades, safeguarding, admissions decisions or communications to families should have a named human checkpoint.
02
Access and data
An agent can only reach what it’s given. Grant the narrowest access the task needs, and know where student data goes, who stores it and whether it’s used to train models.
03
Safeguarding
Agents that interact with students need age-appropriate limits, clear escalation routes to staff and transcripts teachers can review.
04
Learning, not shortcuts
The PNAS findings show that AI which does the work for students can undermine learning. Student-facing agents should guide and question, not complete assignments.
05
Staff confidence
Guidance and training need to come before rollout, not after. Teachers can’t supervise an agent they don’t understand.
06
Accountability
If an agent makes a mistake, the school is still responsible. Keep logs of what agents did and review them.
In development at Faria
What This Could Look Like in Practice
At Faria, we’re developing an AI admissions partner for OpenApply, called Faria Partner, around the same principle: the agent prepares, a person approves. Admissions staff will be able to ask questions in plain language, such as “Who needs a follow-up?” Partner will find the answer in the school’s own OpenApply data and prepare the next step: a batch of document reminders, a note on an applicant’s record or a rescheduled tour.
It’s designed so that nothing happens until a staff member approves it. Each change will appear on an approval card showing who it affects, what will be sent and what will be logged. Partner will work under each user’s own permissions, so it can’t see or do anything that person can’t, and approved changes will be recorded in OpenApply’s activity feed. It organises the work and makes no judgement on applicants: admissions decisions stay with the school’s people.
Agentic AI is the shift from asking AI for help to giving it a job. The schools that benefit will be the ones that decide deliberately what those jobs are. Before your school takes its first steps with AI agents for education, it’s worth thinking about:
Talk them through with colleagues, doubts included. That’s where the most useful conversation starts.
Sources
Teaching for Tomorrow: Unlocking Six Weeks a Year With AI · Gallup and Walton Family Foundation, June 2025
Lesson Planning Using AI Lesson Assistant: Evaluation Report · NFER for EEF, October 2026
Generative AI Without Guardrails Can Harm Learning · Bastani et al., PNAS, 2025
AI in Education Report: From Potential to Practice · Google, November 2026
Bridges 2026: How Schools Are Putting AI Agents to Work · GovTech, July 2026
The Rise of the Agentic AI University in 2026 · Inside Higher Ed, January 2026
School Says It Gets Better Grades With AI Instruction Than Teachers · WSMV, December 2025
AI Leap Programme Announcement · Estonian Ministry of Education and Research, February 2025
