AI can make a task easier for one person and less usable for another. The difference is often decided before a model is selected: who helped define the problem, whose data is used and who has the authority to say no.

FIELD GUIDE

Co-design is a cycle

01 Listen People closest to the need define it
02 Build Test with permission and context
03 Check Review access, errors and unintended impact
04 Return Share control and revise the design
Participation at the start matters as much as testing at the end.

What the case compendium says

Jobs and Skills Australia's 2025 Gen AI case studies include inclusive design for disability and an app co-designed with Indigenous communities. The compendium also describes AI work connected to revitalising the endangered Ngalia language, with community leadership and cultural safeguards.

The public summary does not provide enough detail to copy those projects or claim particular outcomes. That limitation is important. Cultural knowledge and disability experience are not generic training data to be repurposed without authority.

Questions for a project team

Start by identifying the people affected and who can represent their interests. Agree what information can be used, who owns the resulting material, how consent can be withdrawn and who reviews failures. Pay contributors for their expertise where appropriate.

For accessibility, test the whole task with people using the assistive technologies and formats they rely on. A tool that drafts text can still exclude someone if the interface, review step or error recovery is inaccessible.

A respectful first experiment

Define one narrow goal with the people concerned. Prototype with synthetic or explicitly approved material first. Evaluate usefulness, unwanted exposure, errors and who controls the output. Do not put community language, personal histories or culturally sensitive knowledge into a model until the custodians have agreed the boundaries.

Want to explore how this could apply to your organisation?

Explore Private AI ↗