Unless you’ve been living under a rock, you’re well that AI has been worming its way into everyday life for some time now, with seemingly new use cases every week. Much has been made over AI’s encroachment into the world of gaming, from AI frame generation on Nvidia’s latest 5000 series GPUs, to Microsoft’s recent announcement of a generative AI capable of creating game visuals and controller actions. While no self-respecting techie can consider themselves totally averse to technological progress, there is certainly some warranted skepticism on what these sorts of tools mean for the future of gaming, and what their myriad potential applications portend for the future of game development itself.
To my surprise, Razer — yes, the same Razer that makes super-powered gaming laptops and RGB-laden hardware — stepped into this space in a big way earlier this month. The Singapore-based gaming hardware company announced a new tool it is calling AI QA Copilot, an AI-powered game development tool that it says will aid developers in finding bugs faster and more efficiently. This comes after Razer teased an AI gaming coach dubbed Project Ava at this year’s CES in Las Vegas.
The demo at the time, which I saw firsthand, was pitched as an AI gaming coach, and at the time was being trained on MOBAs. It was a disembodied voice that could observe a player’s choices and make recommendations in real time, like when it would be best to buy items, as well as real-time strategic suggestions to help a player win fights. It was impressive but niche, and any specs under the hood were very hush-hush. It’s clearer to me now that Razer’s goal all along was to evolve into the business of game development with an AI-powered toolkit developed in-house. The implications of AI-assisted game testing are far-reaching, and the challenges to bring it to market are great, and the philosophical questions along the way, perhaps greater.
Speaking to Digital Trends, Quyen Quach, VP of Software at Razer and lead of the AI team, as well as Dr. Eric Vezzoli, Razer’s Director of Technology, discussed their vision for AI QA Copilot, the competitive landscape that has suddenly emerged around AI-assisted game development, and what it all means for the future of the company and industry.
Questions were jointly answered by Quyen Quach and Dr. Eric Vezzoli
Addressing concerns
While AI is an exciting frontier for companies like Razer, there are several hurdles that the company must clear to realize its potential. Everything from its unreliability to concerns about job loss stand in its way. Among the biggest concerns is energy use. If the billions of dollars that big-tech is burning buying H100 Nvidia cards and the talks of reopening closed Nuclear plants are anything to go off, AI requires a lot of power to function. I wondered how much compute power AI QA Copilot would require to run, and whether the costs of that power could ostensibly eat into labor cost savings for developers leveraging such a tool.
The team at Razer says that “the compute power usage of AI QA Copilot is only in the cloud, as the rest of the local compute are event implementation in game, which allows the system to be used only when required by the agent, for example in idle state or when simply walking around, there is no resources consumption.” Declining to elaborate further they add “we can’t share metrics right now, but we expect the time saved to be significantly larger compared to the compute costs.” Perhaps this will function as an enterprise cloud subscription tool, much like Open AI’s ChatGPT. Or it might be a token system like working with developer APIs.
On the topic of labor cost savings, I asked the obvious question: Does Razer anticipate that this tech will replace QA jobs? The company says that it views AI QA Copilot as a tool, not a replacement.
“We believe our AI Gaming innovations are more enablers for the industry,” Razer says. “Razer QA Copilot will help QA testers in the most tedious part of their job by streamlining the repetitive tasks. The QA tester which will retain the ability to validate the suggestion of the QA copilot to be more efficient, allowing them to do more, faster. This would eventually mean Games having a faster QA cycle and Gamers having a better gaming experience.”
Razer has already signed a “Memorandum of Understanding” (a non-binding agreement) with QA company Side, to “leverage the new Razer AI QA Copilot to speed up the QA process.” Razer explains “the Razer AI QA Copilot will be integrated into side’s in-house global QA teams to help boost testing efficiency.” Side works many of the leading game studios including Bethesda, EA, and Epic Games.
Ghosts in the machine
One of my first questions when all this tech was unveiled was how close were these tools to functioning unattended. What about hallucinations which are a known issue in the AI world? Here, Razer again frames the narrative around these tools as assistants, and not full-on replacements.
“QA copilot is meant to be a buddy for a QA tester, which have the final ability to decide if a flagged discrepancy should be identified as an issue or not,” Razer says. “The important aspect for us is that the ration between false positive and real bugs is overwhelmingly balanced towards real bugs. The QA tester is finally the one who decides what’s then reported as an issue or not after revising the QA Copilot report.”
Both human and machine are similar; they are not perfect.
I pressed further, wondering what the Razer teams longer term view was on the reality of hallucinations in AI, and what they were doing specifically to address this when building AI QA Copilot.
“Both human and machine are similar; they are not perfect, which is why we designed the AI Copilot to be just that for QA devs, a copilot,” Razer says. “Whether the AI is automatically flagging bugs for QA to review or automatically generating the end report, the goal is making the job simpler and more efficient by reducing the tedious tasks. Either way, the end result remains the same — the AI Copilot works with the QA teams to continuously learn and improve at identifying unintended behaviors in games.”
Much like todays leading “Auto-Pilot” features in vehicles, it sounds like Razer is saying someone still needs to be at the wheel while these tools are running. If not literally while they are running, then at least afterwards to review its work. I imagine a near-term and realistic implementation might look like one or two QA testers overseeing AI QA Copilot running through a game on multiple machines at once, while they supervise and review any flagged bugs. The copilot could also be left running at all hours or overnight, with QA testers then reviewing flagged bugs in batches. At a high level, the point is to save all the time of gameplay needed to find these bugs in the first place, and this will invariably lead to savings on labor costs for the developing studio.
An emerging landscape of AI in gaming
AI in gaming has felt like it suddenly went from nothing at all to everything everywhere all at once. Among the companies that have announced AI tools for use in game development, Microsoft’s Muse stands out as the big kid on the playground. I asked how Razer felt about a behemoth of that size working in such close proximity to their stated goals, and how it felt that its tech stands out.
Razer responded with an air of self-assurance, saying “It has been a fun journey to see how much we have innovated on AI Gaming since Project AVA was announced at CES a few months ago, and it’s great to see others follow and enter the space with their versions.”
I’m left with the distinct impression that such a confident tone from those closest to the project, means that the company knows is has something good on its hands. Addressing the emerging landscape more directly, Razer tells me “the Razer QA Copilot caters to the refinement of the game experience once the gameplay is set, thus serving a different, and somewhat unaddressed, need in the market.”
The ToS of it all
Hallucinations and competition aren’t the only challenges ahead for Razer. There is a fundamental question of how these tools are being trained, and precisely what they are being trained on. When Razer launched Project AVA (now Game copilot), its AI gaming coach at CES 2025, it was clear to anyone who attended the demo that Razer was not in a position to be very forthcoming about what it had used to train AVA.
Most games have clauses in their terms of service that explicitly forbid precisely the sort of in-game scraping and testing needed to develop something like an AI copilot. I asked if that was still the case or if some partnerships with game studios had brought that out of the grey.
Razer’s answer was short and to the point: “At Razer, we work with leading partners from the best studios in the world. Yes, we train them in partnership with studios.” Developing these tools on the up-and-up will be necessary as they transition from prototypes in a dimly lit room with three monitors and a server rack, to fully fleshed out commercially viable products.
Integration is the name of the game
Overcoming these challenges, and the substantial investments required to see it all through, are surely to serve a greater end than helping me better defend lane in Deadlock. Razer’s newly announced Project WYVRN, which houses not only it’s AI products like Game copilot and AI QA Copilot, but also Razer Sensa Haptics, Razer Chroma RGB, and Razer’s prized 2016 acquisition THX, sits at the core of the companies vision. Razer is placing high hopes that this suite of products will allow its tech to be integrated into game development from the ground floor.
“The motto of Razer is For Gamer By Gamers,” Razer says in discussing its vision for the future. “With our latest innovations, Razer is leading the AI Gaming space with meaningful innovations to make the overall game development and eventual experience for the gamers better than before. We believe AI is a tool at the service of the creative minds creating the experiences we enjoy every day.” Razer seems focused on achieving a sort of vertical integration in the marketplace. Help make more games, better, and faster, and end up with more games tailored to Razer’s suite of immersive gaming peripherals.
Showing no shortage of ambition, Razer says “we empower games designers, game developers and gamers to be able to create better and more interactive experiences, as seen in QA copilot and Game Copilot. We believe that Razer is set to revolutionize the game industry, by leading AI in gaming.”
Wading into uncharted waters
Razer paints a utopic view of tools like AI QA Copilot and their role in game development. We are asked to envision small indie developers who can’t afford 100 QA testers on payroll and imagine that AI QA Copilot would allow for testing at what would ultimately be a lower total cost. In a best case scenario, this could be a boon to smaller studios and their ability to pump out titles on time and on budget in a market that can often feel akin to David vs. Goliath. 99% of studios are working with a smaller budget than EA, Tencent, Sony, Activision, and the like. Eric Barone created Stardew Valley on a shoestring budget over a four year period while working an evening job as an usher. AI QA Copilot, in its most idealistic implementation, could pave the way for more success stories like Eric’s.
The other side of this, however, paints a more foreboding picture. Mega-cap game studios could leverage tools like the one Razer is building to lower labor costs and increase profit margins. These savings will certainly not be passed on to consumers, and the same $60-$70 AAA game price tag will simply earn more for the juggernaut studios that develop them. Whether this leads to increased productivity in the field, job losses for testers and entry-level developers, or a balance of both remains to be seen, and those problems stand at the very heart of the debate over AI and its use in any creative endeavor.
Razer’s team made clear to me that CEO Min-Liang Tan is serious about these projects, and investing heavily to see them come to fruition. Time will tell how these tools will change the gaming landscape, and I’m personally hopeful for a brighter version of that future than the one many fear.
