Universities are making consequential decisions about AIthrough familiar machinery: procurement processes, risk assessments, datagovernance and enterprise licensing. These processes are necessary. Theydetermine which tools staff can access, what information can be entered andwhere institutional controls apply.
They do not determine whether AI improves the work of the university.
That depends on what academics and professional staff dowith the technology. A researcher, grants adviser, reporting specialist andpartnership manager may use the same model in quite different ways. Each bringstheir own knowledge of the task, the constraints around it and the consequencesof getting it wrong.
This makes AI adoption more dispersed than a conventionaltechnology rollout. The institution still needs to provide suitable systems andestablish the rules. Much of the value, however, will be found by peopleapplying those systems to work they understand.
Universities have extensive experience implementingenterprise systems. A platform such as Workday or SAP is configured aroundestablished processes, roles and information flows. The aim is generallyconsistency across the institution.
Generative AI behaves differently. Its output changesaccording to the material, instructions and examples supplied by the user. Twopeople using the same model can produce very different results because they aresolving different problems or bringing different levels of expertise to thetask.
An enterprise licence can provide secure access withoutproducing worthwhile adoption. Staff may have little time to experiment,limited guidance on appropriate uses or no practical way to share what theylearn. A general training session can explain how the tool operates. It cannotidentify every task where the technology may be useful or anticipate theexceptions that make an apparently sensible application unreliable.
The procurement decision is therefore only one part ofadoption. Universities also need to consider AI as a workforce enablementtechnology. This means creating the conditions in which people can use itcompetently, question its output and improve the work around it.
Different parts of the university have different roles. ITteams set conditions for access, security and integration. People and cultureteams can support capability development, and institutional leaders setpriorities. Academics and professional staff will determine how the tools areuseful in their work.
An enterprise licence does not settle every future choiceabout AI. Models differ in their strengths, limitations and data-handlingrequirements. If the approved tool is poorly suited to a task, staff may turnto publicly available alternatives, creating a gap between official policy andactual practice.
This does not require an elaborate new governance structure.Universities can establish the privacy, information security and data-handlingrequirements that every approved tool must meet, then allow staff to testsuitable models against defined tasks within those boundaries.
Small trials can compare output quality, checkingrequirements and effects on the wider process. The people doing the work shouldhelp identify the models worth testing and judge whether the results areuseful.
AI is particularly capable at structured text production. Itcan prepare a plausible summary, organise an argument and produce polishedprose quickly. This has immediate value in universities, where written materialsits behind research, teaching, administration and external engagement.
Good writing once provided some indication of the effort orcapability behind a piece of work. That is no longer a given. A fluent grantapplication, briefing or research summary may conceal poor analysis, missingevidence or limited understanding of the subject.
Recent published research supports this distinction. Studiesof professional writing and consulting tasks have found that AI can improvespeed and the rated quality of work that falls within its capabilities.Research has also found poorer decisions when people rely on AI for tasks whereits performance is less dependable.
The difficulty is that the model does not reliably tell theuser which situation they are in. A weak answer can be delivered with the samepolished language as a sound one.
The value of human review depends on the expertise of theperson doing it. Two people can use the same model for the same broad task andget different value from it. The difference is not simply their ability towrite instructions. It is their understanding of the work.
An experienced grants adviser knows when an eligibilityquestion requires a more detailed or nuanced answer. An academic can see when asummary has missed an important distinction in the research. A reportingspecialist knows when apparently consistent data is based on definitions thatdo not align.
This expertise shapes how people use AI. It affects thequestions they ask, the material they provide, the assumptions they test andthe answers they accept.
As competent output becomes easier to produce, the expertisebehind it matters more. Universities will need people who can assess thereasoning, identify missing evidence and recognise when a plausible output iswrong. That judgement is what makes AI-assisted work dependable.
Universities need a way to learn from local experimentation.Teams should record where a model helped, where it failed, how much checking itrequired and whether the wider process improved. Those findings need to bevisible beyond the immediate team.
Leadership also needs to provide direction. If AI createsadditional capacity, what does the university want that capacity to achieve?
A grants team might provide advice earlier, before adeadline turns application development into a rush. Academics may spend moretime developing ideas, mentoring researchers or working with collaborators.Partnership teams may be able to follow up opportunities that would otherwisestall. Other staff may address recurring process problems that immediatedemands have repeatedly displaced.
These choices should reflect institutional priorities.Without that connection, AI may help individuals complete existing tasks fasterwhile leaving the university’s underlying performance unchanged.
Licence numbers and training completion rates provideevidence of access. Usage figures show activity. The more important test iswhether people can now do something useful that the institution previouslylacked the capacity to do.
AI adoption should be treated as workforce enablement.Procurement, security and data governance establish the conditions for access.Institutional value develops through the way academics and professional staffuse the technology in their work.
Useful applications will often emerge locally. Universitiesshould give people enough scope to test new practices while making the resultsvisible across the institution. Central teams should support and connect thatexperimentation rather than assume they can design every application inadvance.
Professional expertise becomes more important as competentoutput becomes easier to produce. Universities need people who can assess thereasoning, identify missing evidence and recognise a plausible answer that iswrong. Academic and professional expertise both have a place, depending on thetask.
The institution must also decide what it wants to achievewith any additional capacity. Faster work is not, by itself, an institutionaloutcome. Better advice, stronger research propositions, greater impact,improved teaching and more effective partnerships are more useful measures.
AI may enter the university through a procurement process.What happens after that will determine whether it changes the institution.
We acknowledge the Dharawal people as the Traditional Owners of this land and pay our respects to Elders past and present. This is, was, and always will be Aboriginal land.
