by 
The subject of AI use in the world of retro gaming has already caused plenty of discussion, and another project has emerged over the weekend to add more fuel to the fire.
As spotted by @ray_castello on social media, retroports has just released footage of GoldenEye 007 (apparently) running on SNES.
The video shows the opening level of the game and, according to the developer, utilises a "polygon-based approach with some fog/occlusion culling techniques." The texture quality used here means only the first stage is possible, and retroports points out that the footage is not precomputed but "live on the SNES," before clarifying it's actually running via the SNES emulator MESEN, combined with the 2MB SuperFX2.
Then comes the almost inevitable admission that retroports has "mainly used [Claude] Opus and then [Chat-GPT] Astra helped bail out a couple of bugs."
The developer adds that they have a massive amount of respect for people who created 3D games on the same hardware without AI assistance. "Hats off to the people who can render a triangle on these things, let alone do creative optimisation. You guys are my heroes. Beyond obsessively watching optimisation videos, the credit lies with Dylan Cuthbert, Rare, Nintendo and our agentic overlords. It’s a really exciting time to try things. I just hope someone with OG hardware skills can do even better. MVG, Kaze Emanuar, James Lambert, Game of Tobi and Joshua Barretto are the actual geniuses!"
As you can imagine, the reaction online has been divided, as is so often the case with vibe-coded projects. There are a lot of positive comments, for sure.
"This might be the most impressive port ever," says @AFat32USBOrigials. "I can't even fathom how much optimisation it took to make this even run!"
"This is what AI does best, folks," adds Brett Viker. "It opens the door for dreamers to make their projects a reality by tearing down the barriers to entry. We should be all for this."
However, others have been slightly less positive, even going as far as to cast doubt on the legitimacy of the video until a ROM is released for people to test out themselves. Even @ray_castello, our source for this story, admits that the world of AI-driven coding makes it hard to navigate. "There's a lot of crap being churned out by AI these days, but I found it interesting."
@Optimus6128 says what a great many people feel when such ventures appear. "There are so many of these 'look I ported this impossible thing to this impossible hardware' releases these days, and most of them are AI-driven. That makes them non-impressive for me; it trivialises the hobby. Like, there is something crazy like this, and many years ago I would be like, "Wow, what did this coder do this time?" and now I am like 'Meh, I'll pass.'"
While AI is helping people do impressive things in the realm of coding, there are still concerns over how untrained coders can possibly debug and tackle issues with AI code they had no hand in creating. We've also seen many vibe-coded ventures hit the market in a poor state, and many argue that the existence of such projects dissuades human developers from creating what could be superior takes on the same project.
Of course, there's also the very real issue of AI being based on the hard work of humans, and its use comes with a stern environmental cost – and the fact that the companies behind these AI tools aren't entirely truthful about a lot of things.