Hello there! Welcome to this week’s AI Gamechangers, your regular look at the people, tools and ideas bringing practical AI into the games industry.
This week, we’re speaking with Andy Tsen, CEO and co-founder of Ramen. The company began life as the VR studio behind Zenith, but has since channelled those lessons into Aura, a multi-agent AI assistant designed to help developers build faster inside Unreal Engine and, following its acquisition of Coplay, Unity too.
Andy’s perspective is especially interesting because this isn’t a theoretical pitch from outside the industry. It comes from someone who has shipped games, wrestled with production pipelines, and is now trying to build the tool he wishes he’d had. In our Q&A, he discusses autonomous agents, keeping creators in control, why AI needs to handle the messy cross-disciplinary nature of game development, and why taste may soon matter more than technical ability.
As ever, scroll beyond the interview for a round-up of useful AI news and links from around the web.
Andy Tsen, Ramen VR
Meet Andy Tsen, CEO and co-founder of Ramen, the company behind Aura, a multi-agent AI assistant for game development.
Andy previously co-founded Ramen as a VR studio, creating Zenith: Nexus, before pivoting towards AI-powered development tools. Aura now helps developers work across Unreal Engine and Unity, assisting with Blueprints, C++, editor manipulation, 3D assets, animation, rigging and more.
In this interview, Andy explains why game development needs specialist AI agents, how automation can remain creator-controlled, and why AI may make taste more important than technical execution.
Top takeaways from this conversation:
Ramen’s pivot from VR to AI was driven by direct production experience: Aura was built as the tool the team wished it had all along.
Aura uses a multi-agent approach, with specialist agents for areas such as Blueprint generation, C++, asset creation, animation and so on, all coordinated by an orchestrator.
Andy argues that automation does not have to mean losing creative control. Developers can approve or reject scripts, files and Blueprints, or use Aura in a more focused side-by-side workflow.
The long-term shift may be from technical execution to creative judgement: as AI handles more of the pipeline, Andy believes “taste” will become a greater barrier than technical ability.
AI Gamechangers: Ramen started life as a VR game studio behind Zenith. Please describe the moment you realised you wanted to pivot from making games to making AI tools – how did it come about?
Andy Tsen: Our mission at Aura is to help the game industry responsibly and ethically transition into AI-assisted game development workflows.
We care passionately about this because we believe AI has the potential to help studios like we used to be create incredibly ambitious and expansive games. We’re building Aura to be the tool that we would have loved to use on Zenith.
At the same time, we can acknowledge that AI is having some very disruptive effects on the industry. If we take as a given that AI in games is inevitable, what can we do to make that transition as positive and empowering as possible? That’s the question that gets me out of bed every morning.
In simple terms, what does Aura do? And why does your approach go further than what game engines like Unreal and Unity themselves are offering with their own AI features?
Aura is a fully autonomous agent that allows game developers to build entire features and games with prompts faster and more efficiently than ever before.
The level of functionality and intelligence it currently offers is beyond anything we’ve seen in a first-party engine to date. It works outside of your game engine, and in many cases, you don’t even need to have the game engine actively open for it to be building for you. Think of it like Visual Studio, Claude Code, or Rider (which we all integrate seamlessly with, too). It’s a platform that plays well with your game engine, but lives outside of it.
We are big fans of both UE and Unity, though, and are excited by the prospect of collaborating with both of them to make Aura even more powerful.
You describe Aura as a multi-agent AI assistant. For readers who might be familiar with single-purpose AI tools but not the multi-agent approach, can you explain what that means in practice? Why is this important for game development specifically?
Inside Aura, we have agents that are specifically good at Blueprint generation, C++, editor manipulation, 3D asset creation, animation and rigging, and more.
We also have an orchestrator that keeps the plan coherent; the specialists do the work.
“The next two years will see AI able to create games of real substance, and taste will become a much greater barrier than technical ability”
Andy Tsen
This matters for games because game development is inherently cross-disciplinary. Even a simple gameplay feature touches code, visual scripting, assets, and editor state. A monolithic model that tries to do all of that equally well ends up mediocre at everything.
The Aura 12.0 beta introduced several new capabilities (Telos 2.0, the Dragon Agent, animation and rigging tools, etc.). Which of these are you most excited about? Do you have examples of how a developer has used one of them?
I think the Python Agent is the biggest unlock in terms of pure functionality. Because the Python Agent can do almost anything a user can do in-editor, and because models are so good at Python, it’s able to do a lot of stuff out of the box that we were surprised at.
Examples of workflows where the Python agent would save you days or weeks of time are when you have a game in production with many assets, and you need to make batch edits to levels, or assets that previously would have required days of manually clicking or multiple production engineers to write one-off scripts. That process now takes minutes.
“The scope of what Aura’s capable of has grown tremendously in the last month. We can now generate materials, build procedurally generated levels, and verify features in-editor”
Andy Tsen
I am also proud of the collaboration we did with Accelbyte, in which we were able to showcase enterprise-level backend integrations that might have taken weeks previously in a single day.
The Dragon Agent enables “autonomous agentic loops” that can construct full gameplay systems without Unreal Engine even being open. How do you balance automation with keeping the creator in control?
By giving them the toggles that they need to stay in control. For example, if you want, you can approve or reject every single Python script, C++ file, or blueprint that Aura creates. While it’s possible to go fully agentic with Dragon, it’s also possible to use Aura in a more scoped, traditional sense, iterating side by side with it on specific problems, or even having it help debug your own code.
Ultimately, it comes down to preference, and we want to support both.
How are you training and refining your models to achieve the best level of accuracy with engine-specific code?
Most of it comes down to evaluation discipline. We’ve built extensive internal evals that measure engine-specific quality: does the Blueprint compile, does it run, does it do what was asked, does it follow the project’s conventions. That eval layer is the only way you know you’re actually getting better and not just moving a metric. There are many proprietary techniques that we employ, including traditional ML techniques as well as ways of encoding knowledge to ensure we pass those evals.
Your team has AAA veterans from Activision, Blizzard, Riot, and Ubisoft alongside hackers and modders. How does that mix of backgrounds shape the way you build Aura? Does having shipped games yourselves change the way you think about what developers actually need from AI?
For sure. When you are working with the best game developers in the world trying to push the envelope of what’s possible, you just think about problems differently. For these companies, it’s about creating perfection, not cutting corners or saving costs.
Practically, what that means is that all our tools are subject to much higher scrutiny, since they can’t just be used for prototyping to be useful for AAA developers.
“We have agents that are specifically good at Blueprint generation, C++, editor manipulation, 3D asset creation, animation and rigging, and more. We also have an orchestrator that keeps the plan coherent; the specialists do the work”
Andy Tsen
Professional developers also want an incredibly high degree of control over their tooling. It’s like the difference between a consumer playing with Midjourney to generate “cool art”, to professionals using AI tools in Photoshop or Illustrator to dial in their meticulously crafted concept art.
Do you personally come to all this from the technical side or the business side? How did you get started with games, VR, and Ramen? Was it always an ambition of yours?
Technical. I came up as a developer, and I still do the work. I think you lose something as a founder if you stop.
Games have been the thing since I was a kid. I modded before I coded, coded before I shipped, shipped before I had any business running a company. VR pulled me in during the early Oculus days, and Zenith came out of that obsession. Ramen existed to make Zenith. Aura exists because making Zenith showed us where the whole pipeline was broken.
Ambition is a funny word for it. I never sat down and said, “I’m going to start a company.” I sat down and said, “I want to make this game”, and the company happened around the work. I think that’s how most of the good ones start.
Looking ahead, where do you see AI fitting into the broader 3D production pipeline? And what’s next on the roadmap for Aura?
Production-ready 3D assets will soon become the norm. Beyond that, the tools around rigging, animating, and rendering them will get much more sophisticated. You will see the end-to-end art pipeline change completely, where individual 3D artists, animators, and concept artists will be able to do 10-20x more than they did before. A team of 10 artists can act as a team of 50-100 under this new paradigm.
“While it’s possible to go fully agentic, it’s also possible to use Aura in a more scoped, traditional sense, iterating side by side with it on specific problems”
Andy Tsen
As for Aura, our latest releases show a lot of what we’re pushing for. The scope of what Aura’s capable of has grown tremendously in the last month. We can now generate materials, build procedurally generated levels, and verify features in-editor using our verification agent.
Agentic loops are still far and away the most important thing, though.
Please gaze into the crystal ball! What further disruptions are likely to come from AI in the year ahead?
World models become increasingly important, not for world generation but for spatial reasoning. We see some combination of the two lineages (WM, LLMs) converge, perhaps even trained end-to-end, to massively increase the general intelligence of models and robotics.
Game engines and agents like Aura help fuel this transition because games are the best real-world simulators outside of the actual real world.
“If we take as a given that AI in games is inevitable, what can we do to make that transition as positive and empowering as possible?”
Andy Tsen
For our industry, fully “vibe-coded” games become possible that express much higher complexity and quality than the games we’re seeing today. Right now, an AI game developer is like preschoolers drawing with finger paint. The next two years will see us graduate into first-year art students, where AI will be able to create games of real substance, and taste will become a much greater barrier than technical ability.
Further down the rabbit hole
Some useful news, views and links to keep you going until next time:
GitHub has announced that all Copilot plans will transition to usage-based billing from 1st June.
Capcom’s game Pragmata has inspired a fake controversy about AI art, writes Abbas Saleem. The android character in the game draws a picture - actually, the picture is handmade by Capcom artists - which has led to amusing accusations of AI art from the community.
Upcoming event: if you’re in Ho Chi Minh City (Saigon) this week for the GameVerse event, why not attend the free Pocket Gamer developer mixer on Friday night, 8th May?
Upcoming event: the GameChanger TLV conference dedicated to generative AI in games takes place on 10th June.
A quirk in the personality setting for OpenAI’s Codex caused it to become briefly obsessed with goblins and other mythical creatures.
Roblox has unveiled Roblox Reality, an AI-driven photorealism upscaler that combines its existing engine and cloud with Video World Models to generate high-fidelity visuals on top of standard Roblox titles.
Razer has expanded its free, open-source AI development toolkit AIKit to support image, video and audio models alongside Arm64 architectures, enabling developers to run omni-modal AI locally across hardware, including its Blade laptops and NVIDIA’s DGX Spark.
The toolkit also powered Razer’s recent AVA Mini April Fool’s campaign in partnership with Akash Network, generating over 11,000 personalised pet images at $0.01 each (up to 15x cheaper than standard cloud APIs).
Infinity Fiction has announced a £250k pre-seed round to launch Tidefall, a dark fantasy RPG, alongside its proprietary Living World AI Engine. Tidefall enters pre-alpha today with a wider alpha launch on 1st June. Look out for a full conversation with the team in AI Gamechangers very soon.
Ben Horwitz built Sinceerly, an app that rewrites your AI-drafted emails to sound more human 💀