If AI is writing the code, what exactly are we training software engineers to do?
by Emma Jones
Something pretty fundamental is happening to software engineering, and I’m not convinced our education and workforce systems have caught up with it.
For years, coding ability sat at the centre of what it meant to be a software engineer. But it seems that, increasingly, that’s changing.
Engineers now use AI to generate code, refine it, test it and troubleshoot it. Their role is shifting towards framing the problem, specifying what’s needed, interrogating the output and deciding whether what has been produced is actually any good.
That sounds like an evolution in tooling. I think it could be much bigger than that.
Specialist tech recruitment agency, Lookahead's new State of AI & Engineering in 2026 research gives us a glimpse of where this may be heading. Among its Australian respondents, product thinking and system design were the two skills most frequently identified as growing in importance. Communication also rose. Coding itself and deep knowledge of particular technology stacks were volunteered as skills in decline.
Now that in itself should make us rethink what we mean by "technical skills".
If AI makes generating code cheap and fast, the valuable human capability increasingly sits elsewhere: understanding the problem, breaking it down, communicating it precisely, making decisions about architecture, identifying risk, testing assumptions and recognising when the machine has produced something confidently wrong.
In other words, some of the capabilities the technology sector has historically dismissed as "soft skills" could become central engineering competencies.
And this creates a challenging question for education:
What happens when students can produce functioning code without developing the depth of knowledge previous generations acquired through writing, breaking and repairing it themselves?
Because there’s a second problem here - someone still needs to know whether the AI-generated output is safe, maintainable, secure and actually solves the right problem. Yet Lookahead found concerns about skill atrophy were far more prominent than fears of outright job loss. It also found that although code is being produced faster, checking it has not become correspondingly easier.
So perhaps we need to teach less towards production and much more towards judgement.
Give students imperfect AI-generated systems and ask them to find the weaknesses. Test whether they can explain trade-offs. Assess their ability to interrogate assumptions, work across disciplines and communicate technical decisions to people who are not engineers.
Then there is the junior problem.
The Tech Council of Australia's recent submission to the Parliamentary Joint Select Committee on AI points out that Australian labour-market data does not currently show widespread AI-driven displacement of early-career workers. It argues instead for greater AI capability, stronger technical training and much more real-world application.
But Lookahead raises the longer-term problem very clearly:
Much of the entry-level work through which generations of engineers learned their craft can now be performed by AI, and nobody yet has a convincing replacement for that apprenticeship pathway.
That may be one of the most important workforce questions we are not talking about enough.
How does someone develop senior-level judgement if the junior-level work that previously built it disappears?
And (of course!) there is a gender dimension here too.
Women have long been expected to bring communication, collaboration and stakeholder skills to technology teams while those capabilities were frequently treated as secondary to "real" technical expertise. If those same skills now become essential to directing, evaluating and governing AI-enabled engineering, organisations need to make sure they are recognised, rewarded and promoted accordingly.
Otherwise, we risk repeating a familiar pattern: the value of a capability rises, but the people who have historically been expected to provide it do not necessarily receive the status or reward that follows.
The conversation about AI and jobs is still too focused on which roles disappear.
A more urgent question may be:
If AI increasingly produces the technical output, what does it mean to be technically skilled?
Because if that definition is changing, we need to redesign education, hiring and early-career development now, not once the old pathways have already stopped working.
Author: Emma Jones is the Founder and CEO of Project F, a social enterprise working to improve gender equity in the technology sector. She is the creator of the T-EDI Standards®, Australia’s first independent benchmark for gender equity in technology workplaces, developed in partnership with the Tech Council of Australia.