Menu PocketGamer.biz
Search
Home   >   News

Do AI tools for game development live up to the hype? Not even remotely

But the tech did enable an experienced engineer to re-create the original X-COM in three months
Do AI tools for game development live up to the hype? Not even remotely
Stay Informed
Get Industry News In Your Inbox…
Sign Up Today

AI tools for game development do not “even remotely” live up to the hype, says AI Guys CEO and CTO Ken Noland.

Speaking at New Game Plus during London Games Festival, Noland discussed his journey using AI tech to build the original X-COM: UFO Defense in North America completely from scratch. To create it, he used the code behind OpenXcom, an open-source clone of the title, as a guide.

The challenge: can a single engineer build a whole new game engine with the assistance of every generative AI tool at their disposal in just three months? The new system also had to be a perfect reaction of the original game down to the last pixel, while also allowing for modders to define their own gameplay experience.

The reason for the project was to find out: what is the difference between the AI hype and the reality of game development?

Practical AI

Throughout the development of the new engine, he used a number of different development environments, including Microsoft Visual Studio, Microsoft VS Code, Anysphere Cursor, Jetbrains Clion and Jetbrains Rider. He noted that Cursor in particular became a necessary tool for development.

Noland utilised a number of AI tools including ChatGPT, Claude 3 Sonnet, GitHub Copilot, Cursor, CodeLlama and DeepSeek Coder.

“Deep seek surprised me," he said. "It was, as far as local models go, on par with the frontier models. And just like Claude it was able to ingest large chunks of code and be able to produce meaningful results. It didn't always compile, but none of the models always produce compilable code.”

After three months, he released the first version of the game.

“There were, of course, some minor bugs that popped up, and there’s ongoing need for support to enable more modding experiences, but overall we delivered,” he said

Though Noland was able to build a new game engine from scratch in three months using AI tech, he said it doesn’t even remotely live up to the hype.

“You can't create a game using generative AI, and at the current level, to create any kind of playable experience, it still requires an experienced engineer,” he stated.

Noland added that without the skills of being an experienced engineer to drive the AI tools, he would never have been able to build a game engine from scratch.

However, noting a more “inspirational” conclusion, he said it was still a “huge accomplishment” to build an engine in just three months – minus an audio system.

For a more detailed look at the development process, you can read the X-COM case study here.

Want to learn more about AI? We regularly run the Practical AI track at our Pocket Gamer Connects conferences. Check out our upcoming shows here.

Our next even is the Dubai GameExpo Summit powered by Pocket Gamer Connects on May 7th to 8th.