How AI is reshaping software development
A quick question for a colleague? Increasingly, developers are turning to AI instead. Researchers at the University of Gothenburg are studying how generative AI is changing everyday work for software developers, revealing shifts in tasks, workplace relationships and perceptions of the profession.
Associate Professor Nataliya Berbyuk Lindström and doctoral student Arvid Lunde are studying how generative AI affects software developers working in large organisations. Since late 2024, they have conducted focus groups and followed developments from the early adoption of AI tools.
“When new tools are introduced, the way we collaborate may also change. AI can, for example, replace a colleague in certain situations. We are looking at how work is changing and which tasks are disappearing,” says Nataliya Berbyuk Lindström.
One change is already noticeable in how colleagues interact. Developers are less likely to turn to each other with simple questions, the kind that previously might have required a quick Google search or a brief conversation at someone’s desk. Now, AI can provide an immediate answer.
At the same time, the researchers have found that interaction between colleagues sometimes increases, as AI-generated answers lead to new discussions.
Whether this is a positive or negative development is difficult to say, according to Arvid Lunde.
“There are several aspects to consider. Much of what AI replaces is information that people could previously find through a Google search. The knowledge and experience built up within an organisation over time are much harder to replace.”
For new developers, AI may also change how they enter the profession. They now have access to a tool that can quickly help them move forward when they get stuck.
“AI makes it easier for new developers to acquire knowledge. At the same time, there is a risk that things move so quickly that they do not have time to develop a deeper understanding of what they are doing,” says Arvid.
This also affects mentoring. Some of the questions a junior developer would previously have asked a more experienced colleague are now directed at AI.
But reliance on these tools is not limited to newcomers. Arvid Lunde raises the issue of deskilling, the potential loss of skills when tasks become automated.
“These effects can also influence more senior developers. Our interviews show that they have become accustomed to using these tools. They feel it would be problematic if they suddenly lost access to them. The tools are becoming deeply embedded in their work.”
Developers are moving at full speed
At an organisational level, the researchers have found that developers are quick to adopt AI tools.
“The developers are racing ahead,” says Nataliya Berbyuk Lindström.
Meanwhile, management is trying to establish frameworks for their use through strategies and policy documents. According to Berbyuk Lindström, this creates a mismatch. The overarching documents can be abstract, while developers have already started using the tools in their daily work.
Arvid Lunde is particularly interested in the role of middle managers. Caught between management policies and developers’ practical work, they become a kind of gatekeeper, responsible for interpreting guidelines and making them work in practice.
The researchers therefore want to get closer to the actual work taking place within organisations and study how AI is used in practice. They also share their findings throughout the research process, including through workshops with developers.
Should students still learn to programme?
What is happening in companies also has implications for universities, according to Berbyuk Lindström.
“What does it take to be a software developer? This is highly relevant to informatics and software development. What knowledge and skills are needed? We have programming courses, should we keep them?”
As AI takes over more time-consuming tasks, the qualities that are valued are also changing. Berbyuk Lindström highlights creativity and innovative thinking in particular.
She also describes how companies are beginning to experiment with so-called pods – small teams bringing together different areas of expertise, where AI can carry out some of the work previously done by programmers.
But even if AI can write the code, someone still needs to be able to judge whether it is any good.
“Can we trust AI to produce good code? To assess that, you need to know what good code looks like. So the question is: should we remove coding and programming from our courses?” she asks.
The researchers also see a generational shift. Experienced developers have learnt their profession by carrying out tasks themselves, tasks that the next generation may increasingly leave to AI.
“Will the code and products become worse? Or will we simply get used to leaving coding in the hands of AI?” says Berbyuk Lindström.
Interview by Agnes Ekstrand