māchine · gaming

AI coding tools speed up retro game ports and strain the decomp community

Hobbyists are using AI to port classic games to browsers, phones and headsets in weeks rather than years. Decomp.dev, a hub for that work, has stopped taking new projects.

Classic games are turning up on hardware they were never built for. Halo runs in a web browser, Wind Waker on an iPhone, Super Smash Bros. in mixed reality. According to The Verge, projects like these have gone viral unusually often over the past month or so. The reason is that hobbyists are now using AI coding tools to take games apart and rebuild them.

The work is called decompilation and recompilation, usually shortened to decomp and recomp. Done by hand, it can take years. With AI assistance, one person can now run many such projects at once. The faster pace has a cost. One of the main community hubs for this work has stopped listing new projects, and its operator says AI has broken how the community works.

What a decomp is and why it was slow

A decomp breaks a game down to its building blocks. A recomp rebuilds it so that it runs on hardware it was never meant for. Done manually, this kind of reverse engineering is slow and painstaking. The Verge notes that one group spent years decompiling the original Animal Crossing for the GameCube.

AI coding tools cut that timeline down sharply. Chris Sotraidis, an AI consultant who ports games as a hobby, told The Verge he got started this summer. He had seen an iPhone and iPad port of Command & Conquer: Generals Zero Hour, which EA made open source last year. He then built a mobile port of The Legend of Zelda: Ocarina of Time on top of an existing decomp made by the community. Within a few weeks he was working on a dozen projects.

"I'm honest enough to say that the pace and volume would not have been possible for me without it," Sotraidis said. He added that some of the ports would have been possible with minimal AI help if he had had much more time. "The tools allow you to move much faster if you know what you want."

A community index closes

Decomp.dev launched in 2024 to track decompilation projects. It has now switched off its main index and stopped accepting new projects. Existing project pages stay online. Its operator writes under the pseudonym Encounter. In the shutdown notice, quoted by The Verge, Encounter wrote that "with the right guidance, human input, and code reviews, AI can expedite the decompilation process in a significant way." They also wrote that "the community aspect of decompilation has been broken."

The site does not blame AI as such. It says AI "can help tremendously when used correctly" and does not ban projects that use it. The operator writes that a single hobbyist cannot define and enforce quality standards across hundreds of projects. The real target is what Encounter calls hostile forks. These are copies of long-running open-source decomps that someone finishes with AI and releases as a port, without ever having contributed to the original. Encounter wrote that these forks "put little to no effort into meaningfully contributing to existing projects, tearing apart our community fabric." The site says its index and leaderboard were "causing more harm than good."

Preservationists point to a second loss. Part of the purpose of a decomp is to understand how a game works and why certain programming decisions were made. The Verge notes that this understanding is lost when AI does the work, which can make it harder to restore games faithfully. The outlet cites an estimate that more than 90 percent of classic games are "critically endangered." It describes a split between preservationists, who are frustrated by low-quality work, and AI users, who are excited about what the tools make possible. Sotraidis expects quality to improve as the tools mature. He adds that "better tools aren't automatically going to make people give proper credit or treat each other well."

Industry interest

Large companies are moving into the same space. On October 13, Xbox CEO Asha Sharma announced Xbox Labs, an internal team meant to "turn emerging ideas into things our players and developers can experience." She wrote that it is easy to see the possibilities in "making great games easier to access, giving old games new life, and allowing more players to become creators and co-creators." According to The Verge, the team has so far teased only an experimental dashboard.

Epic Games has launched a vibe-coded version of its PC games store, meaning one built largely through prompts to AI coding tools. It has also reportedly moved to "AI-accelerated development" for an upcoming collaboration with Disney. Epic CEO Tim Sweeney wrote on X that AI tools will lead to "a new golden age of gaming."

Old games are also commercially valuable. Remakes of Halo: Combat Evolved and Ocarina of Time are among this year's biggest releases. Hobbyists are porting the same titles with AI in a matter of weeks, so official remakes and fast unofficial ports now exist side by side.

What it means for work

AI tools now take over much of the labour-intensive part of reverse engineering and porting. That includes reconstructing readable code from old games, finishing partly completed decomps and adapting the code to new platforms. A job that once took a group years is now something one person can attempt next to a dozen other projects. The tools do not produce understanding of how a game was built, credit for earlier work or the shared code review that open-source projects depend on. The people most affected are the volunteers who did this work by hand, because their unfinished public projects have become raw material for others to complete. For rights holders, porting is no longer a scarce skill. That changes both what official ports and remasters cost and how quickly unofficial versions appear. It is still open how communities will credit and review AI-assisted work, and whether efforts like Xbox Labs will go beyond experiments.

Sources

This article was researched, written and illustrated by AI, and published without human editing. Spotted an error? Let us know.