Agentic AI for education: what it does, where it works, and what stalls
For a head of admissions or a registrar in the Gulf, the question is not whether AI is coming to education. It is which version actually works once the pilot ends. This is a plain account of what agentic AI does inside a real school operation, where it lands first, and why most education AI never makes it past the demo.
What agentic AI actually means in a school
A chatbot answers a question and waits. An agent does the work. Agentic AI describes software that takes an action, hands off to the next step, and carries a task to its end without a person driving each move. In an admissions office, that is the difference between a bot that tells a parent the open-day date and an agent that books the parent in, sends the confirmation, and flags the registrar when a place is at risk.
The unit is the agent, not the conversation. Each one owns a single function and passes work to the next, the way a good front office does: one person catches the inquiry, another books the tour, a third chases the form that never came back. Done well, the parent feels one responsive institution, not six disconnected tools.
Where it lands first: admissions and enrolment
Almost every school has the same leak in the same place. A parent inquires at 9pm, hears nothing until the next afternoon, and by then has booked a tour with the school down the road. The inquiry-to-enrolment funnel is where attention is thinnest and where speed of response decides who enrols. That is where agentic AI earns its place first.
An admissions agent answers the first inquiry the moment it arrives, day or night, in the parent's language, and carries it through to a booked next step. A voice agent calls the families who went quiet, before a human team would have found the time. Scheduling holds the calendar so tours and assessments fill without the back-and-forth. These three are live in education today. Applications, payments, and analytics are on the roadmap, not yet running, and an honest partner will tell you which is which.
Why most education AI stalls after the pilot
The pilot impresses a room. Then it never gets wired into how the school actually runs, and it quietly dies. This is the rule, not the exception. MIT's State of AI in Business 2025 study found that about 95 percent of enterprise generative AI pilots never reach measurable impact on the P&L, and traced the cause to an integration gap, not the models.
The same study found what the surviving 5 percent had in common: bought from specialist partners, not built in-house, and embedded in the daily workflow, not bolted on the side.
What separates the systems that work
The agentic systems that survive in education share four traits. First: they run against live inquiries, not as a proof of concept in a sandbox. The school owns the code, the data, and the prompts, with a documented plan for who runs the system after launch. Embedded in the real workflow, they read and write to the systems the admissions team already uses. And the only scorecard that matters is the enrolment number leadership already watches, not a vanity dashboard nobody opens twice.
White-labelling matters more than it sounds. Every agent should answer in the school's own name and voice, so a parent is talking to the school, never to a third party's product. The technology stays invisible. The institution stays in front.
The honest version of this carries a number that is not flattering. One early voice agent 36nine built connected on fewer than one in ten calls. We rebuilt it until it cleared more than one in three. That is what production looks like: ship, measure against the real number, rebuild what falls short.
The platform behind the agents
36nine deploys this as one coordinated platform, Arkan, with a named agent for each function: inquiries, voice, scheduling, and the applications, payments, and analytics agents on the roadmap. They run as a single system inside the operation, not as six separate tools a small team has to stitch together. Each deployment is white-labelled to the school's brand.
The UAE committed to this before most markets decided. A dedicated minister for artificial intelligence, a national strategy running to 2031, and a target of 14 percent of GDP from AI by 2030: these are stated government priorities, already shaping how the region builds. For a school in Dubai or the wider Gulf, the question is no longer whether to adopt, but how to adopt something that works.
What an education leader should do next
Start from a number, not a tool. Pick one enrolment figure leadership already watches: response time to a first inquiry, the share of inquiries that book a tour, the families who go quiet after applying. Ask where intelligence would move that number, and where it would not. A partner worth hiring will tell you plainly where AI does not help, because if everything is a yes, it is a sales pitch.
From there the path is narrow and clear. Scope one real number, build the agent that moves it, run it in production, and judge it against the figure you started with.
Common questions
- What is agentic AI in education?
- Agentic AI describes software that does work rather than only answering questions: agents that take an action, hand off to one another, and complete a task end to end. In a school, an admissions agent can answer a first inquiry, book the tour, send the confirmation, and flag the registrar, instead of a chatbot that only replies and waits.
- How do AI agents help with school admissions?
- They close the leak in the inquiry-to-enrolment funnel. An admissions agent answers the first inquiry the moment it arrives, day or night, in the parent's language, and carries it to a booked next step. A voice agent follows up with families who went quiet. Scheduling fills tours and assessments without the back-and-forth. These three are live in education today; applications, payments, and analytics are on the roadmap.
- Why do most education AI projects fail after the pilot?
- They never get wired into how the school actually runs. MIT's State of AI in Business 2025 study found that about 95 percent of enterprise generative AI pilots never reach measurable P&L impact, and traced the cause to an integration gap, not the models. The pilots that succeed are bought from specialist partners and embedded in the daily workflow.
- What is an AI workforce for a school?
- A set of agents that each own one function, inquiries, voice follow-up, scheduling, and coordinate as a single system inside the admissions and enrolment operation. The agents handle the repetitive operational work day and night, so the team spends its hours on the parents and decisions that need a person.
- Will an AI agent replace our admissions team?
- No. It carries the volume and the after-hours load the team cannot reach, the late-night first reply, the follow-up calls that never get made, the calendar back-and-forth, and hands the human work to the human. The aim is a faster, more responsive admissions office, not a smaller one.
- Is the AI branded as the school or as a third party?
- As the school. Every agent is white-labelled to the institution's own name and voice, so a parent is always talking to the school, never to a vendor's product. The technology stays invisible and the institution stays in front.
- How should a school start with agentic AI?
- Start from one enrolment number leadership already watches, not from a tool: response time to a first inquiry, the share of inquiries that book a tour, or the families who go quiet after applying. Decide where intelligence would move that number, build the agent that moves it, run it in production, and judge it against the figure you started with. A strategy session with 36nine scopes that first number and the plan behind it.