The Next Era of Games: Mods, Agents, and Infinite Worlds

Microsoft

Across the shows: 44 speakers in other episodes, 53 bullish and 8 bearish calls, past 30 days

• Jonathan Lai · Andreessen Horowitz (a16z)Bullish

  Xbox likely sees AI-driven community ports, such as Halo in the browser, and is embracing them.

  Jonathan links Xbox's announcement about supporting AI and new development frontiers to community projects. Several groups decompiled and recompiled the original Halo, a 2001 Xbox game, into WebGPU versions that play in a browser. Jonathan thinks Xbox has seen this and will use AI to port and remaster older titles for new audiences, as it once did with teams porting Windows games to other platforms.

  “I think Xbox has likely seen things like this happen, and they're saying, hey. Like, we used to have whole teams that would just take a Windows game”at 21:08

• Anish Acharya · Andreessen Horowitz (a16z)Bullish

  Extension-style mods strengthen Minecraft and benefit its owner, Microsoft, unlike forks.

  Anish says Minecraft is the strongest example of a game benefiting from its mod ecosystem. Mods that extend a shared core game and community also strengthen the IP owner. Forking a game gives the owner no benefit and may split the community. Microsoft owns Minecraft, so the ecosystem gain accrues to Microsoft, though Anish also notes Microsoft wishes it had a more network-like asset.

  “the strongest form in my mind is Minecraft. ... Minecraft really benefits from the mod's ecosystem.”at 29:11

Anthropic

Across the shows: 116 speakers in other episodes, 160 bullish and 93 bearish calls, past 30 days

• Anish Acharya · Andreessen Horowitz (a16z)Bullish

  Anthropic's Opus models unexpectedly excel at Three.js browser 3D, enabling AI-driven game generation.

  Anish says a recent Opus model had an unplanned ability to work well with Three.js, a library for browser 3D. Anthropic said it did not design that in. The model lays out scenes element by element, which would be tedious for a human. Anish treats coding as one of two converging mega trends in game generation, alongside tool use such as operating Blender.

  “I think it was Opus four seven, maybe it was Opus four eight, but it just had this property, actually, which it was not designed to have in it”at 5:15

Spawn

• Anish Acharya · Andreessen Horowitz (a16z)Bullish

  Spawn is the most ambitious shared community and AI game generation platform Anish has seen.

  Spawn combines a shared community with game generation. Its embodied agent, Savi, appears in every game, takes bug reports, modifies the game while the player is in it like a copilot, and creates portals to other games. Every game is multiplayer by default within the same community. Anish calls it early but the most impressive and ambitious effort he has seen at this. He says the a16z partners are not investors in Spawn.

  “Well, you should check out Spawn, actually. Now Spawn, we're not investors, but I'm a huge fan of what they're doing.”at 30:16

Media, music & entertainment

• Jonathan Lai · Andreessen Horowitz (a16z)

  AI shortening game dev cycles lets studios take more creative risk and ship new IP.

  Cutting the idea-to-prototype time cuts the cost of making a game, so studios can take more bets. Jonathan cites Grand Theft Auto VI, in production about 12-13 years, as an example of long cycles. He says the cycles suffocate innovation because a producer at a AAA studio spending hundreds of millions over a decade will greenlight sequels and established genres over new IP. Cheaper development, roughly tens of millions over 2-5 years instead of hundreds of millions over 5-10, would let studios fund new IP and first-time developers.

  “get a five to ten year dev cycle to, I don't know, two to five years. It's still gonna take some time, but I think it'll just enable the game developers to take more risk”at 4:46

• Jonathan Lai · Andreessen Horowitz (a16z)

  AI makes modding games easy, letting fans remix existing IP into new games and genres.

  Mods have produced major franchises: Counter-Strike came from a Half-Life mod, League of Legends from a Warcraft III mod (DOTA), and Fortnite and PUBG from an Arma mod. AI now lets an ordinary player point an AI agent at an existing game and ask for changes, such as adding voxel crafting to Ghost of Tsushima. If the mod is fun, other players adopt it and it can become a genre. Starting from an existing game also gives players a framework to remix, rather than a blank canvas.

  “And AI is making it really, really easy to mod games right now, because exactly like the way my son did.”at 19:15

• Anish Acharya · Andreessen Horowitz (a16z)

  AI will spark a renaissance of RTS and other genres that never worked as iPad games.

  Anish notes that many have tried to build a great iPad RTS, but no one has really succeeded, and that high production costs and weak returns have kept such genres scarce. He expects AI to make these genres economical and predicts a renaissance of RTS and similar genres. He says Tim Sweeney has made a similar point.

  “we've never really, really gotten one, and, you know, now I think we're gonna see this complete renaissance of genres like RTS.”at 22:48

• Jonathan Lai · Andreessen Horowitz (a16z)

  AI makes niche game genres, such as job simulators, economically viable to build.

  Niche genres such as job simulators have been unprofitable because of high production costs. Jonathan describes a train-conductor simulator he plays as an example of a genre with a small audience that still attracts paying players. With AI, a studio can build these games cheaply, and players who love them will pay. Simulators are an evergreen genre, so AI could let players role-play almost any job they can imagine.

  “But now you can role play a day in the life of almost any job in the world that you can envision.”at 24:46

• Jonathan Lai · Andreessen Horowitz (a16z)

  Original IP owners have historically not captured value from mods; the business model is unresolved.

  Original creators have historically not benefited much from mods built on their IP. The original League of Legends creators did not gain much from the DOTA-derived game, and it is unclear whether Half-Life's developers earned much from Counter-Strike. As AI makes remixing and recompiling easy, companies face an open question about whether they can monetise player creativity. Jonathan says folks are still early, with ideas such as revenue shares or royalties going back to the company.

  “historically, original owners of IP have not benefited that much from the mods that have been created on top of their IP.”at 28:06

• Jonathan Lai · Andreessen Horowitz (a16z)

  AI-made games need a new distribution platform, since Steam, App Store and Roblox are not built for them.

  Many AI-built games are still released through traditional channels such as Steam, the App Store and Roblox. Jonathan says this leaves an opening for a company to build a distribution platform that natively serves AI game creators. He points to a broader opportunity in persistent communities that connect players across games, as Minecraft and Roblox do.

  “today, like, many of these games are still going through traditional distribution channels”at 29:52

• Jonathan Lai · Andreessen Horowitz (a16z)

  AI is now table stakes for game studios, used mainly to boost productivity.

  Jonathan's conversations with game developers show that almost all of them already use AI. He calls the first wave of adoption productivity: generating design documents and concept art and iterating quickly from idea to prototype. He sees a second wave as more exciting, where AI creates new mechanics and genres that cannot exist today and expand the market for games.

  “in my conversations with game developers, I would say in general, they're all using AI already. I think it's AI has become table stakes”at 35:01

• Jonathan Lai · Andreessen Horowitz (a16z)

  AI world models could let MMOs adapt in real time, generating personalized quest lines that current scripted games cannot.

  Jonathan argues that today's MMOs are scripted linearly, so they cannot handle a player who ignores the intended path, such as one who stays on the beach and builds a ship to sail the ocean. He proposes an adaptive MMO built on world models that could generate an entire quest line for such a player, including pirate attacks and naval combat. He frames this as a possible future rather than a near-term product, and he links it to the 'never ending game' idea from his earlier article.

  “in an adaptive MMO, one that, maybe is built using world models, for example, could literally generate an entire quest line for someone who really likes the ocean.”at 36:41

• Jonathan Lai · Andreessen Horowitz (a16z)

  Small audiences deter studios from making RTS games; cheaper AI development could spark an RTS revival.

  Jonathan says few developers are making new RTS games because a pure market-sizing analysis shows the audience cannot support them. He expects that to change and predicts an explosion of RTS games. His reasoning implies that cheaper development, which the episode attributes to AI, would reduce the cost of serving a small niche. He gives no numbers and no timeframe, so the view is loosely argued.

  “There's not a lot of developers that are trying to make new and modern RTS games... because the audience size doesn't support it if you just look at it from a pure market sizing analysis.”at 38:57

Frontier AI labs & model race

• Jonathan Lai · Andreessen Horowitz (a16z)

  Recent model can operate Blender, Unreal and Unity directly, enabling one-shot game prototypes.

  Earlier models only told users which steps to do in Blender or Unreal. The newer model can operate those tools itself, so a hobbyist can draft a design document, generate concept art, build and rig 3D models, add music and export a playable game. The model also QA-tests its own output by reviewing screenshots and fixing bugs, which is why early prototypes on X are surprisingly good. Jonathan calls this a huge technology unlock.

  “it was very, very good at directly using very popular game development software”at 2:20

AI & jobs

• Jonathan Lai · Andreessen Horowitz (a16z)

  AI will not replace human taste and creative judgment, which remain critical for making games fun.

  The first 80% of a game, a working prototype, is easy for AI to produce. The last 20% makes it fun, including how the world feels, what a shotgun sounds like versus a pistol, and how a laser gun recoils. Jonathan argues this polish will stay the province of humans for the foreseeable future because it depends on how players feel and what they want.

  “I think the AI model is not going to replace human taste and creative judgment. I think that is still really, really critical.”at 10:30

Enterprise AI adoption & AI software

• Jonathan Lai · Andreessen Horowitz (a16z)

  Command-line agent management is non-intuitive, so a mass-market visual interface for agents will likely emerge.

  Jonathan says the average consumer finds agent orchestration non-intuitive when it runs through a command line or text interface. He thinks a more mass-market visual interface will probably emerge, possibly drawing on game-inspired GUIs like StarCraft or Age of Empires. He acknowledges he is less sure of the exact implementation, and says the field is early in designing interfaces for agentic workplaces.

  “average consumer finds agent orchestration to be non intuitive”at 27:07