You remember the magic of the 1980s and 90s computing scene. The hum of the CRT, the click of the mechanical keyboard, the raw, unfiltered thrill of making a machine do exactly what you wanted. For many of us, the Amiga wasn’t just a computer; it was a cultural moment. So, when a project announces it’s bringing Amiga Unix back to life, your first reaction is probably a rush of pure, unadulterated nostalgia.
But then you read the documentation. You read the project’s page. And a cold feeling creeps in. The syntax is too clean. The phrasing is too sterile. It’s the textual equivalent of the uncanny valley. You realize, with a sinking heart, that this resurrection wasn’t built by human hands. It was generated by an LLM.
There is a difference between resurrection and reanimation. One honors the dead; the other just makes a corpse move.
One commenter on the project summed it up perfectly: they called it “pet-cemetery vibes.” The AI brought it back to life, but it came back wrong. This isn’t just a technical critique; it’s a fundamental crisis of trust. When you’re dealing with vintage computing, you aren’t just looking for a binary that boots. You are looking for the ghost of the original creators. You want to see the human ingenuity that fought against brutal hardware limitations.
Take the user who ran NetBSD 0.8 on an Amiga 500 with a 68030 w/MMU, a 68881 FPU, and a SCSI hard drive sidecar. That wasn’t just software running; that was a human being wrestling with hardware, making magic happen in the same timeframe Commodore released their own Unix. That history matters. When an LLM steps in to bridge the gaps in our knowledge, it doesn’t recreate that struggle. It hallucinates a frictionless, soulless approximation of it.
When an AI writes the code, it isn’t preserving the soul of the machine. It’s just putting a skin suit on a modern algorithm.
The defenders will say, “It works, doesn’t it?” They’ll point out that Debian 3.1 supported Amiga hardware until 2007, and that modern tools can easily patch the gaps. But they miss the point entirely. The biggest threat to vintage computing isn’t trademark disputes or technical difficulty. It is the erasure of the human fingerprint.
We are so obsessed with the idea that AI can do everything that we are forgetting what it means to preserve something. Preservation is an act of care. It requires a human to look at the past and say, “This mattered because someone built it.” An LLM doesn’t care. It just predicts the next most likely token. It doesn’t feel the weight of history. It doesn’t respect the architecture.
If you are working on an AI-assisted project to resurrect a piece of computing history, my advice is simple: do it a bit less AI-assisted. Use the LLM to format, to suggest, to brainstorm. But do not let it write the narrative. Do not let it generate the core.
If we let machines write our history, we won’t get the past back. We’ll just get a very convincing ghost.
FAQ
Q: What's wrong with using AI if it gets the OS running again?
A: Getting it to run is only half the battle. In vintage computing, the human context—the struggle, the workarounds, the specific decisions made by original engineers—is the actual value. AI strips away that context, leaving you with a functioning but soulless imitation.
Q: How much AI is too much in software preservation?
A: When the reader or user can immediately tell that the documentation, comments, or architecture were machine-generated, the trust is broken. AI should be a research assistant, not the lead author. If it writes the narrative, the project loses its historical authenticity.
Q: Is AI actually worse at preserving old code than humans?
A: Yes, for historical purposes. An AI might write more 'efficient' code by today's standards, but it will completely miss the hardware-specific quirks and cultural constraints of the era. It optimizes for the present, erasing the reality of the past.