Technologies to Identify Game Issues

Matthew Tighe, founder of Do Games, has spent nearly a decade porting games to consoles, including titles like Cult of the Lamb and Art of Rally. His new venture, Gameworks, leverages this expertise to make developers' lives easier by providing a platform that analyzes games for common performance issues and certification problems, then generates a report for the user. The system utilizes a combination of AI techniques, including large language models, custom models, and computer vision models, to identify issues such as resolution or frame-rate dips. Tighe explains that the goal is to test throughout the development process, rather than just at the end, to catch bugs and problems early. The Gameworks platform allows developers to add their own issues and clip specific parts of a video timeline for a test session, making it easier to collaborate and address problems. There are plans to integrate the system with tools like Jira and Trello, and it has its own internal issue system that resembles Trello cards. The system can recognize similar issues and group them together, making it easier for developers to prioritize and address problems. Tighe emphasizes that the goal of Gameworks is not to replace human developers but to make their lives easier by automating tedious tasks and providing valuable insights. PerfCop, or Performance Copilot, is another tool working along similar lines to Gameworks. It's a performance analysis tool based on statistical analysis with a thin layer of generative AI, designed specifically for Unreal Engine. PerfCop goes beyond existing performance analysis products by providing deep statistical analysis to identify abnormal execution patterns, isolate the most impactful issues, and break them down to the function or scope level. The AI component generates a report based on deterministic analysis, communicating structured findings in a way that accelerates debugging and decision-making. According to Ken Noland, co-founder of AI Guys, PerfCop saves developers around 15-70 hours of time per month. The tool has a chat interface called Sherlock, which allows developers to ask specific questions about the results and extract actual performance metrics from the trace file. Noland sees this kind of back-end application as the future of generative AI in the games industry, helping developers identify performance bottlenecks and optimize game development.