One person can also develop games, and classic old games can be revived by AI: The gaming industry is ushering in a new transformation.

Authors: Elena Burger, Anish Acharya, Jonathan Lai, a16z

Compilation: TechFlow

TL;DR: When disruptive technology enters an industry, it’s often the incumbent archetypes that get hit first. But in our latest discussion, the a16z team argues that AI will not destroy the status quo in gaming; instead, it will accelerate the evolution of three core forces: indie developers, tool makers, and world-building studios.

Drawing from two dimensions—“new games vs. existing games” and “reducing costs vs. unlocking new possibilities”—this piece maps out four quadrants of AI-empowered game development. Paired with real-world examples like Blender modeling, decompiling Halo into code, and personal projects reviving classic games, it showcases the emerging origin stories of a new generation of games as creative barriers fall.

If you'd like to hear more from Elena Burger, Anish A and Jon Lai on this topic, click here to listen to the companion podcast for this piece.

The Three Archetypes of the Gaming Industry

If you want to understand an industry, the best way is to study its most revered archetypes.

In the gaming industry, we start with independent developers. This category includes Eric Barone (creator of Stardew Valley), Jonathan Blow (author of Braid, The Witness, and Order of the Sinking Star).

Then there are small teams like Tarn and Zach Adams, co-creators of the epic world simulation game Dwarf Fortress.

People love these developers and their games because they manage to achieve commercial success outside the traditional studio greenhouse despite immense hurdles. They're like the “geek wins the girl” underdog stories of the gaming world. That these games succeed at all is part of their charm, of course, but they're also genuinely fun.

Next are tool makers with a prophetic spirit, epitomized by figures like Tim Sweeney and John Carmack. Through programming experiments close to the hardware, they constantly push past performance boundaries, expanding the creative space for everyone else. Their work births engines, editors, and simulation systems that eventually evolve into development environments for others to build upon.

Finally, there are world-building studios, represented by teams like Rockstar Games (Red Dead Redemption 2 and the upcoming Grand Theft Auto VI) and Larian Studios (Baldur's Gate 3).

They spend years crafting richly detailed, highly realistic game worlds packed with tiny easter eggs, whose depth often far exceeds what a single player can notice.

For example, wildlife in Red Dead Redemption 2 includes possums that play dead and bears that bluff charge, while GTA VI will feature six in-game radio stations, including one hosted by a fictional character voiced by Lana Del Rey and her husband. People adore these games precisely because they feel incredibly real and alive.

These archetypes point to different value propositions: indie developers immerse themselves in refining unique visions and achieve fame when those visions resonate; prophetic tool makers push technological boundaries; and world-building studios create deeply immersive works.

Typically, when a daunting disruptive technology enters an industry, it's these established archetypes that get disrupted first. With AI, however, that might not be the case. In fact, AI's most compelling value in gaming lies in its ability to accelerate the growth of all three archetypes simultaneously.

The same technology can help individuals turn wild ideas into reality, assist engineers in dissecting the underlying structure of games, and give large studios with hundreds of experts more room to experiment with new ideas during production.

This was a recurring theme in our recent conversation among the three of us (Anish, Jon, and me) about AI and gaming. Each of us has several games we've wanted to make but put on hold for years, held back by reasons ranging from lacking specific technical skills to struggling to maintain momentum on a massive project alongside a full-time job unrelated to daily game dev.

Lowering these barriers means we (and developers far more professional than us) can test more ideas, while also raising the ceiling for ambition on the projects themselves.

Four Quadrants of AI-Empowered Gaming

These opportunities unfold along two dimensions: whether AI assists in making existing games or creating new ones, and whether its primary contribution lowers costs or unlocks new possibilities.

For aspiring indie developers, the immediate opportunity lies in bridging the gap between “having a unique idea” and “possessing every skill needed to fully express it.” Even if someone excels at one part of game design, like scripting core gameplay loops, missing skills in other critical areas still prevents them from building the game they envision.

On our podcast, Jon describes how agents collaborate across disciplines: helping draft design docs, building assets in Blender, assembling scenes in an engine, reviewing screenshots, and troubleshooting issues. This allows solo developers to use tools from the “Produce” quadrant to quickly build playable prototypes, revealing where an idea is truly fun much earlier.

Of course, there's a lot of work between spotting a bright spot and shipping a polished game. Anish's record store experimental project (mentioned on the podcast), inspired by Habbo Hotel, stemmed from a DJ friend's complaint: online record stores all look like generic e-commerce sites.

Anish wanted to recreate the experience of browsing record stores in Toronto in the 1990s, complete with loud music, mixtapes, rave flyers, and chance encounters.

Anish fed the model visual references from Habbo Hotel and began rebuilding that experience. Because the production assistant directly serves an end-to-end new game, the record store project sits in the “Invent” quadrant on our chart.

Eventually, Anish made his record store and surrounding streets moddable, allowing other players to build their own storefronts down the street. His son added a pet adoption center inside (noting this might be an homage to the popular Roblox game Adopt Me).

Animals in the record store could be adopted and follow their owners around—a behavior Anish hadn’t included in the original project. He says a coding agent automatically implemented it and made the existing record store codebase backward-compatible with these new mechanics. That’s “Extend”: players bring fresh ideas into a world they already know.

Jon’s eight-year-old son provided a smaller, more contemporary example. He wanted to play Ghost of Tsushima, but Jon found it too violent. So he started asking ChatGPT about its mechanics and boss fight designs, then used Codex to help craft a Minecraft mod inspired by them.

Many of these emergent model capabilities, such as operating Blender or using 3D libraries like Three.js and Spark.js, only truly crystallized over the past few months.

When OpenAI released Astra, people were immediately struck by its complex computer-use abilities, executing long-horizon tasks across graphic software and game engines like Blender, Unity, and Unreal.

Meanwhile, Anthropic’s Opus models have shown particular strength in end-to-end game design and code implementation. Just a few months ago, you might only ask a model to generate usable assets or game spec documents.

Today, you can basically just type “make this game more fun,” and it actually becomes more fun. Simon Willison wrote an excellent post detailing some of these dynamics.

Tool Makers and the Decompilation Wave

The continuous improvement in model capabilities provides a great entry point for discussing the “prophetic tool maker” archetype and its role in the AI era. AI is getting increasingly strong at assembling games, and equally impressive is its prowess at taking them apart.

This process, known as decompilation, involves reverse-engineering compiled programs to reconstruct human-readable and modifiable source code. Achieving this requires creatively applied engineering practices.

In Maurice Heumann’s account of an AI decompilation project, he directed a swarm of agents to decompile a first-person shooter. Initially, the code the agents generated was readable but functionally incorrect.

This forced his team to implement a verification loop: recompiling refactored functions and diffing them against the original binary, while handling references prone to shifting memory addresses. It perfectly illustrates the new breed of tech pioneers: acting as supervisors who design constraints that allow agents to solve problems reliably. This is a textbook example of the “Recover” quadrant.

On the podcast, Jon noted that, by his count, there are already at least four browser-run decompiled versions of Halo. We are well on our way toward a new gaming benchmark shifting from “Can it run Doom?” to “Can it decompile Halo?”

Corporate Attitudes and New Studio Opportunities

For AAA studios, it would be easy to outright dismiss AI, as many players have done recently; it’s equally easy to panic amid the rising decompilation wave and reach for lawyers. Seeing major studios show genuine openness and excitement makes for a particularly intriguing contrast.

This October, Tim Sweeney outlined an organizational restructuring around AI-assisted development designed to help employees produce more and higher-quality work, extending those benefits to teams using Unreal Engine. Meanwhile, Xbox announced the launch of Xbox Labs, focusing on porting games for new hardware, developing browser ports, and remaking/remixing titles.

Xbox formed a dedicated task force to explore these possibilities and share tools. The significance lies in legacy majors recognizing both the opportunity to expand new game production and the value of revitalizing their existing game libraries.

For world-building studios, AI brings a dual opportunity: the ability to drastically compress development cycles, and the chance to reassess what’s truly worth experimenting with. As Jon pointed out on the podcast, investing years and hundreds of millions into a single game pressures producers to lean on familiar IPs.

AI opens new possibilities by shortening the time from concept to prototype. By compressing development cycles, studios can prototype more game ideas faster, taking greater risks on original IPs or experimental genres that would barely get greenlit under traditional timelines.

The Act of Creation Is Its Own Game

Now, all three of us are actively making games ourselves. At the tail end of the podcast, we discussed where we hope to take this creative freedom. Anish has been working tirelessly to resurrect Solar Realms Elite, a multiplayer game he played back in his early BBS days, whose source code was long lost.

Fortunately, a design doc and executable remained. He kept chipping away, kicking off the project anew with each generation of models, and now runs a highly faithful recreation on his own machine.

Jon wants to revisit Civilization II, giving the citizens their own agendas so ruling an empire becomes a process of constantly negotiating cooperation. Elena is considering resurrecting some retro dating sims she made in Unity before the AI era.

The most obvious criticism of this current wave of AI-generated games is: they might amuse the creator and a small circle of friends, but we’re still a long way off from truly grand, coherent AI-generated games.

Critics might argue that these early AI games are fun primarily because of the novelty of “making a game by hand,” rather than their actual entertainment value. But is that really a bad thing? From building Lego sets and drawing with crayons to constructing worlds in Minecraft, the act of creation has long been its own game.

This means we’re witnessing one of the things that has always fascinated us about gaming: its capacity to spawn a grassroots ecosystem built on tutorials, tacit knowledge, and creative exchange. Gaming history is filled with stories of people stumbling upon brilliance while tinkering with others’ works, with Counter-Strike standing as the most famous modding example—it began as little more than a mod for 1999’s Half-Life.

We can honor this history without expecting every Minecraft mod or vibe-coded game to become the next Counter-Strike.

Ultimately, vastly more games will emerge. Someone who gets into development by refactoring old games may develop a deep interest in the underlying systems, since it’s hard to truly master decompilation without understanding game binaries. Someone who builds a game world for friends might keep expanding it for years, eventually attracting a massive player base.

These are the classic tropes and origin stories belonging to the industry archetypes we already know and admire. As an industry obsessed with “persistent people making incredible things,” it should feel nothing but thrilled to see more people getting their foot in the door.