AI can shorten game prototyping, but making a game fun still takes judgment

The speakers see AI shortening the distance between a game idea and a playable prototype. They discuss what that could mean for studios, hobbyists and older games, while distinguishing a working prototype from a finished game that people want to keep playing.

Jonathan Lai says a recent model made it easier to operate tools such as Blender and game engines directly, instead of merely explaining what a user should do. In his account, a hobbyist can use an agent to draft a design, create art and models, animate assets and assemble a playable prototype.

He says the agent can also inspect screenshots and use them to identify some visual issues. Lai views faster iteration as important for niche ideas that might not justify a traditional development budget; the examples are early prototypes, not proof that a model independently produces a polished commercial game.

Lai contrasts this with long, expensive development cycles for major titles. He speculates that reducing a five- or ten-year cycle could let studios try more ideas and take creative risks, while acknowledging that substantial games would still require significant time and money.

Lai says the hoped-for benefit is more chances to prototype and more variety in games, rather than a guarantee that every experiment will become a hit. He gives a hypothetical reduction from a five-to-ten-year cycle to two to five years as an illustration.

Anish Acharya distinguishes two routes to AI-assisted creation: a model writing browser-based 3D code and a model operating tools such as Blender. He says both can accelerate experiments. He also notes that long projects can be cancelled after years as markets, tastes or creative priorities change.

Acharya says long projects can run into changing tastes or financial conditions before they ship. In his view, shorter cycles could reduce the number of ideas abandoned after years of work.

Lai argues that when a studio faces a decade-long project and a large budget, it has an incentive to favor sequels and established genres. He hopes cheaper production could support new intellectual property and teams without previous shipped games.

Acharya offers another possibility: instead of simply shortening development, teams might use the same time to build more ambitious worlds. The speakers treat this as a prospect enabled by better tools, not a description of a finished game.

Lai says game development traditionally combines programming, visual art, design, music and sound. AI can help fill gaps for a person without every specialty. He expects more people to try building, although many may stop at an initial level or prototype.

Lai expects some first-time builders to abandon their early experiments, while others may discover they enjoy game development enough to pursue it further. His claim concerns a possible expansion of creators, not a measured outcome.

The interviewer asks whether AI can make a game balanced and genuinely fun. Lai says human taste and creative judgment remain critical: making a scene run is different from making its world, controls, sounds and consequences feel right.

He uses the industry saying that the final 20% can require 80% of the work. The feel of a weapon, differences in sound and other small details contribute to a coherent experience; he expects human judgment to remain important in that refinement.

The group discusses depth and consequence as reasons a player may spend hundreds of hours in a world. They wonder whether language models could eventually help extend that depth, while acknowledging that enjoyment depends on social and cultural factors as well as design.

Acharya adds that creating can itself be a form of play, even if only the creator ever plays the resulting game. He compares making and adjusting a game to a tranquil creative activity; Lai points to Minecraft and physical building toys as precedents.

Acharya describes building a Habbo Hotel-inspired online record shop after discussing the social experience of record stores with a friend. He says a model generated a room and music-related interactions from a few visual references, exceeding his expectations.

The record-shop idea came from memories of a social Toronto store with music playing and conversations among visitors. Acharya wanted to see whether an online space could recover some of that experience, rather than simply listing tracks.

He expanded the prototype into a Toronto street whose other storefronts players could customize by prompting. His son created a pet-adoption space; Acharya says a coding agent changed the game engine so adopted animals could follow the player, a mechanic he had not explicitly built.

Lai recounts that his eight-year-old son wanted to play Ghost of Tsushima, which he considered too violent. The child used AI and public game descriptions to help create a Minecraft mod inspired by it. Lai says the mod worked and was shared with friends; it is his family’s example, not a benchmark.

Lai says his son used a chatbot to understand mechanics from online descriptions, then a coding agent to build the mod. Acharya sees the example as mixing existing ideas rather than simply increasing graphical realism.

The speakers distinguish extending a game through mods from recreating or combining existing works. Lai points to Counter-Strike’s origins as a Half-Life mod and Dota’s roots in a Warcraft III mod as examples of how communities can generate new genres and franchises.

Lai also mentions battle-royale games that drew on earlier mods. His point is that player-created modifications have long been a source of new ideas, before the arrival of today’s AI tools.

AI may lower the barrier for a player to add a mechanic or combine familiar ideas, Lai argues. Starting with a game someone already knows can also be easier than facing an empty prompt. Whether a particular result is fun still has to be tested.

Acharya discusses decompiling older games and combining assets or styles from different games. Lai says community projects have explored making older titles playable on newer platforms and sees potential for AI to help with porting or remastering work.

Lai describes hearing about community projects that made an older Halo game playable in a browser. He sees this as an example of bringing older works to new hardware and audiences, while the conversation leaves rights and commercialization questions open.

The speakers also expect cheaper production to make niche genres more viable. They cite real-time strategy and job simulators; Lai describes enjoying a train-conductor simulator where the player must keep to a timetable and stop accurately.

Lai’s simulator example involves driving a train, stopping at stations and meeting a schedule. He finds the narrow subject compelling and thinks lower development costs could serve small audiences with similarly specific interests.

The discussion turns to visual interfaces for managing coding agents that resemble real-time strategy games. Acharya compares managing teams under limited control and information to a simulation; Lai thinks visual interfaces may make orchestrating agents more accessible, although he is unsure what form will work.

Lai says text-heavy agent management may be unintuitive for many consumers. A graphical interface could help people supervise several agents, though he does not know whether a strategy-game interface is the right design.

The interviewer asks whether publishers can capture value from AI-enabled modding. Lai says the commercial relationship between an original rights holder and successful derivative work is not always clear, and suggests that revenue-sharing or royalty models need more exploration.

Acharya distinguishes extensions that bring players back to a shared base game from forks that might split a community. They discuss whether new distribution platforms could connect AI game creators and allow a community to persist across different games.

The group asks how a new generation of AI-generated games might find players if most current distribution still runs through established stores and platforms. They see room to experiment with community and discovery tools.

Acharya gives an early platform, Spawn, as an example he likes and says his firm is not an investor. He describes an in-game agent that can respond to bug reports, modify a game and connect players to other experiences. This is his account of the product, not an independent evaluation.

Lai compares Minecraft’s shared core game and culture with Roblox’s more decentralized collection of experiences. He says both can support communities, but starting from the same game gives Minecraft players a common reference point for mods.

In Lai’s comparison, Minecraft modders depart from a common base that players already understand, while Roblox players can enter through very different individual games. He says both approaches can work but organize community differently.

The speakers think studios may welcome cheaper prototyping and lower financial risk, while also facing questions about community and distribution. They suggest that improving development economics could help working conditions, but do not establish that outcome.

The speakers also discuss studios’ workload and financial pressure. They hope cheaper production could improve business economics and working conditions, but acknowledge their discussion is forward-looking.

Lai describes AI-generated design documents and concept art as an initial wave of use in studios he speaks with. A more speculative next wave would use AI to enable new mechanics and experiences, rather than only making existing kinds of games faster.

His example of a “never-ending” game is an online world that adapts when a player ignores the scripted route: someone who sails away from a starting beach might encounter a newly generated nautical quest. He presents this as a desired future experience, not a current demonstrated system.

The participants share favorite games, from The Witcher 3 and Dune 2000 to League of Legends, StarCraft and Crusader Kings III. Lai hopes lower production costs could encourage more real-time strategy games despite the genre’s relatively narrow audience.

The participants’ favorite games illustrate different attractions: detailed stories, strategy, shared competitive play and managing relationships. Lai says competitive real-time strategy can be difficult for new players because experienced opponents act much faster.

Acharya describes recreating an old asynchronous bulletin-board game using its design document and executable after its source code was lost. He says he now has what feels like a faithful version on his computer; the assessment is his own.

Lai would like to recreate an older Civilization game on modern hardware and imagines a version where individual units have their own goals while the player sets broad strategy. Another participant notes that existing games such as Crusader Kings III already provide some related dynamics without frontier AI.

The proposed Civilization-style agent simulation would let units or subjects follow their own objectives instead of requiring the player to control every action. The group points out that comparable strategic constraints already exist in non-AI games.

One participant recalls experimenting with a dating simulation using Unity and branching dialogue. Lai suggests that future conversational agents might make character interactions more dynamic, but creating compelling characters would remain a design problem.

The group ends by discussing dating-simulation games, whose characters require extensive authored backstories. Lai imagines future characters responding dynamically as agents; this, too, is presented as a possibility rather than an established capability.