LearnfulThis fall, the University of Waterloo will be piloting Axiom across more than 8 confirmed courses, reaching up to 3,000 students. It’s the largest single deployment of the platform to date, and it comes with a research component attached.
What’s actually happening
Starting in the Fall 2026 term, instructors across multiple UW courses will have access to Axiom to build and deploy AI learning agents for their students, things like feedback agents for in-class activities, self-testing agents to help students prepare for assessments, and admin agents that answer routine course-logistics questions. Instructors choose whether and how to use the platform in their own courses; this is a pilot, not a mandate, and it reflects interest from individual instructors and departments rather than an institution-wide policy position.
To be clear about what this is and isn’t: this is a pilot deployment, not a formal partnership, and it should not be read as the University of Waterloo endorsing or vouching for Axiom as a product. UW’s role here is that of a host institution making the tool available to instructors who’ve opted in, alongside the standard scrutiny that comes with any tool touching student data.
How it’s set up on UW’s side
A few technical and governance details, since these tend to be the first questions we get from other institutions:
- Privacy and information risk review. Axiom went through the University of Waterloo’s information and privacy risk assessment process before this deployment, and cleared it.
- Institution-hosted, not third-party cloud. UW’s instance of Axiom runs on UW’s own servers, rather than on infrastructure we manage externally.
- Integrated with existing UW systems. The deployment connects to UW’s LMS and single sign-on, so students and instructors are using their existing university accounts rather than a separate login.
- UW’s own Azure environment for AI processing. LLM inference and document processing run through the university’s own Azure platform, using GPT-5.6 Sol, rather than a general external API.
That combination, institution-hosted, SSO-integrated, and running through UW’s own Azure tenancy, was a deliberate design choice on our end. Axiom’s architecture is built to sit inside an institution’s existing infrastructure and governance rather than ask the institution to route data out to us.
The research side
Alongside the broader course deployment, a subset of six STEM courses is part of a formally funded research project through UW’s Centre for Teaching Excellence LITE Grant program: Axiom: A GenAI Agent Integration Platform to Help Minimize Cognitive Offloading, running September 2026 through February 2028, led by a project team spanning Physics and Astronomy, Biology, Chemistry, Psychology, Optometry, and UW’s Integrated Teaching Support Unit.
The research asks a fairly pointed question: does constrained, pedagogically-guardrailed GenAI use support student learning, or does it just offload the thinking students are supposed to be doing themselves? The study will measure student performance, student perceptions, and instructor experience across the three agent types (feedback, self-testing, and admin) to try to find out.
You can read the full project description, team, and goals on UW’s Centre for Teaching Excellence site.
Why this matters to us
A pilot at this scale, 8+ courses and thousands of students in a single term, is a real test of whether Axiom’s architecture holds up outside a handful of early-access courses. It’s also, honestly, the kind of scrutiny we think a tool like this should get before anyone treats it as proven: a formal privacy review, real integration with institutional systems instead of workarounds, and an independent research study designed to measure outcomes rather than take them on faith.
We’ll share what we can as the term progresses and as the LITE Grant research produces findings.
Learn more about Axiom:
The Learnful team