Recently, AI within software development has evolved from an experiment to a sought-after technology for working more efficiently. Yet, it can be difficult to make it grow further into the standard within your organization. Techzine spoke with Visma CTO Alexander Lystad about how AI is changing the development process, what strategies are being deployed and how the developer’s role is changing.
Lystad began his career at Visma as a developer, gradually working his way up to CTO. Before his current position, he was Director of Cloud & Engineering. For a company like Visma, such a position was crucial for guiding the hundreds of different software solutions under the brand through software architecture and cloud technology. From his background, Lystad, now as CTO, places a lot of emphasis on the next phase of software development, in which AI is and will continue to be a defining role.
Soon a reality
Visma began exploring AI applications within software development several years ago. A major turning point was the early and virtually overnight rollout of GitHub Copilot to many developers. GitHub Copilot was a logical first tool for broad rollout because it integrates directly into developers’ workflow. They get help writing code, often from the IDE development environment they use. “We have spent a lot of time since the rollout of the GitHub Copilot pilot project both implementing and training our people,” Lystad says. These early experiments were designed to test technological capabilities. “We test new tools first in a limited setting and only expand deployment after successful pilots.”
Initially, questions around compliance and the responsible deployment of AI were on the radar in this regard. This was to establish internally what was acceptable and unacceptable use, to manage risk and ensure a safe environment. This to build in safeguards around a controlled rollout. As a result of the testing, there are safeguards for development teams, making them more willing to embrace the new tools.
From a strategic perspective, it is essential not to see AI as a stand-alone project. At Visma, it has evolved into something utilized at every stage of software development, from planning and code generation to testing and maintenance. There is often a way to increase efficiency and raise software quality. But it has to make sense because while there are plenty of opportunities, it is sometimes overly presented as a solution to every development problem. Knowing what you gain from it, building guarantees and going for a gradual and controlled implementation seems to be the advice.
Grow into a standard
In addition to the technical challenge to be considered in the experimental phase, AI also has the human aspect to think about with immediate attention. Awareness and training are crucial as far as Lystad is concerned. Therefore, courses and webinars are offered within Visma. “We involve the best developers to share their experiences so that others see the value. In the beginning, there were many questions about compliance, and some of them were justified. We first established for ourselves what is acceptable and unacceptable use and then clearly communicated this within the organization. A lot is invested in courses, webinars and sharing best practices,” the CTO explains.
It was important to explain what was not allowed and, above all, emphasize what was possible. This stimulates experimentation. Once you get people enthusiastic, the ball starts rolling. Developers can then enthuse each other about using the new AI tools.
Getting people excited through training and other sessions applies primarily to top-down initiatives. For Visma, a bottom-up approach can also help. Then AI is not only introduced from the top down, but also developed and refined locally in teams. Successful approaches are then scaled up and shared within hundreds of companies under the Visma umbrella. It depends on the size of an organization whether such a bottom-up approach can actually work.
Also read: Increase 2FA adoption among employees and customers
Tool for modernization
Once the technical frameworks are set and developers are on board, you can start looking at additional ways to accelerate and improve development. Indeed, GitHub Copilot proved to be a good way for Visma to assist developers in writing code. However, the software company is constantly looking at other types of applications. Tools used internally for static code analysis, dynamic code analysis and CI/CD are all getting better in their way because of AI. In addition, Lystad sees an opportunity to identify better and manage technical debt thanks to AI. This technical debt refers to code in software that does not have the right quality standards, for example, because it was rolled out quickly to meet a deadline.
Lystad explains that AI offers an opportunity to address such technical debt. “We are working with the University of Oslo and other companies on a project that uses AI to identify technical debt in code. There are commercial tools for this purpose, but we are looking beyond,” Lystad said. “We are developing AI that not only recognizes technical debt in code, but also analyzes within backlogs. This helps determine from a backlog of thousands of items where technical debt is located, how important it is, and its impact.”
Visma’s CTO argues that such projects can also go further. Comprehensive detection of technical debt is primarily something that developers can benefit from at first glance. However, AI can also convert identifications into business-oriented insights. This makes technical issues understandable to non-technical stakeholders, which promotes decision-making.
Moreover, AI is proving effective in modernization projects, such as upgrading outdated tech stacks. Thanks to the ability of language models to “translate” different programming languages and optimize code, such projects can accelerate by sometimes as much as half. While AI is now less suited to developing entirely new functionality, its power lies in streamlining existing processes and improving software quality.
From code writer to product engineer
According to Lystad, all such developments will cause traditional code writing to fade more into the background. AI has that potential, but it is certainly not flawless. A developer’s focus is increasingly on strategic insight and translating customer needs into workable solutions. “It is no longer just about writing code, but about understanding product strategy, customer needs, and the core value of the software,” Lystad observes. This new reality calls for a hybrid role, best described as a product engineer.
This professional combines his technical knowledge with product strategy and customer value. Essential, then, is the ability to manage and optimize AI-enabled tools. This shifts the emphasis from purely technical competence to a broader perspective focusing on technology choices’ commercial impact. The AI can take over repetitive tasks and provide insights into large data sets – leaving the developer time to focus on the more complex and strategic issues. Loose lines of code can already be generated with AI, but picking up test cases independently will also become possible. The product engineer then controls all of this.
The enduring standard
However, despite the practical applications and potential already existing, there are challenges. A major concern is the need to monitor and adjust AI systems continuously. AI models depend on the quality of the data on which they are trained, and input errors can lead to suboptimal results. This makes regular evaluation and quality control indispensable. “We are still far from a situation where we can trust AI blindly,” warns Lystad.
In addition, ethical and legal issues continue to play a role. The deployment of AI raises questions of accountability and transparency. Organizations must establish clear guidelines and frameworks so developers and end users know how the technology is being used responsibly. Visma’s approach, combining internal regulations and external compliance agreements, can be an example for other companies.
As an organization, you must also be prepared for the change it brings. Transitioning to an AI-driven development environment requires a culture of continuous learning and experimentation. Investing in training and knowledge sharing is, therefore, a priority to maximize AI’s potential.
All in all, AI has become an integral part of development processes. Visma’s experiences, such as the early deployment of GitHub Copilot and the investment in training and compliance, illustrate how to grow from experimentation to a new standard. Staff must go along with that, who will eventually see their roles change. By investing in both technology and human talent, organizations can embrace the opportunities of AI and prepare for the challenges and opportunities of the future.