Companies undergoing major operational changes must navigate the complexities of integrating new systems and processes while adapting staff roles. Effective organization and thorough preparation can facilitate this transition, yet uncertainties about employee reactions often linger. In April, Cubeia, a developer of iGaming platforms, decided to fully implement AI-assisted coding across its systems, prioritizing the use of Claude's large language model (LLM) to enhance the speed of development and the rollout of updates for clients.
By mid-July, Cubeia’s Chief Operating Officer (COO) Stefan Grenstad reported to iGaming Business that the supplier’s front-end operations were entirely reliant on AI, while the back-end systems approached a 90%-95% AI integration. Grenstad's earlier assertion that this shift would accelerate update releases proved accurate, enabling more frequent updates than the previous reliance on manually written code.
A significant unforeseen hurdle was how this transition would challenge the skill sets of the development team. "It’s clear that some developers struggle with context switching; to boost productivity, they need to multitask effectively," Grenstad acknowledged. He elaborated that managing several agents simultaneously allows for launching multiple tasks, but achieving the right balance is crucial.
"Finding the optimal number of parallel tasks a developer can manage is essential, and this has been an unexpected challenge," he added. Part of the transition included restructuring and providing additional training to adapt to these new workflows.
In response to these challenges, Grenstad noted that the team has engaged in discussions about setting daily task limits. While some tech leads have resisted the shift towards entirely AI-generated code, others have embraced it, including skeptics who have found value in using Claude as a collaborative tool. "One senior developer who was initially doubtful has actually started using Claude as a discussion partner and adapted positively to AI-assisted development," Grenstad remarked.
Conversely, there are concerns regarding a potential loss of control due to the rapid increase in release frequency. Before the transition, Cubeia typically handled four to six releases per month with an average of 50 support tickets raised. This dynamic shifted dramatically; in May, for instance, there were only five releases accompanied by 86 tickets, and by June, the team achieved ten releases and 96 tickets.
Recently, the transition has opened a discussion about how to reorganize teams to better accommodate new processes and expand AI utilization across different functions. Grenstad is focused on improving the integration of the quality assurance (QA) team into the development cycle, especially as release speeds increase. "The QA function is currently part of the development team, but its handover process is ineffectual with AI-driven development. We need to start integrating AI into QA concurrently with development.
Under the new framework, developers will oversee tasks or releases until they reach production, making the QA phase an integral part of that process," Grenstad explained. Ensuring the teams receive adequate support during this transition is vital. He noted, "Developers largely support the goal of a full transition to AI-assisted coding. My tech leads have been pivotal in helping those struggling adapt.
While I believe the team acknowledges this is the future of our work, there remains skepticism regarding how to maintain quality at such a rapid pace of development.
In summary, Grenstad emphasizes a collective acceptance of this direction and the importance of addressing concerns surrounding quality assurances amidst the shift to an AI-driven coding environment.
