Hello, and welcome to the latest AI Gamechangers conversation, your weekly newsletter where we talk to the people building the future of AI in games. It’s your guide to the practical and innovative use of AI across the games industry. Feel free to browse the archive and share what you find. Scroll to the end for our curated news round-up.
This week, we’re looking at a brand new AI-powered action game. We’re sitting down with someone who’s asking what happens when AI stops being a production shortcut and starts being the game itself. Robin Qin is the lead game designer at Meshy Game Studio, the new publishing and development arm of Meshy (the 3D generative AI platform with over ten million users). Its debut title, Black Box: Infinite Arsenal, was revealed at GDC, and is built around an agent that generates weapons in real time, on the fly, from player input. No pre-scripted outcomes or defined weapon pool. Just AI weaving the experience as you play.
Robin Qin, Meshy
Today we’re chatting with Robin Qin. He is the lead game designer at Meshy Game Studio, the new game development and publishing arm of Meshy, a 3D generative AI platform serving over ten million users.
Robin spent much of his career designing large-scale live-service titles, including MMOs and SLGs at Tencent and FunPlus, before joining Meshy roughly a year ago. His debut project, Black Box: Infinite Arsenal, is a survivors-style roguelike built around a proprietary AI model that generates weapon mechanics in real time.
We had a chance to play Black Box: Infinite Arsenal during GDC in March, and then followed up with Robin to dive into the team’s philosophy and ambitions.
Top takeaways from this conversation:
Meshy, famous as a leading AI-powered 3D model generator, has launched a dedicated game development studio, with its first game coming to Steam soon.
Black Box: Infinite Arsenal uses a proprietary AI model trained specifically on gameplay mechanics, enabling real-time generation of new weapons.
The “Designer Agent” approach keeps players in control through simple material-combining choices, hiding the generative complexity while still delivering unpredictable weapon outcomes.
The game is a proving ground for Meshy’s core platform: future versions will use Meshy’s 3D pipeline to generate assets, textures, and animations in real time, directly feeding learnings back into its tools.
AI Gamechangers: Meshy has built its reputation as a 3D asset generation platform. What prompted the move from “AI for production” into developing and publishing an original game?
Robin Qin: Meshy is the leading 3D generative AI company, serving more than 10 million users at both individual and enterprise scale, generating more than 100 million models to date. While our platform kept getting better at producing assets, we noticed that AI’s potential goes far beyond speeding up production - it can actually reshape the foundational logic of games.
As AI adoption matured across the industry, we watched it evolve from early “chatty NPC” experiments into something that can drive a game’s core loop. We launched Meshy Labs as our experimental AI incubator, and Black Box: Infinite Arsenal is its first title, built to show that AI isn’t only about saving developers money. It can deliver player-facing experiences that traditional scripting simply can’t.
It’s both a push on what’s technically possible and our early answer to what an “AI-native game” can be. As our CEO, Ethan Hu, puts it, “If AI for 3D is transforming how we produce, then AI for Fun is the next frontier, changing not just how games are made, but how they are played and experienced.”
Black Box: Infinite Arsenal is described as an “AI-native” game. What does that mean in practice compared to more traditional procedural or systems-driven design?
The core difference is that AI shapes the gameplay loop itself, not just assists in content production.
In a traditional roguelike, the weapon pool is finite, and designers hand-write every interaction. In Black Box, the heart of the game emerges dynamically: AI generates the logical core and mechanics of weapons in real time, opening a design space traditional procedural approaches can’t fully enumerate.
“We’re already exploring what the next titles look like, with each one designed to push the boundary of AI-native experiences in a different direction”
Robin Qin
It’s genuinely open-ended rather than pre-scripted. That gives players far more freedom to experiment with.
Can you walk us through a typical gameplay loop? What is the player actually doing moment to moment, and where does the AI step in?
Black Box uses a familiar survivors-like framework to keep the learning curve gentle.
What players do: fight on the battlefield, gather resources, and pick materials to combine.
Where AI steps in: it acts as a real-time “Designer Agent.” When you combine materials, the Designer Agent reads the semantic intent in real time and assembles atomic mechanics (trajectories, physics, elemental effects, damage parameters) and wires up the VFX on the spot.
You might end up with something never explicitly coded, like a “gravity trap” or “chain lightning”, and AI interprets that logic on the fly, turning it into in-game numbers and effects.
Which AI models are you using behind the scenes?
We built a proprietary model trained specifically on gameplay mechanics, not a general-purpose LLM repurposed for games. The model is purpose-built to output balanced, playable weapon logic in real time, with inference fast enough to feel seamless mid-combat.
We chose to develop in-house because off-the-shelf models don’t understand game-design constraints the way a specialised system can. They can generate something technically interesting, but they lack the intuition for what actually feels fun to play.
The idea of generating weapons and mechanics from text prompts is very powerful, but also potentially chaotic. How do you ensure the results feel coherent, balanced, and fun rather than random? How do you keep everything consistent with the world, and avoid losing the sense of authored design that gives games their identity?
Our approach pairs real-time logic generation with dynamic assembly on the presentation layer.
Keeping chaos in check: AI outputs a logic core. If someone wants a “40-metre greatsword,” AI takes existing assets (say, a curved VFX) and stretches, retunes, and reassigns their attributes based on semantic intent, so results stay grounded in what the game already owns.
Preserving authored design: we don’t put long, fiddly text prompts at the centre of combat. In the main loop, players build by choosing materials, so we carry the burden of generation complexity and balance tuning, while players explore AI-driven possibilities through simple, intuitive choices. That’s how we keep pacing and identity feeling designed rather than improvised.
Games like this could mean players become co-creators in a sense. Do you see this as a new genre emerging, or an extension of existing sandbox and roguelike design?
It’s more like pushing the frontier outward.
Black Box may wear a roguelike shell today, but it points toward a future of extreme freedom - closer to Westworld than a hand-authored branch chart: interactions that emerge from an AI-driven substrate rather than designer-written trees.
It’s not about replacing traditional 2D or 3D games but about opening a new pattern where players and AI weave the experience together.
Meshy has many millions of users on the tooling side. How does this game connect back to your core platform? Is it a showcase, a testbed, or something that feeds directly into your product roadmap?
Black Box is both a testbed and a preview of where Meshy is headed. The current build is 2D-focused, deliberately so, because we wanted to validate the core AI gameplay logic before layering on production complexity. But we’re already moving toward 3D, and that’s where the connection becomes very tangible: in future versions, when a player combines materials and the Designer Agent assembles a new weapon, Meshy’s platform will generate the corresponding 3D assets in real time – models, textures, animations – rather than pulling from a pre-built library.
So the game isn’t a side project; it’s a live proving ground for the next generation of our platform capabilities. Everything we learn about real-time generation under gameplay constraints – latency budgets, visual coherence, style consistency – feeds directly back into improving Meshy’s core 3D pipeline for all our users.
Does Black Box: Infinite Arsenal represent the beginning of a broader pivot into game development? Or even the start of a games publishing label for Meshy?
Absolutely. Black Box: Infinite Arsenal is the debut title from Meshy Game Studio, our dedicated game development and publishing arm. The studio was created to explore what becomes possible when generative AI isn’t just a production tool but the engine behind the gameplay itself.
“AI generates the logical core and mechanics of weapons in real time, opening a design space traditional procedural approaches can’t fully enumerate”
Robin Qin
We’re already exploring what the next titles look like, with each one designed to push the boundary of AI-native experiences in a different direction. Black Box: Infinite Arsenal is heading to Steam later this year as Meshy Game Studio’s debut title, and it’s just the beginning.
What is your own background? Where did your journey as a game developer begin, and how long have you been at Meshy?
I’ve spent most of my career designing large-scale live-service games – MMOs and SLGs at Tencent and FunPlus – where you learn quickly that the boundary between “designed content” and “player-generated chaos” is where the most interesting things happen. That tension always fascinated me.
When I saw what Meshy was doing with generative AI, it clicked: here was a team that could actually push that boundary wide open. I joined about a year ago and am now the lead game designer on Black Box: Infinite Arsenal. It’s the kind of project I couldn’t have built at a traditional studio because the AI infrastructure simply didn’t exist before.
Further down the rabbit hole
Some useful news, views and links to keep you going until next time:
ComfyUI has raised $30 million in a funding round led by Craft Ventures, Pace Capital, Chemistry, and TruArrow, at a $500 million valuation.
Google DeepMind is hiring game makers in the UK: job ads for Game Designer, Product Manager, QA, Research Engineer and Software Engineer went up last week.
OpenAI released ChatGPT Images 2.0 last week, to much acclaim.
Ace, Sony’s AI-controlled ping pong robot, has defeated top table tennis players.
Anthropic revealed Claude Design. Using its most capable vision model, Opus 4.7, the tool (currently available to paying users as a research preview) will build app prototypes, pitch decks, marketing assets and more. It exports to Canva, PowerPoint and PDF, but looks more like a threat to Figma and tools like Lovable.
Speaking of which, Lovable responded to revelations that its “public” setting made entire chat histories visible to other users, not just the finished product.
Meta is reportedly cutting 8,000 jobs as AI spending surges to $135bn.
Firstpoint has launched with a €50 million target to invest in AI-driven gaming and entertainment startups. The new venture capital firm focuses on emerging markets in Türkiye, Eastern Europe and Asia. The firm is led by industry veterans Burak Yılmaz and Mike Fischer.
Jordan-based publisher Tamatem has acquired Playable Factory from Istanbul. It’s a move that sees the company step into the AI-powered ad tech space.
Tamatem dominated in last year’s MENA Games Industry Awards, picking up trophies for Best Arabic Localisation, Best Publisher, and Middle East Game Of The Year (for mobile RPG Land of the Brave).
