A Blue Robotic Owl Is Helping Israeli Students Catch Up on Missed Lectures

Key points

• Bar-Ilan University in Israel piloted an AI learning agent across 71 courses and roughly 2,400 students in February 2026.
• The agent runs inside Microsoft Azure (Microsoft's secure cloud computing service), built using Microsoft Copilot Studio with help from KPMG and OpenApp, and draws only from materials uploaded by each course's own teaching staff.
• The university plans to expand the tool to another 100 courses from October 2026, the start of its 2026/27 academic year.
• One lecturer saw average exam scores rise three percentage points after his students used the agent to generate their own practice tests.
• The agent works in Hebrew, Arabic and English, making it accessible across BIU's diverse student body.

Yarden Shay used to spend five to six hours a day studying after lectures. Now she spends about three. The difference is a blue robotic owl.

That owl is an AI agent, software that can carry out multi-step tasks on its own, embedded inside Bar-Ilan University's Moodle learning platform. When Shay missed classes in her first semester studying criminology and psychology, catching up meant chasing classmates for notes while new lectures kept piling up. "It's like a gap that is growing through time," she told Microsoft AI, which first reported on the pilot. The agent fixed that. All her course materials, recorded lectures included, sit in one searchable place.

She now uses it less for catch-up and more for active revision. She asks it open-ended questions, runs through multiple-choice quizzes it generates on demand, and gets told immediately when she's wrong and why. Then she shares her AI-assisted summaries with classmates who ask for them.

What does the agent actually do?

It answers questions in plain language, but it is strictly limited to whatever the teaching staff has uploaded: recorded lectures, transcripts, slides and research papers. It cannot browse the internet or pull in outside sources. That boundary is deliberate, and it matters for students who need to trust that what the agent tells them matches what they will be tested on.

Eliyahu Biton, a practicing lawyer who has lectured at BIU since 2017, spent weeks preparing his healthcare law and ethics course materials before the agent went live. He rewrote bullet-point slides as detailed papers, fed in video transcripts and tested the agent's answers until he was satisfied. His conclusion: if you want the agent to teach well, you have to teach it first.

The payoff showed up at finals in August. His fifty-odd students averaged three percentage points higher than in previous years. Biton is careful to note that could reflect a stronger cohort, not the tool alone. What he is more confident about is the change in atmosphere: students arrive at exams calmer because they have already put themselves through dozens of practice tests the agent generated.

His own role is shifting, too. Routine questions have dropped. He now holds four to six deeper discussion sessions a year rather than fielding basics. "Knowledge is no longer the problem," he says. "But what I can do better than a chatbot is to mentor."

What about privacy?

The agent runs entirely inside BIU's Microsoft Azure environment, meaning student interactions and course materials stay within the university's own secure cloud space rather than flowing to a public AI service. That is worth knowing before you assume a commercial chatbot and a university-grade deployment work the same way.

This fits the pattern we have been tracking across AI2Day's Microsoft coverage: the company is pushing Azure as the foundation for institutional AI deployments, while security questions around that infrastructure remain live.

For now, the owl stays in its lane. It supplements lecturers; it does not replace them. As Shay puts it: "It's something that goes with the lecturer, not instead of the lecturer."

Common questions

Can any university set this up?

The BIU agent was built specifically by KPMG and OpenApp using Microsoft Copilot Studio and requires the university to upload and maintain its own course materials. It is not a plug-and-play product available off the shelf, though the underlying tools are commercially available.

Is there a cost to students?

The source material does not give a per-student price. The pilot was a university-run initiative; students accessed it through the existing Moodle platform at no extra charge.

Could the agent give wrong answers?

It is grounded exclusively in staff-uploaded materials and cannot pull from the internet, which limits hallucination risk (the tendency AI tools have to confidently state things that are not true). Biton spent considerable time testing responses for accuracy before the pilot launched, precisely because students would be relying on those answers for exams.