Across four game-development attempts, one gameplay has launched, one has entered real rooms, and two remain in testing or demonstration. We see AI compressing the time required to build game prototypes, but not yet the last mile of delivering a finished product to players.

Huajiao Tech Team’s practice is closer to frontline operations than the spectacle of “generating a game in real time”: the team uses AI to help with gameplay design, code development, 3D modeling and model Q&A, then has engineers fill in physics rules, networking, exception handling and deployment. Rather than saying AI has taken over game development, it has made “building a playable version first” cheaper and faster.

What This Is

The team made four types of attempts. “Guess Together” uses AI to answer “yes,” “no,” or “uncertain” based on the answer key; it has launched in party rooms and solo-stream rooms, with hundreds of rooms starting each day. “Pool” uses Codex—an AI tool that can read and write code and assist with development tasks—to handle shots, collisions, pocketing and rule decisions. It is still refining full matches, reconnection after network loss, and launch preparation. “Ghost Attack and Defense” tests complex mechanics involving resources, construction and offense; its online-integrated version is already being used to support business operations, while its in-house 3D prototype continues to validate the development path. “3D Farm” remains a demo focused on modeling and interaction.

Industry View

The noteworthy shift is that AI tools can now handle code, interfaces, model integration and basic assets simultaneously, allowing nonprofessional teams to validate game ideas relatively quickly. Once the guessing gameplay entered real rooms, the team could continue optimizing it using feedback such as repeated questions and answer smoothness. This shows that AI is no longer used only before development; it is also becoming part of runtime experience.

The risks are equally clear. Code that runs does not mean a game is fun. Pool still requires human playtesting for game feel, numerical balance and control guidance. Full matches, reconnection after network loss and synchronization across multiple rooms are not yet complete. Real-time generation also faces latency, failures and difficulty for players in judging the effects of their actions. If model responses are unreliable or the system cannot continue when the model behaves abnormally, an entire match can be disrupted. Combined with inference costs, companies will not assess the returns solely by “how much code was generated.”

Impact on Regular People

For enterprise IT: AI is splitting development work into two phases: “prototype quickly” and “deliver seriously.” The former can deliver clear efficiency gains; the latter still requires engineers to address stability, testing, business rules and deployment costs.

For individual careers: Product, operations and business staff will be better able to turn ideas into demonstrable versions, lowering the barrier to collaborating with developers. But defining requirements, validating experiences and deciding whether to launch remain core human responsibilities.

For the consumer market: Players will see more AI-assisted mechanics that allow real-time variation, potentially producing richer interactive content. At the same time, generation latency, unstable logic and uneven content quality could affect actual retention.