What happens when you ask a 2025 AI to think like a 1998 computer?
A Claude user found out by prompting the model to rebuild their childhood PC from scratch - specs, interface quirks, era-appropriate software, and all. The result was a working simulation of a late-90s Windows machine, complete with the kind of period-accurate details - boot times, program names, desktop layouts - that only someone who lived through that era would recognize as correct.
The project isn't a serious engineering achievement, but it's a good demonstration of something Claude does unusually well: maintaining a consistent, detailed persona or environment across a long conversation. The model didn't just list Windows 98 features. It stayed "in character" as the machine, responding to commands the way a real system from that era would have.
For everyday Claude users, the more practical takeaway is about prompting strategy. The user essentially gave Claude a constrained world - "you are this specific machine in this specific year" - and let it operate within those rules. That kind of constraint-based prompting tends to produce more coherent, useful outputs than open-ended requests, whether you're simulating a 1998 PC or drafting a legal brief in a specific jurisdiction.
It's also a reminder that Claude's training data goes deep into the pre-LLM internet. The model has solid recall of 1990s software, hardware, and computing culture - enough to pass a nostalgia test from someone who was actually there.