Side, a seasoned services provider formerly known as PTW, has entered into a six‑month memorandum of understanding with Danish artificial‑intelligence specialist Modl.ai to co‑develop an AI‑driven quality‑assurance (QA) platform for video games. The partnership merges Side’s extensive experience in outsourced services with Modl.ai’s cutting‑edge AI technology, aiming to create a solution that enhances testing efficiency while keeping human expertise at the core. Modl.ai, founded in 2018, secured $1.7 million in seed capital two years later and has since focused on applying machine learning to the games industry. Historically, AI‑based QA tools required developers to hand over full game builds—or even source code—so the AI could analyse the software internally.
This “integration‑heavy” approach, which relied on plugins, SDKs, and deep access to proprietary assets, proved cumbersome and often stalled adoption because developers were reluctant to expose sensitive code and the process introduced significant friction. The new system, explained by Modl.ai founder Christoffer Holmgard, eliminates the need for any internal code access. Instead, the product is treated exactly as a human tester would receive it: a compiled build delivered to a device that mirrors the end‑user’s environment.
The AI runs the game, watches the screen, simulates inputs, and records every observable event. A second AI layer then reviews this data, flags anomalies, and escalates potential bugs to human reviewers for final validation. The first phase of the workflow is called “building game knowledge.” Developers upload assets, design documents, screenshots, or any other relevant material and converse with the AI in natural language.
By feeding the system contextual information, the AI constructs a mental model of the game’s mechanics, narrative, and visual style. This conversational interface allows teams to ask questions like, “How does the inventory system behave when the player reaches capacity?” and receive an AI‑generated understanding that will guide later testing.
Once the AI has a solid grasp of the title, developers define testing objectives. These can range from traditional test cases—such as verifying level loading times—to novel scenarios that exploit the AI’s ability to generate edge‑case situations. Holmgard stresses that human insight is still vital at this stage: testers must craft prompts and instructions that steer the AI toward meaningful exploration, translating domain expertise into language the model can act upon.
During a test run, the AI produces a comprehensive bug report that includes textual descriptions, short video clips (the “evidence”), and full log files. These artifacts are handed to a senior QA specialist who reviews them, determines the severity, and decides on the appropriate remediation path. The human reviewer also provides feedback to the AI, labeling each finding as a genuine defect, an intended feature, or a false positive.
Over time, this iterative loop refines the model’s ability to distinguish between bugs and design choices, gradually reducing the rate of misclassifications. One of the persistent challenges in automated testing is the ambiguity between a bug and a feature. For example, an unexpected visual effect might be an intentional easter egg rather than a flaw. By incorporating continuous feedback from experienced QA professionals, the system learns these subtleties, improving its predictive accuracy and aligning more closely with the creative intent of the developers.
Side envisions a broader transformation of the QA workforce. Rather than remaining confined to repetitive, low‑value testing tasks, staff can evolve into “AI engineers”—professionals who design prompts, fine‑tune models, and interpret AI‑generated insights. Martin McBride, Side’s chief commercial officer, argues that this shift moves employees up the value chain, offering higher salaries, new career trajectories, and the possibility of repatriating jobs that had migrated to low‑cost regions. By automating the monotonous smoke tests and regression checks, testers are freed to focus on strategic analysis, gameplay evaluation, and creative test design.
Benedikte Mikkelsen, Modl.ai’s chief product officer, echoes this sentiment, noting that many current test procedures involve mind‑numbing repetition. "You don’t need a human to click the same button sequence a hundred times; the AI can handle that, allowing testers to concentrate on assessing game feel, balance, and emergent behavior," she explains.
This reallocation of effort not only boosts morale but also adds tangible value to the development pipeline. Both companies are careful to stress that AI will not replace QA professionals. The price point for manual testing in certain offshore markets remains low, making a wholesale substitution economically unfeasible.
Instead, AI serves as an augmentation tool that enhances productivity while preserving human judgment for complex, nuanced decisions. McBride describes the approach as a “sliding scale”: the proportion of AI involvement can be adjusted based on project needs, but a human will always remain in the loop. Regarding applicability, Side’s chief operating officer Oliver Deslandes admits that the current solution is best suited for simpler titles—puzzle games, card games, and other experiences with repeatable mechanics and modest graphical intensity.
Fast‑paced, 3D open‑world or fighting games present additional challenges that the team plans to tackle in future iterations. For now, the focus is on partners willing to pilot the technology and help shape its evolution, balancing expectations for cost savings with the desire for reliable, high‑quality results. In summary, the Side‑Modl.ai partnership aims to blend AI’s capacity for rapid, repeatable testing with human expertise in interpretation and decision‑making. By delivering a system that operates on finished builds, learns through natural‑language interaction, and continuously refines its bug‑identification algorithms via human feedback, the collaboration aspires to redefine quality assurance in gaming.
The ultimate goal is a more efficient, satisfying workflow for testers, a faster path to market for developers, and a sustainable model that can adapt to the diverse range of games emerging in the industry.