The first thing you notice opening a nunu.ai test report is the replay. Not a log file, not a stack trace: a video of an AI agent moving a character through the opening tunnel of Hogwarts Legacy. Mouse twitches and keyboard taps timestamp against its internal monologue [nunu.ai, retrieved 2024].
The agent narrates what it sees, decides where to go, flags issues. Thirty minutes of tutorial, no scripting, no training. For a QA lead, the bug report has a witness.
That artifact sells nunu.ai. Founded 2022 by Kyrill Hux, Nicolas Muntwyler, Jan Schnyder, the San Francisco company builds agents that play like humans. Findings go through nexus portal [Y Combinator, 2024].
Developers, artists, producers write plain-language plans. Agent runs them. Bugs hit Slack/Jira with recording, trace, reasoning [nunu.ai, retrieved 2024].
In March 2025, a16z speedrun led $6M seed atop $2M pre-seed [Crunchbase, March 2025] [Mobidictum, retrieved 2026].
The bet
Game QA is labor-intensive, unloved. Studios hire contractors for repetitive tests: clipping bugs, quests.
Costs scale linearly. nunu.ai bets pixel-perceiving agents absorb regression testing before patches.
Proof of generalization: Hogwarts Legacy, livestreamed Pokémon Emerald world-record AI speedrun [nunu.ai, retrieved 2024].
Pricing tiers indie to AAA, enterprise with manager/support [nunu.ai, retrieved 2024]. Real SaaS shape.
Why it could be big
Tailwinds align: climbing budgets, slipping schedules, recent multimodal reliability.
a16z speedrun backs games infrastructure [Crunchbase, March 2025]. TIRTA, YC complete cap table [TIRTA Ventures, retrieved 2026] [Y Combinator, 2024].
Ambitious: Rob the Robot transfers agent to physical robot [FOV Ventures, retrieved 2026]. TIRTA sees "unembodied minds" to robotics [TIRTA Ventures, retrieved 2026]. Optionality draws investors.
The team and traction
Pre-Seed 2024 | 2 | $M
Seed Mar 2025 | 6 | $M
Total disclosed | 8 | $M
Team small: five in SF [Y Combinator, 2024]. YC W23 [LinkedIn]. StormForge case shows customer-built surface [nunu.ai, retrieved 2024].
The honest counterfactual
Competition heats: labs ship computer-use agents. Studios could build on off-shelf models.
Risk: compression between generals above, incumbents below.
Bulls: moat in scaffolding (perception, emulation, nexus), relationships. Demos prove on commercial titles [nunu.ai, retrieved 2024].
What to watch
Next 12 months: named AAA logos? Rob the Robot to product?
Series A late 2025/2026: generalist signals embodied; strategic signals QA compounding.
Cultural question: AI finds bugs, what is human QA job?