Twelve years ago, Vikram Chandra’s Geek Sublime: The Beauty of Code, The Code of Beauty was published. In terms of actual history, that isn’t long at all; in terms of technology, it’s an eternity. In Geek Sublime, Chandra drew on his experience as an author of both prose and code to explore where the two converged.
You might not see the two as having any connection. Before reading it, I—someone who had also had a foothold in both worlds for a while—was unsure of how convinced I would be. But I left convinced by his argument, one that emphasized elegance and beauty in both worlds, and cited examples of precisely how that could be achieved.
Chandra’s book was a finalist for the National Book Critics Award for Criticism, with Anne Trubek hailing “the subtle and revelatory scope of this book.” While the literary world Chandra wrote about in Geek Sublime is not far removed from its counterpart in 2026, the world of coding has been transformed in seismic ways. Vibe coding—using AI to generate code for online projects—has become widespread, for good and for ill. (“Vibe Coding Is the New Open Source—in the Worst Way Possible” read one WIRED headline.) An episode of the New York Times podcast The Daily asked whether human coders were even necessary any more. What does Chandra himself think of all of this?
Over the course of several weeks, we exchanged a series of emails in which I learned that he remains as concerned as ever with the aesthetics of code—even as he also recognizes the changes taking place in the tech world. Tobias Carroll: In Geek Sublime, you eloquently argued for the elegance of well-written code. Now, we seem to have entered the era of “vibe coding.” What do you make of the rise of AI-first coding, and how would you say that it has affected the principles you outlined in your book?
Vikram Chandra: The principles I talked about in the book are still crucial—you still want your code to be elegant and easy to maintain in the long run. The problem is getting AI agents to write that kind of code. I’ve only used AI to write little scripts after giving it strict instructions pointing to a list of those principles, and in the small that seems to work.
I don’t have any first-hand experience with large AI-written codebases, but from everything I hear—even from experienced programmers—is that you end up with large-scale spaghetti code if you’re not very careful. I have a friend who is making steady money from reviewing and fixing that kind of slop. The models are all trained to help and please, and that sometimes results in them writing code that seems to work brilliantly but is a shambles internally.
Extract — continue reading at the source.
