Epic Games Turns to AI Coding for the Unreal Engine

(spectrum.ieee.org)

45 points | by rbanffy 4 hours ago ago

50 comments

  • vblanco 2 hours ago ago

    As much as epic seems to like AI coding, they are still failing at the basics. I develop games with them, and know other devs who also do, all of us using AI extensively. None of us use the official unreal engine MCPs and similar tools, because they are simply almost useless. We all run our own, and deeply optimize the engine.

    So many MCPs and others and they still dont have a good way of having text assets, which lets the AIs properly use their agentic greps to find things quickly.

    The engine is also a behemoth, which causes problem due to sheer bloat that slow down the iteration cycles, and that becomes the key bottleneck. Half of my dev friends are just swapping to godot, where its a smaller and leaner engine and we can just implement whatever feature we miss, but with much faster iteration loops.

    • thunky 2 hours ago ago

      > None of us use the official unreal engine MCPs and similar tools, because they are simply almost useless

      Curious if you've tried 5.9 preview yet and if that might change your take at all?

      • vblanco an hour ago ago

        We are still upgrading to 5.9 but for what we are seeing, no point to it, it wont work better. Our MCP main trick is that its a asset-exporting step that writes text versions of every blueprint to allow the bots to read the asset data much more easily, and then for the writes those are done by just exposing the editor python scripts. Works very efficiently.

      • ftchd 2 hours ago ago

        Probably not.

        They already made their own workflow which presumably works much better for them.

    • chiengineer3 an hour ago ago

      In the process of creating a new 3D runtime that is rebuilt from the ground up for web - works but not ready

  • _spduchamp 2 hours ago ago

    I recently spent a couple days making a music video inside Godot engine using Claude, and in the end it was a perfect implementation of the vision I had in my head, but it also gave me a weird feeling. Is it MY art? It doesn't feel right, so I tossed out the project and I'm not releasing it.

    I will from now on, keep AI out of the creative pipeline. Sure, I'll use it to brute force tune biquad filters and other low level grunt work to build tools, but then put the AI away, and focus on the art of making.

    • FunkyMushroom an hour ago ago

      > Is it MY art?

      The answer's in your post:

      > in the end it was a perfect implementation of the vision I had in my head

      If you are doing conceptual art, then it is quite your art. Duchamp's Fountain is not any less Duchamp's because the urinal was commercial off the shelf.

      But if you're not doing conceptual art, then almost certainly not. Art is not a vision astrally projected out of the brain, it's forged in the artistic process, and that inevitably changes the ideas. The restrictions of the medium and the act of creation are just as much part of the art as the original concept.

      This is what is meant when people call AI art "soulless". There is nothing in it, merely the empty shell of an idea. People look into art and expect to see the imprint of another human, of their struggles against the medium or glorious success with it. To be the audience is to be in conversation with the artist, whether that is theatre or the cave paintings of myriad years ago.

      The problem with AI art is that unless you go through all this effort, the effort one seeks to avoid by reaching for AI, all you make is some conceptual art. "Huh, neat idea". You might as well just print the prompt on a piece of paper. An empty shell so devoid of everything there is not even an abyss to stare back.

      • kirth_gersen 39 minutes ago ago

        "An empty shell so devoid of everything there is not even an abyss to stare back." Very nice turn of phrase there.

    • kirth_gersen 35 minutes ago ago

      I think that the generation of art and code end up feeling fundamentally different. Code is more of an engineering task to optimize a solution under a given set of constraints, so it tends to converge towards some optimal solution. Whereas art is about taking the constraints and diverging from the starting point in some interesting way. Which is why AI feels bad and empty, it is converging rather than diverging in some individualistic way. I think that's why there wasn't similar outcry against procedural generation, because pg was fundamentally about diverging in an interesting way rather than converging to a similar mush.

    • worried_individ an hour ago ago

      is it really a painters art when they use a brush that was manufactured by someone else? the exact color mix they picked up in a shop somewhere? when they paint on a canvas which as a resource has been standardized by a group of people they don't even know?

      art is modified on a range from basically not at all to quite a lot depending on the tools you use, and nobody is in full control over the entire process.

      I guess you have to decide for yourself which parts of the result are actually important to you.

    • ftchd 2 hours ago ago

      How does a music video in godot look like?

      I guess like any 3d scene in godot but when thinking about a music video I imagine something specific (more or less).

      Do you have a snippet or something?

    • setr an hour ago ago

      Isn’t art all about taste? And taste is really curation. It is in deciding what not to present that it becomes yours. I don’t think the invocation of the AI matters, unless you’re so tasteless as to have made few/no decisions in the process (ala vibe-coding). Only then have you presented the tasteless taste of the machine as your own tasteless taste ( or your taste is indistinguishable from the machine, a different shame )

      Though I’m of the opinion that selection and presentation of procedural art is perfectly owned as well, an apparently contentious position

    • TheRoque an hour ago ago

      I kinda have the same issue. I tried to vibe code some stuff, but so many decisions taken were not my own. I can question all of them but most of the game framework is already there so it will inevitably influence the result. Anyways, it helped me realize quickly that what I had in mind is not fun, and I don't know how to make it fun no matter how much I prompt it

      • andyfilms1 an hour ago ago

        I think this is not going to be a popular opinion on HN, but this ambiguity vanishes when you just reframe your relationship to AI. Instead of thinking, "I made this! Wait, did I make this?" I instead prefer to think "AI made this, with my instruction." It's not only more technically accurate, it also resolves that moral dilemma (for me, at least.)

    • acomjean an hour ago ago

      Yeah.

      A few friends like to play with procedural generation. Can it make art? It kinda can but to us it felt soulless (maybe because we knew how it was created?)

      I think about this older song a bit: it’s oddly heavy: “No prints can come from fingers if machines become our hands” Horizon has been Defeated -Jack Johnson

    • skeledrew an hour ago ago

      As AI becomes more prevalent, the concept of "ownership" will have to be updated. Or discarded.

  • Grombobulous 2 hours ago ago

    Start with fixing basic problems with the engine that show up in almost every game that utilizes it.

    I think the launch of GTA VI shows the value in having your own engine. I don’t believe that game would have been possible to make with Unreal in its current form, even with a small army working on it.

    Sure, comparing to the most expensive piece of media of all time is a poor comparison, but it’s not just Rockstar who sees benefit from their own engines.

    Rimworld just recently switched off of Unity and to their own custom engine, greatly improving performance.

    Epic is right to be worried about AI eating their lunch. Why bother with a third party company when the barrier for custom engine design is falling?

    • piker 2 hours ago ago

      Same reason as always? The barrier for creative expression is also falling. You can spend your time re-inventing something that is mostly solved to eek out that extra 1% if you think you're Rockstar, but if you want to focus on the 99% of value, buy don't build.

      [Edit: I strongly support and love folks building their own engines, too, for non-economic reasons, just suggesting the answer to the QP.]

      • Grombobulous 2 hours ago ago

        I agree with you here, but I am also thinking of large studios who have the level of output that justifies having their own engine.

        Many of those studios have still decided to use Unreal out of workflow convenience, ease of hiring, etc.

        But will that out-of-the-box platform advantage continue into the age of AI? We’ll soon find out.

    • tempoponet 2 hours ago ago

      This is a balanced take. The anti-ai folks will say it creates bugs they can't fix because nobody understands the code. But UE5 has terrain traversal stutter in every single game. Huge AAA games releasing across platforms, and they release it in that state and never fix it. A 5090 can't grind past it.

      The worst case scenario is already here, the optimistic take is that they have a new tool that might fix it.

    • djtango 2 hours ago ago

      I have been nerding out over Square Enix's story where they burnt a decade or two chasing engines and platforms only to wave the white flag and go to Unreal.

      In the same period of time Capcom managed to hone their engine and reap the benefits of having an in-house engine.

      More recently Chinese and Korean studios have seemingly popped up out of nowhere and my understanding is that UE massively lowered the barrier to entry to shipping AAA games.

      So China and Korea basically leapfrogged companies like Square Enix that basically spun their wheels for two gaming generations while they're also getting squeezed by games like BG3 and Ex33.

    • tancop 2 hours ago ago

      There's still value in using the same engine as everyone else so you get pre-made solutions for hard things like lighting, and don't need to fix all the problems yourself.

      Epic has been working on performance because they're very invested in The Witcher 4. CDPR moved off their custom engine because it could barely handle Cyberpunk and they partnered with Epic to build UE6 together.

      If the effort fails both companies will be in trouble, but if it's successful Unreal can fix its reputation and stay #1 for ten more years.

      • afavour 2 hours ago ago

        > There's still value in using the same engine as everyone else so you get pre-made solutions for hard things like lighting, and don't need to fix all the problems yourself.

        In fairness this feels like exactly the situation where AI would shine.

    • esperent an hour ago ago

      > I think the launch of GTA VI shows the value in having your own engine

      Your own engine and 1.5 billion dollars to spend on developing a single game.

      For those of us with merely a few tens to hundreds of millions, maybe Unreal Engine is a better choice?

      See for example CD Projekt who gave up on their own engine from Cyberpunk and switched to Unreal for future games. They're not exactly small or poor, and it's reasonable to say they're one of the most technically capable dev studios in the world, yet they still decided to stop rolling their own.

    • V__ 2 hours ago ago

      I am really curious how the next Cyberpunk will turn out performance wise, since they want to switch over to Unreal.

    • doctorpangloss 2 hours ago ago

      Rockstar and Rimworld both use LLMs to author engine code. Core parts of Rimworld are authored with LLMs (https://github.com/ahwootech/BRUTAL for example). Their port away from Unity was almost certainly facilitated by LLMs.

      Unreal Engine and editor sources are available, people modify them (like Valorant), almost certainly with LLMs at this stage.

    • sublinear 2 hours ago ago

      Of course an engine is important. It is just as much of an artistic expression as the assets. Looking to AI for answers might be even more misguided.

      You'd think that since so many hallmarks of AAA games (graphics, etc.) have stagnated, they'd also realize it's not so important to attracting gamers anymore.

  • ferguess_k 14 minutes ago ago

    Tim seems to be very positive on it. I don't have the credential to question him, but like it or not, this is definitely what the future looks like.

  • bemtevi 2 hours ago ago

    I've stopped consuming new media...pretty much anything after 2023 unless it's verifiable anti-AI. Luckily there are a LOT of games out there that are AI-free, like ROMs of old console games and stuff. It's just no longer fun to experience something that another person (or group of people) didn't make themselves.

    • Grombobulous 2 hours ago ago

      That seems incredibly compromising to your own enjoyment. There’s a lot of media that very specifically doesn’t involve AI in its production.

      GTA VI is a pretty good example, a literal army of humans made it.

      https://www.rockstargames.com/VI/credits

      Another example, games on Steam are required to disclose the use of AI, and even when the developer is fibbing, reviewers are quick to find it.

      • officeplant 2 hours ago ago

        >GTA VI is a pretty good example, a literal army of humans made it.

        And some of us are refusing to buy it because Rockstar's anti-unionization efforts and mistreatment of that "literal army of humans" that is still playing out in court[0]

        [0]https://www.eurogamer.net/gta-6-rockstar-tribunal-awaits-ver...

        • skeledrew an hour ago ago

          These collisions are funny. Employee costs rise due to things like unionization, companies move to AI so they can do more faster and avoid rising employee costs. Customers raise pitchforks over company anti-unionization and/or pro-AI efforts, and also complain about the cost of the product, or switch to smaller vendors doing things the way they prefer and complain about all that's missing that the other vendor products - which they boycotted - have in spades...

      • benoau 2 hours ago ago

        Gaming industry is full of carveouts to use AI without admitting AI is used.

        You can have AI one-shot an entire game for you but as long as you trace over the graphics and substitute the audio you don't have to declare anything on Steam.

        Take Two only said that they didn't use generative AI in GTA VI, but it sounds like they otherwise use AI wherever else it is beneficial:

        > We’ve been in the AI business and machine learning business since the beginning of time.

        > Obviously we’re well aware of this. We’re trying everything we can to make better games more efficiently and I have some confidence that that can occur (using AI).

        https://insider-gaming.com/take-two-addresses-ai-in-gta-6/

      • glitchc 2 hours ago ago

        Only time will tell if GTA VI is the new modus operandi or the swan song to a bygone era.

        • Grombobulous 2 hours ago ago

          It’s never going to be the modus operandi, since it’s the most valuable media franchise in the world.

          But it only takes a few counterexamples to make the idea of abstaining from new media seem silly.

          For example, if you believe that movies have turned to shit since they switched to digital over film, well, there are a handful of directors like Christopher Nolan who prefer film and so the idea that you’ll just never watch a new movie again over that issue doesn’t make a lot of sense.

          In fact, that absolutist abstinence actively harms the niche that you want to see succeed.

      • 2OEH8eoCRo0 2 hours ago ago

        Why would older media compromise enjoyment? Were people that were born in 1890 robbed of GTA VI?

        • Danox 19 minutes ago ago

          He’s confused. a great story will always win out slop. The same goes for great gameplay. You can put as many bells and whistles as you want if the game play isn’t any good it won’t last and it certainly won’t sell at least not for long, once word gets out.

          Gameplay Gameplay Gameplay. Story Story Story.

    • CuriouslyC 2 hours ago ago

      How's the air way up on that high horse?

  • skeledrew an hour ago ago

    Yes, this is the way. ArtCraft all the things so proprietary software owners wake up and start creating value again and reduce the enshittification of their products, or become irrelevant.

  • rob74 2 hours ago ago

    > Why Buy an Engine When You Can Build One?

    Maybe because you prefer to spend your tokens on building stuff that sets your game apart from others rather than on rolling your own engine?

    • dathinab 2 hours ago ago

      and because there are _a lot of subtle trick parts/unknown unknowns/etc._ and in turn a lot of chances for bugs you could have trivially avoided by re-using an existing engine/code.

      unknown unknowns are a huge pain when you blindly rebuild something as you _can't_ fix them as you _simply don't know_ that you messed up or even what knowledge you are missing.

      This are the kind of things where you roll out a perfectly working game, and it then has major crashes for a large part of you customers because of differences in the execution environment between testes and end users and your code accidentally having relied on something it shouldn't have. (E.g. lets say you have some subtle broken concurrency code, then having subtle different cpu pipelines/instruction timings/background os processes/antivirus/etc. can be the difference between "it always works" and "it crashes/bugs frequently". E.g. technically correct code which just doesn't optimize well even through it looks like it should. E.g. very common state management issues. etc.).

      A lot of that stuff is at least somewhat handled in existing game engines after hard won experience (accidentally shipping bad code => fixing it) and while you can "transfer" some of that know-how by pointing the AI at an existing engines code there are limits to how well that works.

      Anyway in the end the main point is still why high risk reinventing the wheel, when you can just buy it for a much lower cost, risk, option to buy additional support, and know-how dev can transfer from other projects?? (and/or the AI can transfer from other projects through careful crafted skills and similar)

      Naturally that only works if you don't charge too much for a game engine (but then it's not that there aren't lower cost/free options with some limitations sure, but starting with that is still a better choice then rolling your own engine).

  • Based-A an hour ago ago

    The irony of ripping out core systems devs are used to using and then admitting that owning your own engine to reduce dependence on a third party is really palpable. UE5 dented Epics credibility by producing "samey" looking games, now I can imagine that UE6 games are going to homogenize across all areas.

  • ozgrakkurt an hour ago ago

    Games built with Unreal already started to look like washed-out garbage. Hopefully this will further reduce the quality so we will see more well built games.

  • holoduke 2 hours ago ago

    As an experiment I am vibe coding a 3d engine where I try to implement things used in AAA games without ever looking at the code. The goals is to get GTA6 quality graphics. Its amazing what you can get already. But reality is that things are still very hard. For example a displacement effect like you see in a game called Crimson Desert. I think I have spend at least 1 month of max account Claude tokens. Hundreds of prompts. Still not having the desired effect. https://github.com/holoduke/vibe

    • skeledrew an hour ago ago

      The approach matters a lot. I don't think it should take that much if you have the agent first do deep research on the effect you'd like replicated (there should be papers, open source code, etc out there on it), and provide it with examples that it can use to verify that it's gotten it correct.

    • LetsGetTechnicl 2 hours ago ago

      Why tho

      • wilkystyle 2 hours ago ago

        > As an experiment

        Also, I think you might be on the wrong site if this is your mindset.

  • lovedaddy an hour ago ago

    Slow, clunky, unreal engine + slop. Winner.

  • avaer an hour ago ago

    The trend with AI seems to be away from bloated engines: in web, GUIs, and games. You really want to cut out the fat, use safer languages, reduce dependencies on companies, processes, and bullshit. Because that all gets in the way of the agentic loop, which is slowly becoming all that matters.

    So, I'm not convinced we'll need Unreal.

    Besides, "Do it the Unreal Engine way" often "just works": rendering arch to shaders to animations to actors, literally everything I've tried in the last month or so. Even a screenshot plus knowing what you're talking about can get you there these days with a frontier model.

    I'll offer an alternative (which Epic could adopt if they are smart): Nintendo

    Nintendo doesn't have an "engine". Their ModuleSystem is basically a set of libs that each game borrows from and customizes (it's very interesting to study all of the decomps and code evolution). Basically Nintendo will pick and choose and rip apart different pieces of their games to make what's best for the next game. Each game learns and borrows from the previous ones. But they are not beholden to maintaining some big fat thing called "engine" that's supposed to serve the entire gaming world (plus movie world, etc). I imagine this makes Nintendo perfectly set up for AI coding, where transmogrifying one game into another is how everything is done. But it really only works where everything about the game (including the assets) is "open" (which it presumably is within the Nintendo product org).

    • Danox 17 minutes ago ago

      Nintendo depends upon gameplay and more gameplay…