A new retrospective from Gwern.net has taken fresh aim at one of computing’s most famous unrealised visions: Ted Nelson’s Project Xanadu.
Published as “Project Xanadu: Even More Hindsight,” the essay revisits the history of the hypertext system Nelson began advocating decades ago and argues that the project’s collapse was not a mystery of destiny but a consequence of ordinary product failures. In the author’s telling, Xanadu never turned a striking concept into something durable because it lacked steady design iteration, did not settle on meaningful use cases and struggled to become practical software rather than an idea that lived mainly in presentations, debates and nostalgia.
The essay is framed partly through a personal memory of a 2024 San Francisco gathering marking the 50th anniversary of Computer Lib/Dream Machines. The author describes seeing vintage Xanadu machines, rare Nelson memorabilia and hearing stories from a former Autodesk-era Xanadu programmer. That backdrop matters because the piece uses it to contrast the romance of hypertext with the hard constraints of older hardware and software development. The point is not just that the ideas were ambitious. It is that the environment in which they had to be built made a great many promising systems harder to finish, test and ship.
One of the essay’s central themes is that many computing disappointments look different when you remember how little RAM, storage and tooling earlier systems had. The author uses that lens to explain why elegant-sounding software often did not become practical software. In the Xanadu case, the argument is that the vision was compelling enough to influence generations of technologists, but the implementation never resolved the long gap between conceptual elegance and a product people could actually rely on.
The retrospective also places Xanadu in a broader discussion about scalability, distributed systems and how software architectures age. The linked material makes the case that “scaling” is not a single property but a set of trade-offs across compute, storage and networking. That framing helps the essay’s larger claim: projects fail not only because they are technically difficult, but because they miss the combination of performance, developer experience and real-world usefulness needed to survive.
Gwern’s conclusion is not that hypertext was a bad idea. It is that a good idea can still lose if it cannot survive contact with engineering reality. The essay repeatedly points to iteration as the missing ingredient. Without repeated refinement on actual tasks, a design can remain compelling forever and useful never.
For readers outside the software-history niche, that makes the piece more than a tribute to a famous dead-end. It is a warning about how technology myths are formed. Systems become legendary when they inspire people, but they become infrastructure only when they can be built, debugged and adopted at scale. Xanadu, in this account, became the former and never the latter.
That is why the article resonates beyond retrocomputing circles. It is as much about the habits that turn technical ambition into lasting software as it is about Ted Nelson’s original hypertext dream. The essay’s thesis is that vision alone is not enough. To matter, a system must also be iterated into something practical enough for ordinary use.
