Opinion
AI is changing the work engineers do and perhaps what we should expect of graduate engineers as they enter the profession.
Engineers have never been employed simply to produce answers. Ultimately, we are employed to make decisions.
For a long time, those two things sat comfortably together in engineering education. Learning to analyse a structure, model a system, apply a standard or develop a design was also part of learning how to become an engineer. Technical competence developed alongside professional capability, while judgement continued to deepen through experience in practice.
AI is beginning to complicate that relationship.
Engineers can already use AI to assist with analysis, coding, design alternatives, documentation and tasks that once took considerably more time. None of this makes technical knowledge less important. Someone still needs to know whether an answer is plausible, whether the assumptions make sense, what might be missing and what the consequences are if it is wrong.
But it raises a more interesting question than whether AI can do engineering work.
What happens to the engineer who is learning to do it?
In Australia, the four year Bachelor of Engineering (Honours) has long been the principal educational pathway into professional engineering practice. Graduates are expected to solve complex problems, exercise judgement, and understand risk, responsibility, and the broader consequences of engineering decisions.
But nobody becomes a finished engineer at graduation. Professional capability continues to develop through doing the work. Young engineers calculate, design, review drawings, encounter site problems, make mistakes and work alongside people who have seen similar problems before. Over time, they start to notice things they would once have missed. They become better at recognising which information matters. They learn when a calculation looks wrong, sometimes before they can immediately explain why. They see patterns. They become more comfortable with uncertainty and gradually take responsibility for more difficult decisions.
That accumulation of experience matters. There is value in doing a calculation yourself before learning to check one produced by somebody else. There is value in struggling with a design rather than immediately selecting between several generated alternatives. There is value in getting an answer wrong, finding the mistake and understanding why you made it.
Repetition matters too. After enough calculations, reviews, site visits and design decisions, an engineer starts to recognise what is ordinary, what is unusual and what deserves another look.
What appears to be routine work is not always merely routine. Sometimes it is apprenticeship.And that is where AI presents a problem that is easy to miss.
If AI increasingly assists with calculations, coding, preliminary analysis, option generation and documentation, graduate engineers may become more productive. They may be able to contribute to complex projects much sooner.
At the same time, some of the work being automated is also the work through which generations of engineers learned.
We could therefore find ourselves expecting young engineers to exercise greater judgement earlier, while giving them fewer opportunities to develop it the way previous generations did.
This is why I am not convinced that the most useful question is whether AI will replace graduate engineers. It may instead change what graduate engineers are there to do.
If a machine can produce an analysis in minutes, the engineer's task increasingly becomes deciding whether to trust that analysis. Were the assumptions appropriate? Does the model adequately represent the physical system? Is the result reasonable? What has been overlooked? What happens if an assumption fails?
AI can increase the amount of technical work that can be produced. It also increases the amount that someone must verify. That shifts some of the engineer's work from production towards evaluation. And evaluation requires judgement.
There is a tension here. The less time a young engineer spends producing certain kinds of technical work, the sooner they may be expected to judge work they did not produce themselves.
We cannot solve that simply by saying graduates need more judgement. Judgement is not another topic to add to a curriculum. It rests on technical understanding and develops through repeated encounters with uncertainty, competing constraints and consequences. Nor can we simply bring it forward by raising expectations. If we need engineers to exercise more sophisticated judgement earlier, we also need to think about how they will acquire it.
This comes at an interesting moment. Engineers Australia's new National Competency Standard for Engineering brings entry to practice and independent practice into a common architecture and makes engineering mindsets such as systems thinking, sustainability, risk and human centred thinking more explicit.
What interests me is not another list for universities to map into curricula. We have become quite good at mapping. The harder part is what happens when these different considerations point in different directions. A lower carbon solution may introduce another form of risk. A safer design may require more material. A technically elegant solution may be difficult to construct or maintain. A design may comply with the relevant standard and still be inappropriate for its context.
No competency framework can resolve those tensions for the engineer. At some point, someone has to bring them together and decide. That is judgement.
AI creates another problem on the other side of the entry-to-practice threshold: before graduates enter the profession, universities have to establish that they are ready to do so.
For a long time, a reasonably useful relationship has existed between the work a student produces and what we think that student can do. It has never been perfect, but much of assessment depends on it.
AI makes that relationship less straightforward. A student can produce sophisticated engineering work with considerable assistance. The analysis may be sound, the risks identified, alternatives compared, and the final report excellent.
What does that tell us about the student's own capability?
Keeping AI out of engineering education is not the answer. Engineers will use these tools, and using them well will itself become part of engineering competence. But a student's competent performance with AI is not quite the same as knowing what that student understands independently.
Universities increasingly need to establish both: can the graduate work effectively with these tools, and do they understand enough engineering to recognise when the output should not be trusted?
That makes the reasoning behind the work increasingly important. We may need to ask less often, What did you produce? and more often, Why did you decide that?
This brings me to the question underneath all of this.
What exactly are we saying when we award someone an accredited engineering degree?
We are not simply saying that they have completed four years of study or can produce technically correct work. We are saying they are ready to begin professional engineering practice.
There is a chain of trust in that statement, from universities and accreditation bodies through to the profession and ultimately the public.
AI is now putting pressure on both sides of that point of entry.
Before graduation, it makes it harder to infer individual capability simply from the quality of the work produced. After graduation, it may change some of the early career work through which that capability traditionally develops toward independent practice.
That is why I think the question is bigger than how we incorporate AI into an engineering curriculum. It is whether AI is beginning to change what we mean by being ready to enter the profession.
I do not think we yet have enough evidence to say the entry to practice threshold has moved. But there is enough happening around it to ask whether the capability expected at that threshold is beginning to change.
If graduate engineers move sooner from producing information to evaluating it, from following a defined method to questioning it, and from solving a bounded problem to helping decide what the problem actually is, then entry level expectations will have changed even if the qualification itself has not.
Universities will need to respond, but probably not with another subject called Artificial Intelligence for Engineers. Students may need to encounter ambiguity earlier and more often. They need opportunities to make assumptions, defend them, change their minds and choose between alternatives that are all technically credible. Industry experience may become more important, not less, because consequence, constraint and responsibility are difficult to reproduce in a classroom. And we may need to be much more deliberate about preserving the learning hidden inside the work AI now makes easier.
In an earlier piece, I argued that engineering has always been socio-technical. In another, I asked what success looks like in a profession where the best work is often invisible: the bridge remains standing, the water remains clean, the building remains safe, and most people never know the names of those who made it possible.
Judgement connects the two.
Engineering requires technical knowledge, but it also requires us to understand systems, people, risk and consequences. Society trusts engineers to bring those things together and make decisions.
We can teach equations, standards and software. We can teach students to use AI. But eventually every engineer encounters a situation where the information is incomplete, several answers are defensible, and the consequences are real.
Someone still has to decide.
The question is not simply whether our graduates can find the answer. It is whether they are ready to own the decision.
Because that is where the curriculum ends and the engineer begins.
Professor Zora Vrcelj is Head of Built Environment & Engineering at Victoria University