At Bendigo Kangan Institute, 18 teachers support 4,500 students enrolled in literacy and numeracy support. That was the starting point for a Cogniti trial: giving students another way to get help within their courses, including when staff were unavailable.
At Manukau Institute of Technology and Unitec, growing use of AI agents raised a different question: once an agent is in a course, who checks that it is still doing its job?
Our latest Cogniti Connect session brought these two experiences together. Bendigo Kangan Institute was the 2025 Australian Large Training Provider of the Year Gold winner. Manukau Institute of Technology and Unitec received a Tertiary ICT Award specifically for the Cogniti AI governance solution shared in their presentation.
Bendigo Kangan Institute: what students’ questions revealed
Rebecca Laspatzis shared how the team developed course-specific Cogniti tutors to help students work through unfamiliar language, calculations and workplace communication. The tutors were designed to guide students through problems rather than simply supply answers.
Some applications gave students a chance to practise before working with another person. A role-play agent, for example, played a disgruntled student receiving feedback from a trainee educator. An assessment checker helped students identify what needed attention and directed them back to their learning resources before submission.
But the insights into the conversations, an important Cogniti feature, were useful to teachers, too.
When students repeatedly asked about the same concept or assessment question, teachers had something they could act on. Did an explanation need more work? Were the task instructions clear? Reviewing the conversations helped the team look at teaching and assessment as well as individual students’ support needs.
Rebecca also described how the team involved colleagues who were doubtful about the tools. They invited sceptical teachers to test the agents with difficult subject-specific questions first. Some became advocates.
Watch Rebecca’s presentation below.
Manukau Institute of Technology and Unitec: Taming Agents in the Wild West
You would not appoint a new employee without defining their role, checking their suitability and reviewing their work. So why build an AI agent, put it in a course and leave it unchecked?
That comparison shaped the approach James Oldfield, Rochelle Irving and Sharnell Aumua shared in “Taming Agents in the Wild West”. Their award-winning governance solution brings structure to the development and ongoing management of AI agents and mini-apps.
The team does not suggest that agents are employees. But when students are having conversations with an agent without a teacher in the room, someone needs to remain responsible for how it performs.
Their workflow follows a recognisable pattern:
- Give it a clear role. Identify the need, define what the agent should do and assess the risks before building it.
- Check it is ready. Work with subject matter experts, test with staff and students, and complete quality assurance before launch.
- Review its work. Monitor use and gather feedback to check whether it is doing the job it was designed for.
- Keep it up to date. Revise prompts and test again as models change, using feedback to develop improved versions.
That last step matters. An agent can still be running without working as intended. A change to the underlying AI model can alter its behaviour, even when nobody has changed its instructions.
At the time of the presentation, the team reported 81 deployed agents and more than 11,000 conversations. Their bespoke platform keeps each agent’s purpose, risks, approvals, versions and deployment locations in one place. It also helps them spot when an agent built for one course could be adapted for another.
The result is a practical way to make accountability part of everyday work, rather than something that ends with approval to launch.
Watch James, Rochelle and Sharnell’s presentation below.