ManaMind raises $1.5m for AI agents that replay games to find bugs

On 30 April, London-based ManaMind said it had closed a $1.5m (£1.1m) pre-seed round led by SVV (Sure Valley Ventures), with EWOR, Ascension, Syndicate Room and Heartfelt taking part. The company builds AI agents that play through video games, find the bugs and write the reports that human testers would otherwise produce by hand. SVV published the announcement on its own site the same day.

The buyer is the games studio and the budget line is quality assurance (QA). SVV’s announcement puts QA at 10–15% of a game’s total budget, a figure it gives without a source. Much of that work is regression testing: replaying a game after every change to check that nothing which worked in the last build has broken in the new one. It is repetitive, it scales with the size of the game and it is what ManaMind’s agents are built to take over.

The company was founded by Emil Kostadinov, its chief executive, and Sabtain Ahmad, its CTO. According to the announcement, Kostadinov is an Oxford MBA and EWOR fellow who “experienced the pain of manual QA firsthand as a game tester”, and Ahmad holds a PhD in machine learning. ManaMind Limited was incorporated at Companies House on 3 January 2025. Both founders have been directors since that day, and both are registered as persons with significant control.

At $1.5m the round is small, and the evidence behind it comes from early deployments. Kostadinov’s stated ambitions reach as far as robotics, but the case SVV makes is narrower: hours per regression cycle and the share of critical bugs caught, with design partnerships at two studios behind it. Whether those numbers hold at studios that pay for the service is what the $1.5m has to show.

Agents that test from the screen

ManaMind’s agents work from what a player sees and hears. Its home page, in a version archived on 30 April, said its bots use “the same inputs (audio and video)” and “the same outputs (keyboard/mouse, joystick)” as human players, and are “platform, device, engine, and code agnostic”. The same page said that this let them test “like actual game testers, hence being able to catch bugs other automated solutions miss”. For a studio, the claim is that testing can start without anything being built into the game’s code.

Underneath sits a model the company calls HiveMind. “We’ve developed our own proprietary visual model specifically for virtual environments because gaming demands that level of precision,” Kostadinov said in the announcement. The system generates reports for each issue it finds, so that testers can move from documenting bugs to fixing them, and it runs around the clock alongside development rather than after it, according to the company.

“We’re automating the manual, time-consuming parts so studios can focus on building amazing worlds,” Kostadinov said.

Company figures from early deployments

The announcement says early deployments completed full regression cycles in six hours rather than days, with 86% of critical bugs caught before shipping. These are the company’s own figures. The announcement does not name the games or studios involved, or say what total the 86% is measured against.

Brian Kinane, founding partner at SVV, built the investment case on the same figures. “Their autonomous agents complete in six hours what takes manual QA teams days – and they catch bugs that human testers miss,” he said. “That’s the kind of measurable improvement we back.”

ManaMind’s own home page makes bigger claims. A version archived on the morning of 30 April still promised studios savings of “up to 80% of your QA budget”, said its bots could cut “up to 8 months of testing” and carried a waitlist sign-up. Neither figure appears in SVV’s announcement, and the page does not say how either was measured.

Two studios and a pre-seed round

ManaMind has design partnerships with Included Games, a mobile games studio, and CrazyLabs, which the announcement calls a hypercasual developer. When the announcement makes the wider case for automated QA, it reaches for AAA titles that ship with more than 1,000 hours of content and for Cyberpunk 2077, whose troubled release, it says, wiped $1bn (£740m) from CD Projekt’s market value in days. Neither named partner is described as a maker of games on that scale, and the announcement does not say whether either partner pays for the service.

Companies House filings show ManaMind allotted shares on 12, 16 and 30 March 2026, a month before the round was announced. The company says it will spend the money on its technical team, on developing its own model and on scaling “across key geographies”, which it does not name.

Kostadinov’s ambitions run well past game studios. “Gaming is our launchpad, but our vision is to build the autonomous testing layer for all software and, ultimately, robotics,” he said. The company’s own description of itself talks of powering robots that understand the world around them.

Kinane’s case rests on a regression cycle measured in hours and on bugs that human testers miss, at a company whose named users are two studios on design partnerships. The round gives ManaMind time to turn those partnerships into paying contracts and to put its six-hour figure in front of studios with bigger games and bigger QA bills.