As the Generative AI Revolution in Games progresses, it will completely reshape User Generated Content (UGC) creating a world where anyone can build games and expanding the games market beyond what many thought was possible. In the coming years, deep technical knowledge or artistic mastery will no longer be essential skills needed to develop games; instead, creators will be limited only by their energy, creativity, and imagination. Ideas won’t be cheap; they’ll be precious. Most importantly, game creation will become truly democratized and millions of new game-makers will be minted.
Ideas are cheap. It’s only what you do with them that counts. – Isaac Asimov
Based on the history of UGC platforms, the recent availability of consumer-facing LLMs, and observations about previous technological shifts, we believe that the evolution from UGC games to AI-powered UGC (what we’ll refer to hereafter as AIGC) will occur in two phases.
In this blog post we’ll cover the history of and lessons learned from existing UGC platforms, a market map of where UGC companies are today and a framework to think about them, what we expect companies to be building and how they will compete in AIGC’s first phase (AI-powered tooling), and how companies might emerge in AIGC’s second phase (AI-powered engines).
The world of gaming has undergone a tectonic shift in recent years with the rise of UGC platforms like Roblox and Minecraft (56M DAU and 17M DAU respectively). These platforms have enabled millions of people to experience the thrill and challenge of making virtual experiences and games for others by making creation tools more accessible. The games built have scaled with the power of the tools, now rivaling professional development teams (see gameplay from Roblox Ultimate Paintball vs. January’s Roblox Frontlines below).
How did these two platforms become dominant players in the UGC space? Under the hood, Roblox and Minecraft are very different products and they took very different paths to grow. Both, however, are rooted in the history of video game mods – dating back to the community of hackers that just wanted to bring their own ideas to life in the games they loved.
One of the first popular mods was Castle Smurfenstein in the early 1980s, a mod of Id Software’s popular Castle Wolfenstein game. Id followed with the release of Doom in 1993, which included a WAD file package for Doom’s maps, sprites, textures, assets, etc. And of course, there’s Counter-strike, the most popular mod for Valve’s Half Life, and Defense of the Ancients, itself a mod of the popular Warcraft game and the precursor for Riot’s League of Legends. The tradition is still well and alive today; Skyrim, a game that launched in 2011, has over 60K mods with billions of downloads.
Mods require a more sophisticated understanding of the game’s underlying architecture and programming, but Roblox and Minecraft simplified and abstracted the game creation process. Roblox launched in 2006 as a UGC games platform for younger children, spawned by founder David Baszucki’s insight that some of the educational physics tools he built were used to create games. Roblox was designed to be composable, with shape primitives imitating lego blocks and a simpler scripting language in Lua. In comparison, Minecraft drew players into its creator loop through its core building and survival gameplay. Minecraft started in 2009 as a simple indie game called “Cave Game,” created by Swedish programmer Markus “Notch” Persson, who drew inspiration from base-building and block-mining games. As the Minecraft tooling became more powerful, players created grand cityscapes like this one of Minas Tirith from The Lord of the Rings.
There are a few lessons that can be drawn from Roblox and Minecraft, though both had different origin stories, go-to-market strategies, and corporate outcomes:
We can extend the lessons from these two games to create a structure by which to evaluate current and future UGC platforms, depending on how open, platform-first they are like Roblox and how on-rails, game-first they are like Minecraft.
Despite their different approaches, both games are also highly composable and horizontal, which allows for different games to be created across genres (fighting, MOBA, racing, etc.). However, there are some UGC games that are highly verticalized for a specific genre. In Halo Forge for example, gamers are encouraged to create levels and rulesets but within the mechanics of the Halo game. Another example, Roll20, is a customizable platform tailored specifically for table-top role-playing games (TTRPGs). The map below segments many popular UGC platforms by how vertical vs. horizontal they are, and whether they’re game-first or platform-first.
AIGC’s first phase will represent a transition from UGC 1.0 to AI-powered creation, where generative AI will be used to dramatically improve existing UGC workflows. There are two main reasons why a transition phase will occur:
First, the generative AI space is still evolving rapidly – Large Language Models (LLMs) only recently became good enough to meaningfully improve text and 2D asset workflows, and 3D asset models are still a work in progress. As a result, the first wave of AIGC platforms will likely be built flexibly, as the infrastructure layer changes over time (see below). Second, initial tools will also likely be built as evolutions or optimizations of existing toolsets and UI. Incumbents like Roblox are incentivized to streamline rather than completely transform its existing creation pipeline, and startups may choose to take the path of least resistance rather than teaching new development paradigms to creators.
UGC incumbents are already exploring adding generative AI capabilities to their toolset, with Roblox adding generative AI tools to Roblox Studio. Details are scant so far and Roblox faces multiple challenges including innovator’s dilemma and tech debt that has accumulated for more than a decade. But they also have significant scale advantages with their creators, players, and development teams. What will companies in this transition phase be focused on building? And what will they need to succeed?
If AIGC’s first phase is generative AI accelerating existing UGC tools, the second phase will be generative AI powering the underlying creation engine. We predict that creation engines built from the ground up for generative AI might enable new creation paradigms and UX, have custom rendering capabilities, and/or be built with a programming language made specifically for AI-powered creation. These AI native engines may be cloud-based with reimagined technical and data architecture oriented towards rapid iteration and creation on any device at runtime. As a result, it will be extremely difficult for today’s UGC incumbents to win in this phase – they would have to rewrite all of their underlying technology and port their existing entrenched ecosystem over! So what are the potential paths that startups might take? We’d like to postulate two potential avenues – as with UGC 1.0, a vertical path and a horizontal path.
Companies that take the vertical path will have a narrow initial scope. This focus could take the form of a purpose-built set of creation capabilities for a specific genre of game (serviced by a specific subset of creators). Companies like Hidden Door (storytelling games), Roleverse (tabletop RPG games), and Regression Games (competitive battlebot games) are initially building focused creation tools around single genres. A narrow focus affords the opportunity to ship product, acquire users, collect feedback and ultimately achieve product market fit more quickly which in turn allows for purposeful, feedback-driven reinvestment into building better tools. Building purpose-defined tools in a narrower sandbox makes creator onboarding easier, but these same creators may struggle (and eventually churn) when trying to expand beyond those boundaries into new genres. But, because of their depth in a specific genre, they might be able to reinvent the creation process as Minecraft did with digital lego blocks.
Horizontal AIGC startups will look like game engine companies. Bringing generative AI capabilities down to the foundational infrastructure layer enables novel creation workflows and tooling. If a new search engine were built today, it would likely start with an “answer a user’s question” rather than an indexed keyword paradigm. AIGC-enabled game engines may entail similarly fundamental changes to game creation; what if these new AIGC engines make Scene Graph paradigm obsolete for instance? Just as we see creators starting to generate video without traditional animation software and rendering pipelines, new techniques will likely arise which may displace real-time rendering as we know it today. Or take asset creation – companies like Luma Labs are building novel 3D scanning and asset generation technologies that could end up powering new game creation engines. What if, instead of text-prompts, we could take a video of a space and AI would automatically generate the meshes, textures, and fully rendered level in a game? The horizontal path is the most risky and will likely require more capital and stronger research teams but is also the path that can completely redefine game creation writ large.
Generative AI is set to transform and disrupt the UGC games space by democratizing game creation. Everyone is a gamer at heart, and every gamer can become a game maker. The AIGC era will empower millions of people to make their first game and this new generation of game developers will unleash a wave of game design creativity that will change the games industry forever. More creators, more diverse games, more gamers.
If you’re a founder excited to build generative AI tools that will unlock this new generation of creators, feel free to reach out!
Joshua Lu is a partner on the A16Z GAMES team at Andreessen Horowitz, where he focuses on game studios and the intersection between games and social applications.
Robin Guo is a partner on the A16Z GAMES team at Andreessen Horowitz, where he focuses on web2 and web3 studios, games adtech, AI applications, and the intersection of healthcare and games.