At a GDC 2025 booth, Julian Park heard a developer say something that stuck: "ChatGPT can't see how 1,000 scripts interact. It's like fixing a car blindfolded." Park had already built the answer. He just wasn't done shipping it yet.

Park is the CEO and co-founder of Bezi, the Unity AI assistant that indexes your entire project - scripts, assets, scenes, packages - before it touches a single line of code. Not a generic code copilot. Not a chatbot with a Unity theme skin. A tool that knows what's in your scene hierarchy, which packages you're using, and how your thousand scripts interact. That specificity is the whole bet.

"We're all fighting the same fires, just with different tools."
- Julian Park, after GDC 2025

Oculus to the Garage

Before Bezi, Park spent years inside the machine. At Facebook he built design tools in WebGL and WebRTC - the kind of graphics-heavy browser work that most engineers avoid. Then he moved across the campus to Oculus, where he led VR multitasking and UI framework projects. That's not a glamorous role. It's the part of VR no one writes think-pieces about: making sure multiple windows stay coherent when you turn your head. Getting that right requires understanding exactly how software behaves under novel constraints.

When Park, Cecilia Uhr (now Bezi's design lead), and Denys Bastov (CTO) decided to start a company together, they did it with eyes open to a specific gap. Figma had changed 2D design. Unity was powerful but gated behind engineering. Nobody had built the obvious middle ground for 3D and spatial work - and having all worked at Oculus, they'd felt that gap personally.

"Figma excelled in 2D design but lacked the necessary features for 3D, and Unity, while powerful, posed challenges for designers without engineering assistance."

- The founding thesis, per Uncork Capital

Uncork Capital saw it. They led a $3M Seed round in December 2021. Benchmark, one of the most respected VC firms in software, led the Series A in November 2023, bringing total funding to $13M. Benchmark's Chetan Puttagunta doesn't back a lot of companies. The fact that he backed this one means something.

The Pivot That Wasn't Really a Pivot

Bezi started as a collaborative 3D design and prototyping platform - a Figma for spatial computing, built with Apple Vision Pro's emergence in mind. The company spoke at HarvardXR, Stanford XR, DICE, and GDC. Park ran hands-on workshops showing designers how to build interactive 3D prototypes without writing a line of code. The tool found an audience.

But the deeper story was always about eliminating friction between idea and execution. When AI capabilities accelerated, Bezi evolved. The product today is a Unity AI assistant with Actions (it operates directly inside the engine), Vision Tool (it can see your game), and Workspaces (cross-project context for teams). The original instinct - make 3D creation accessible - didn't change. The surface did.

"An artist, building tools that improve his own workflow, without asking anyone for permission."
- Julian Park, on what Bezi makes possible

The Quality Ceiling Argument

In March 2026, Park published a piece called "The Quality Ceiling of Games Is About to Rise." It's worth reading as a statement of what he actually believes, beyond the usual AI hype cadence. The argument isn't that AI writes better code. It's that AI makes experimentation cheap enough that more ideas get tested.

Park points to three games: Breath of the Wild (Nintendo built a 2D prototype to test physics before touching 3D), Hades (two years of player-driven Early Access iteration), and Astro Bot (a third of development time devoted purely to prototyping). What do they share? The bandwidth to run experiments most teams never get to run.

Most studios don't kill good ideas on purpose. They kill them because someone asks "do we have the bandwidth?" and the honest answer is no. Traditional prototyping cycles take days or weeks, which means teams commit early and settle for adequate instead of exceptional. Park's thesis: when prototyping becomes as fast as conversation, creative quality compounds.

The Lost Arcade case study is the proof point he leans on. The studio achieved 2x engineering output and slashed tool development time by 95% using Bezi. More specifically: a non-programmer artist started building their own workflow tools, without waiting for an engineer, without asking permission. That one anecdote does more work than any benchmark chart.

What He Heard at GDC 2025

Park's GDC 2025 writeup is a methodical survey of developer pain. He identified four recurring problems across studios of every size: asset and documentation chaos (one consultant called their Notion "a graveyard of outdated tutorials"), debugging inefficiency (consuming 20-30% of development cycles), generic AI producing code that doesn't fit specific codebases, and brutal onboarding when contractors cycle in or senior engineers leave.

The observation that unified all of them: the scale of the problem doesn't change the nature of it. Indie studios and AAA shops fight the same fires. Bezi is built on that insight. Context-aware AI - AI that knows your project - is the answer that works at every budget level.

The Founder Who Shows Up

Park has spoken at HarvardXR, Stanford XR, DICE, and GDC. He writes for Bezi's blog. His Twitter handle - @julianparc rather than @julianpark - has the energy of someone who registered it slightly off and decided to commit. He built companies with his Oculus colleagues, which is less common than it sounds. The shared context of building something as unusual as VR multitasking at scale creates a trust baseline most founding teams spend years trying to manufacture.

Bezi is SOC 2 Type 2 certified and makes a pointed guarantee: your game's IP and data are never used to train their AI models. In a market where that's far from standard, it's a deliberate positioning choice aimed squarely at studios that can't afford a leak.