Production game studios ship enormous productions into a market with zero tolerance for mistakes. Due to the amount of games published every year, today’s players have very little tolerance for issues like broken quests, menu crashes, or unfinished assets. At the same time, games have become so complex that quality assurance has become one of the most tedious and costly jobs in development.
ManaMind founders Emil Kostadinov (CEO) and Sabtain Ahmad (CTO)
Manamind
ManaMind, a London startup, says AI agents can take over this work. The company has raised $1.1 million in pre seed funding to launch a platform built around autonomous agents that play and test games using only video and audio, the same way human players perceive the world. The agents watch. They listen. They decide what to do next inside the running game.
Founder and chief executive Emil Kostadinov lived the pain firsthand when he worked as a tester in his teens, walking into every wall of a level to see whether the world stitched correctly. “I felt scammed the first time,” he told me. “I thought I would be paid to play games and instead I spent hours doing the most repetitive tasks imaginable.”
Metamind’s AI will replace the tedious job of Q&A to test each iteration of the game to find bugs.
Manamind
That manual grind still defines most QA. In the interview Kostadinov said that on some productions QA has reached twelve percent of the total budget. Script-based tools help, but they depend on engine access and don’t scale well across different genres, pipelines, or platforms.They do not behave like players and they do not generalise well.
Play Puzzles & Games on Forbes
ManaMind’s approach comes from its technical co-founder, Sabtain Ahmad, the company’s CTO. Ahmad holds a PhD in machine learning from TU Wien and has spent more than ten years building perception systems and deploying them under platform constraints. He built ManaMind’s proprietary vision language model after concluding that public systems could not interpret games reliably. In internal tests the model outperformed systems from OpenAI, Google, and Anthropic on bug detection tasks.
Metaminds monitors its AI’s decision-making in the game.
Manamind
Ahmad said the breakthrough came from abandoning the idea of universal automation. “It became obvious pretty early that no existing model could actually understand or move through a game the way we needed,” he told me. “Studios want AI automation, but they all want to automate different things. That is what pushed us toward teaching an agent to behave like a real tester using only audio and video. No code, no engine hooks, nothing special. And because of that, we have been able to stay flexible in ways other approaches cannot.”
Kostadinov showed me a demo of the platform running a vertical sync test. Vsync is the setting that keeps a game’s frame rate aligned with the monitor’s refresh rate to prevent screen tearing. The agent realised it could not verify the setting from the options menu, exited into gameplay to collect evidence, and returned to the menu to complete the evaluation. “It came up with that on its own,” he said. “I would never have thought of that.”
ManaMind is still just a two person founding team. Kostadinov, thirty, handles business and product. Ahmad, also thirty one, built the technical system. They began just ten months ago, in January, 2025, and spent the better part of the year testing the technology so the company is still in its earliest pre-revenue state, with its first commercial rollouts planned for January.
The platform is already in use with four early access partners, including THQ Nordic and several unnamed studios of different sizes and pipelines. The system is engine agnostic, runs entirely from captured audio and video, and drives tests across many different genres without switching frameworks. It produces its own logs, evidence, and reports that plug into existing QA workflows.
Investors say the complexities of modern game design make it a natural environment for training general purpose agents. EWOR chief executive Daniel Dippold compared ManaMind’s approach to early DeepMind and OpenAI work in games, with the distinction that ManaMind is delivering commercial value rather than research prototypes. Heartfelt Capital investor Imti Basharat said the agents’ ability to operate in unfamiliar digital environments gives the company a broad foundation to expand beyond gaming.
“Games are the perfect proving ground for AI,” said Kostadinov. “They combine complexity, interactivity, and scale, the same ingredients needed for AI systems to understand and act in the real world.”
Kostadinov says the company’s long term plan is to evolve the perception and reasoning stack into a platform that can support general software testing and eventually robotics. For now the focus is on the games industry’s most effortful and repetitive work. “QA is an innately boring, repetitive, expensive job. I know because I have done it,” he said. “People who want to build games should not spend their best years walking into every wall to see what breaks.”