Cubeia’s road to AI-assisted code (part two): How is the development team adapting?

When implementing a seismic operational change, a company inevitably is put through a process of adapting to new systems, new processes and new roles for staffers. While remaining organised and carrying out significant due diligence before introducing that change can help new systems integrate quickly, there is always an element of the unknown for senior teams.

People are ultimately unpredictable, and how they react to change can impact how well new systems are implemented. In April iGaming platform developer Cubeia took the decision to adopt entirely AI-assisted code across its systems, with a focus on utilising Claude’s LLM to speed up the development and release of new updates for its client base.

By mid-July, Cubeia COO Stefan Grenstad tells iGB, the supplier’s front-end was 100% AI-driven, while its back-end systems were only slightly behind at 90%-95%. Sure enough, Grenstad’s estimation that the shift would speed up releases and allow it to push out more updates for clients than when it was reliant on human-written code, has come true.

Multi-tasking

However, one challenge the senior management team did not anticipate was exactly how much the transition would test the development team’s skillset. “What’s been very obvious here is that those of my developers that have a hard time with context switching are struggling because to increase the productivity, you need to be a multi-tasker,” Grenstad explains candidly.

“[When you multi-task you have several agents [running] at the same time, and they’re being done at different times, so in principle, you could start hundreds of them,” he continues. Notably, he believes there is a balance to be met. Because setting up an agent can take minutes, there is room for developers to potentially carry out lots of releases or tasks within a day. But multi-tasking is a crucial skill in ensuring all these can run in parallel and another developer will be available to review the AI-written code.

“There’s a sweet spot for how many parallel tasks [a developer] can actually orchestrate all at the same time. That becomes very important, and we need to find the perfect amount. That’s a side effect here that I didn’t see coming,” says Grenstad. As part of the transition, restructuring and additional training are needed to adapt to the updated processes.

“It’s a prioritisation skill, really, in terms of people knowing and understanding what to focus on versus what is less of a priority.”

Setting limits

The COO says the challenge has led to an open discussion internally on setting limits on how many tasks the team should carry out within a day. When asked whether there has been clear pushback within the team from any tech leads who are less enthused by the shift to a fully AI-written code, Grenstad says “of course”. In contrast some prior skeptics have taken to the new model extremely well.

“One of our senior guys that was one of the big skeptics has actually been using Claude as his sounding board. He has really adapted to AI-driven development in that way. So he’s one of the, you know, I think he’s one of the guys talking very positively about it.

“[When senior deveopers] start asking questions to AI the model can correct itself or make the solution even better, and that’s amazing.”

Others however, have concerns about losing control following the speed of releases increasing significantly. Prior to Cubeia’s road to AI-written code transition, it made between four and six releases per month, with an average of 50 tickets raised.

“We really increased the speed in May and June. In May we only had five releases, but we had 86 tickets. In June we had 10 releases and 96 tickets,” Gransted notes.

Notably, what’s really come to light in the last few weeks is the opportunity to reassess and restructure the team, to better integrate new processes and spread the use of AI across various teams. Ultimately, the intention is to have AI used across all functions.

Better integrating the QA function

One consideration Grenstad is making is how to better integrate the quality assurance function (QA) into Cubeia’s development cycle, particularly now that releases are happening at speed.

“[The QA function] is part of the development team, but there is kind of this handover in the middle. And that doesn’t work very well now that we have AI-driven development. When we try to start with AI-driven QA, they should go hand in hand at the same time.

“Under the new structure developers will be responsible for a task or release up until it gets into production, and the QA phase is part of that,” says Grenstad. QA, he believes, also needs to become a part of the planning process.

A significant part of the process is ensuring the teams involved have the necessary support required to help them through the transition, and he believes that largely, the developers do support the overall target of transitioning to a fully AI-assisted code. “My tech leads are doing a tremendous job with those that are struggling a bit, and you know supporting them, helping them,” notes Grenstad.

“I think everybody agrees now that this is the way we should be working. This is the future. I don’t think they are thinking this is the wrong way to go. I think they still are, as am I, skeptical about so many things like how do we ensure quality remains the same when we have such a high pace.”