The Cherno Is Building a 3D Game Engine With AI

If you spent any significant time in the gamedev pipeline on YouTube — like actually trying to learn this stuff, not just watching to feel productive — there's a real chance Yan Chernikov (aka The Cherno) is part of your origin story. His original game engine series was the thing. Hundreds of hours of line-by-line C++ that turned "I want to understand how games actually work" from a vague dream into a concrete skill tree. The comment sections on those old videos read like a support group slash dojo. People still reference specific episodes in gamedev Discord servers. That's not hype — that's just the record.

So when he dropped a 28-minute video announcing a brand new 3D game engine series, built with AI, the reaction wasn't just "oh cool, new content." It was more like when a retired player announces they're suiting back up. The comments flooded fast. The Discord lit up. People started posting engine name suggestions before the video was even done processing. That parasocial debt runs deep, fr fr.

Here's what he's actually proposing — and it's more thought-out than a lot of the AI-in-gamedev discourse that usually floats around.

The Design of the Gap

Chernikov is explicit about what he's not building. A full AAA engine is off the table — not because of ambition, but because of basic math. Frostbite, Unreal, id Tech: these codebases have been shipping games on millions of different hardware configurations for well over a decade, accumulating fixes for race conditions and edge cases that you simply cannot manufacture from scratch. Frostbite first shipped in production around 2008 with Battlefield: Bad Company, and even that timeline barely scratches what "mature codebase" actually means. Chernikov says it plainly: "It is just impossible." He's right, and I appreciate that he doesn't dress it up.

But he's also bored by the other extreme. A minimal indie engine — render a cube, slap an editor on it, done — isn't interesting to him because, with his background, it isn't hard. What he's going for is the middle: a 3D engine that's genuinely simple but incorporates some of the architectural thinking that makes AAA systems good. The kind of thing where a junior dev watching can actually absorb how professionals think about structure, not just how to make pixels appear on screen.

That's a real and specific educational target. It's also genuinely difficult to hit. "Simple but architecturally honest" is harder to execute than either extreme, and it's going to be fascinating to watch whether the AI-assisted workflow helps or complicates that goal.

What "AI-Assisted" Actually Means Here

Here's where Chernikov separates himself from the discourse that's mostly just noise. He's not doing "AI writes game engine, man watches." He's not doing "I type every character myself, AI is cheating." He's designing a specific methodology, and it's more layered than a thumbnail would suggest.

The plan involves using Claude Code as the primary agent (unsponsored — he makes a point of that), writing detailed architectural documents for the AI to read before it writes anything, building validation tools and test scripts, and doing full code reviews of every line the agent produces. He's also going to let AI run loose sometimes, on purpose, to show what that looks like and why it fails. The whole thing will be 100% open source on GitHub, which means the community gets to audit every decision in real time.

"We will read every line of code that it writes and we will make sure that it is written the way that ultimately I would have written it," he says. That's a meaningful constraint. It means the series isn't a vibe-coded experiment — it's a controlled demonstration of a specific development philosophy.

The part I find most interesting: he's going to try putting himself in the shoes of someone with less experience. Not pretending to be a beginner — he acknowledges he couldn't validate what he can't verify — but slowing down and using the AI to surface reasoning and resources he'd normally skip because they already live in his head. Whether that actually produces different educational value, or whether decades of intuition just quietly override the exercise, is a genuine open question that the series itself will have to answer.

The Junior Dev Debate (Let Me Actually Take a Swing at This)

Chernikov spends real time in the video responding to audience pushback on AI and junior developers, and this is where I want to plant a flag instead of just mapping the terrain.

One commenter argued that juniors "will barely learn anything" using AI, that they'll just blindly accept generated code and the verification burden falls on everyone else. Chernikov disagrees, and so do I — but I want to be specific about why, because the disagreement matters.

The "juniors won't learn" argument treats learning as something that only happens through friction with the void. You struggle, you google, you struggle more, you eventually understand. That model was already broken before AI. Chernikov's counter-example is personal: as a 19-year-old university student in Australia obsessed with C++ and graphics programming, he had basically nobody to ask. His lecturers didn't have the specialization. Google searches for difficult low-level questions returned documentation and Stack Overflow threads that assumed you already knew 80% of what you were trying to learn. The situation only changed when he landed an internship on the engine team at EA around 2015 — suddenly he could ask questions and get real answers, instantly. "It was like a superpower," he says.

AI, at its best, is that superpower made accessible. A junior developer can now point a Claude agent at the Unreal Engine 5 source code and ask it to explain why specific architectural decisions were made. That was not possible before. That is a genuinely different world for people trying to break into a specialized field from outside an industry network.

The real problem — and this is the part worth sitting with — isn't that AI gives wrong answers. It's that you need enough existing knowledge to recognize when it does. A junior who doesn't know what good C++ architecture looks like will accept bad C++ architecture because it compiles and seems fine. The discipline gap is real. But the solution isn't to take the tool away; it's to understand that the tool has a skill floor, and part of the learning process is building that floor. Chernikov's series, done right, could actually be part of building that floor for people watching.

There's also a real thing he concedes, which I think gets glossed over too quickly in the AI-maximalist camp: "I don't feel like I own the code that the agent writes for me, which makes it difficult to even remember." That's not a minor complaint. The muscle memory of typing, the cognitive engagement of construction — those matter for retention. This series is going to have to figure out how to address that for viewers who aren't just watching but trying to learn, and that's not a problem Chernikov has solved yet. It's a problem he's walking into live.

What Hazel Was, What This Isn't

Quick context, because it matters for understanding the stakes. The original Hazel series — the 2D engine, built line by line on camera — was how The Cherno's community got formed in the first place. The 3D version of Hazel was built separately, off-camera, eventually with a small team under the Studio Cherno banner. Chernikov says that represents roughly seven years of development work before AI tools became central to his workflow (the series started around 2019; your mileage may vary on when "AI era" actually began for any given developer). That codebase now underpins Lucky Robots, a robotics simulator project. It's still accessible through Patreon.

This new engine is something different. It's not a continuation of Hazel. It doesn't have a name yet — that's actually a crowd-sourced decision, and the comment section is currently a graveyard of increasingly unhinged suggestions, which is correct behavior. The code will be free and open from day one.

Episode one is expected within the coming weeks, based on Chernikov's stated timeline, and he's signaled it might not even involve AI at all — just diagrams and architectural planning, OG series energy. The community already has opinions about what systems it should prioritize. Some of those opinions are coherent. Some of them are "add a Vulkan backend immediately." You know how it goes.

The name they land on is going to tell you something about how seriously the community takes this. I'm pulling for something that sounds like it could actually ship a game, not something that sounds like it was generated by the very AI being used to build it.

---

— Derek "D-Block" Washington, Gaming & Interactive Media Correspondent, Buzzrag