Cubeia, a software company based in Sweden, has reached the third phase in its journey toward AI-assisted development, transitioning from letting developers use AI on their own terms to establishing a structured AI-driven development process. This shift prompts a critical question for the organization: how to function when coding is no longer the primary barrier.
"We’re doing this 100%," asserts COO Stefan Grenstad regarding the transition to an AI-driven development model.
The focus has now shifted to how the company can maximize the capacity that AI has unlocked. Only six months prior, Cubeia’s efforts centered on substituting human-written code with AI-generated alternatives. Now, the company has evolved to introduce a distinct development pipeline, redefine the roles of developers and quality assurance (QA), and foster a new relationship with clients.
In its initial phase, Cubeia adopted an open approach, allowing developers to use AI as they wished. The second phase imposed a structured framework where all team members relied on the same AI agents and operated through a unified pipeline. This transition necessitated addressing issues around quality and reliability, as well as enhancing comfort among employees with this new working method. Grenstad feels confident that these challenges have largely been addressed.
"During the hybrid period in Q1, Cubeia solved 259 issues. After fully implementing the AI-driven process, that number jumped to 421—a 62% increase. The number of larger projects rose from 17 to 58. So clearly, our shift to AI-driven development has significantly boosted our output," Grenstad explains.
However, the more pressing question concerns how Cubeia will utilize that newfound capacity. Grenstad remarks, "Because we’re not dedicating as much time to coding, we’re focusing more on understanding business value and the domain.” Teams are now expected to grasp not just what they are constructing but also the value it is meant to produce.
Paul Crisp, Cubeia's head of marketing, emphasizes the importance of measurement in achieving objectives. “If our goal is to increase traffic by 10%, we have to establish a baseline, implement the changes, and evaluate the outcomes. If traffic only rises by 1%, we may need to iterate again to reach our objective.
As the company evolves, customer input is increasingly shaping the development trajectory. Grenstad mentions thatCubeia recently began asking customers what they could achieve with AI on the platform. In fact, one client built a customized casino landing page and sought to utilize Cubeia’s APIs for integration. Another client aimed to enhance its offerings using Cubeia’s player account management system to self-build functionalities like personalized bonuses.
The aim is not for Cubeia to deliver another AI tool, but instead to facilitate customer access to their own AI infrastructures utilizing Cubeia’s platform and data. They are currently testing a pilot project inspired by the landing-page example.
This proactive communication with clients marks a departure from the previous six-month outlook. Grenstad is optimistic that the company can now confidently respond affirmatively when customers inquire about integrating their AI solutions with Cubeia’s capabilities. “We’ve realized we will be part of this evolution,” he states.
Cubeia’s burgeoning capacity is also prompting reorganization within its development teams. Grenstad reveals that the company is experimenting with a rapid-response team, addressing smaller customer needs while another group focuses on larger projects and strategic planning. Current teams consist of five to eight members, but Grenstad envisions moving toward smaller units. “In my ideal scenario, teams would comprise just two or three individuals,” he adds.
As developers transition between rapid-response and long-term projects based on their skills and interests, the increase in capacity has not fully eliminated bottlenecks. With AI agents generating multiple streams of work simultaneously, there remains a challenge in determining how much work each individual can effectively manage and review.
“The choke point is now the individual reviewing the output,” Grenstad explains. He emphasizes the company's careful approach when assessing critical or sensitive parts of the code but acknowledges that overall, manual code reviews are not necessary for every task.
The shift toward AI also influences Grenstad’s hiring strategies. For example, in constructing the player journey team, he now prioritizes candidates who possess knowledge of the casino and iGaming sectors alongside skills in product experience rather than purely focusing on senior developers with limited domain understanding. “Technical capabilities remain crucial, particularly for assessing system architecture to maintain reliability, but domain expertise is becoming immensely valuable. I would prefer hiring someone knowledgeable in the domain, even if they lack expertise in Java,” he notes.
This pivot raises questions about how prospective developers will enter the field. With diminished demand for traditional junior Java developers, the conventional path from junior programmer to senior engineer may evolve.
“We need to consider how we prevent a scenario where we have many seasoned Java developers while new entrants lack fundamental understanding because they were never onboarded. That’s a compelling issue, one we’ve been examining,” Grenstad reflects.
Additionally, Cubeia is departing from the old model where developers completed a product and then handed it over to QA. Grenstad describes the former workflow, where product owners defined the project, developers built it, QA tested it, and operations managed release. Now, QA is integrated into the process from the planning phase, focusing on what can be automated and evaluating testing coverage.
Ownership of a task is now broader, encompassing problem identification, development, deployment, and results measurement. While this transformation has alleviated several of Cubeia's concerns, it hasn’t come without challenges. Some coders lament the reduced emphasis on traditional coding, with many having enjoyed the hands-on problem-solving aspect that has diminished.
There’s also a strategic consideration regarding Cubeia’s reliance on AI providers. Currently, the company is heavily dependent on Claude, and Grenstad is concerned about potential price hikes or quality fluctuations from such providers. Nevertheless, he views this unpredictability as manageable: “We’re confident that as we navigate these challenges, solutions will emerge. We’ll encounter issues, learn to manage them, and adapt accordingly.”
Looking back, Cubeia’s original objective in its AI journey was straightforward.
"When we started this journey, my only vision was: ‘Let’s not be writing code in August,’" Grenstad recalls. By striving to eliminate coding as a bottleneck, the company has unearthed that coding was merely one constraint within a more extensive system. As the hindrances shift, the organization must adapt to new demands surrounding domain expertise, product-oriented thinking, quality assurance, prioritization, and customer understanding.
Consequently, the technology itself has become a lesser focus in this narrative. What Cubeia chooses to do with the technology in the coming years is now at the forefront of its ongoing evolution.
