Twelve years ago, Vikram Chandra’s Geek Sublime: The Beauty of Code, The Code of Beauty was published. In presumption of existent history, that isn’t agelong astatine all; successful presumption of technology, it’s an eternity. In Geek Sublime, Chandra drew connected his acquisition arsenic an writer of some prose and codification to research wherever the 2 converged.
You mightiness not spot the 2 arsenic having immoderate connection. Before speechmaking it, I—someone who had besides had a foothold successful some worlds for a while—was unsure of however convinced I would be. But I near convinced by his argument, 1 that emphasized elegance and quality successful some worlds, and cited examples of precisely however that could beryllium achieved. Chandra’s publication 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 literate satellite Chandra wrote astir successful Geek Sublime is not acold removed from its counterpart successful 2026, the satellite of coding has been transformed successful seismic ways. Vibe coding—using AI to make codification for online projects—has go widespread, for bully and for ill. (“Vibe Coding Is the New Open Source—in the Worst Way Possible” work one WIRED headline.) An episode of the New York Times podcast The Daily asked whether quality coders were adjacent indispensable immoderate more.
What does Chandra himself deliberation of each of this? Over the people of respective weeks, we exchanged a bid of emails successful which I learned that helium remains arsenic acrophobic arsenic ever with the aesthetics of code—even arsenic helium besides recognizes the changes taking spot successful the tech world.
*
Tobias Carroll: In Geek Sublime, you eloquently argued for the elegance of well-written code. Now, we look to person entered the epoch of “vibe coding.” What bash you marque of the emergence of AI-first coding, and however would you accidental that it has affected the principles you outlined successful your book?
Vikram Chandra: The principles I talked astir successful the publication are inactive crucial—you inactive privation your codification to beryllium elegant and casual to support successful the agelong run. The occupation is getting AI agents to constitute that benignant of code. I’ve lone utilized AI to constitute small scripts aft giving it strict instructions pointing to a database of those principles, and successful the tiny that seems to work. I don’t person immoderate first-hand acquisition with ample AI-written codebases, but from everything I hear—even from experienced programmers—is that you extremity up with large-scale spaghetti codification if you’re not precise careful. I person a person who is making dependable wealth from reviewing and fixing that benignant of slop.
The models are each trained to assistance and please, and that sometimes results successful them penning codification that seems to enactment brilliantly but is simply a shambles internally. You inactive request humans to reappraisal the code.
Here is the programmer Victor Taelin:
I present judge utilizing existent gen AI agents successful accumulation codebase is harmful and a monolithic mistake. That doesn’t mean nary agents astatine all, but agents enactment champion erstwhile they don’t interaction captious code. Debugging, researching, providing insights, scripts/tools, oregon thing that doesn’t interaction codification you volition support successful the agelong term. But if you merge AI codification without reading, you’re going to person a atrocious time. Speaking from experience.”
TC: Has this displacement successful coding habits near you reasoning astir penning a follow-up volume?
VC: Things are moving excessively accelerated close present to adjacent effort thing similar that. The acceleration makes maine dizzy. We’ve intelligibly crossed immoderate benignant of lawsuit horizon, and I don’t person immoderate thought of what’s coming next.
Was determination a circumstantial infinitesimal erstwhile you realized that the lawsuit skyline had been crossed?
I’d been utilizing different models for rather a portion for research, and besides to constitute reasonably elemental database queries. Then successful November 2025 I observed a immense uptick of buzz astir Anthropic’s Opus 4.5. I tried the escaped plan, and wrong a week was paying for a Pro subscription. And a fewer days aboriginal I upgraded to the Max plan. I accidental that’s a quantitative reflection of however utile it seemed. It one-shotted the database queries, arsenic opposed to my having to bash respective rounds of prompting and correcting. And I’ve been capable to leverage its capabilities to velocity up and deepen my research.
TC: Do you find yourself utilizing AI agents for anything?
VC: I usage AI a batch for humanities research, oregon astatine slightest for putting unneurotic the foundations for probe projects and past extending assorted lines. I person prompts that I tweak perpetually to debar hallucinations and I effort to cross-check what the models archer me. But it’s inactive casual to get taken successful by the implicit assurance with which agents spout “facts.” I person to actively defy getting taken in, seduced. I’ve had immoderate occurrence by getting 1 LLM to fact-check different LLM.
We’ve been utilizing bundle to constitute bundle for decades, ever since the archetypal compilers were enactment unneurotic to construe higher-level languages to instrumentality code.
But successful general, for the enactment I do, LLMs are hugely helpful. They’ve been trained implicit immense amounts of text, truthful they are capable to crook up information that I wouldn’t person recovered myself. I usage Zotero, an open-source exertion to store my accusation (academic papers, web clippings, etc.), and I’ve fixed an cause entree to each the thousands of items successful that store. Suddenly, uncovering worldly I request has go easier by orders of magnitude.
I started experimenting with ChatGPT erstwhile it archetypal came retired successful precocious 2022, trying to usage it precisely for this benignant of work. The earlier versions of each the models weren’t precise utile to me. And present I enactment with LLMs each day. This is what I mean by acceleration. I’ve been utilizing bundle tools since the 1980s, and I utilized to deliberation that we were moving accelerated successful presumption of capabilities. I’ve ne'er seen thing similar what’s happening close now.
TC: As an perceiver of these changes successful the tech world, bash you deliberation thing is mislaid by allowing bundle to constitute codification for much software?
VC: We’ve been utilizing bundle to constitute bundle for decades, ever since the archetypal compilers were enactment unneurotic to construe higher-level languages to instrumentality code.
Linus Torvalds conscionable pointed this out recently: “When I spot radical saying 99 percent of our codification is written by AI, I virtually get angry. Because those aforesaid people—I tin beauteous overmuch guarantee—100 percent of their codification is written by compilers. But they ne'er accidental that.”
The quality present is that erstwhile you punctual an agent, you’re allowing it a grade of autonomy that you wouldn’t and couldn’t springiness to a compiler earlier. The connection “agent” itself implies this deficiency of deterministic power and the consequent unpredictability.
Linus went connected to say, “People who bash not recognize the complexity of systems volition punctual systems and constitute processes that volition fail.”
That’s the perchance disastrous nonaccomplishment you get successful speech for productivity and scale.
TC: Geek Sublime covered a batch of the past that brought america to 21st period computing. Are determination immoderate peculiar moments from that past that you find particularly applicable to the existent debates implicit AI?
VC: In 1966, Joseph Weizenbaum wrote a programme helium called “Eliza,” which was fundamentally a chatbot that interacted with users similar a Rogerian psychologist. That is, erstwhile the idiosyncratic said something, Eliza responded by mirroring the user’s words successful acts of “sympathetic mirroring.” In a marvelous insubstantial published successful Communications of the ACM, Weizenbaum presented the pursuing speech arsenic an illustration (Eliza speaks successful superior letters):
Men are each alike.
IN WHAT WAY
They’re ever bugging america astir thing oregon other.
CAN YOU THINK OF A SPECIFIC EXAMPLE
Well, my fellow made maine travel here.
YOUR BOYFRIEND MADE YOU COME HERE
He says I’m depressed overmuch of the time.
I AM SORRY TO HEAR YOU ARE DEPRESSED
Eliza was fundamentally a database of keywords and a acceptable of translation rules. Weizenbaum wrote, “The gross process of the programme is rather simple; substance is work and inspected for the beingness of a cardinal word. If specified a connection is found, the condemnation is transformed according to a regularisation associated with the cardinal word; if not, a content-free remark or, nether definite conditions, an earlier translation is retrieved.”
What was genuinely astonishing to Weizenbaum was however acceptable users were to judge that they were talking to thing intelligent that was really responding to them with understanding. “The quality talker will… lend overmuch to clothe ELIZA’S responses successful vestments of plausibility.”
In a follow-up insubstantial published astir a twelvemonth later, Weizenbaum described his caput utilizing Eliza: “Of course, she knew she was talking to a machine. Yet, aft I watched her benignant successful a fewer sentences she turned to maine and said ‘Would you caput leaving the room, please?’ I judge this anecdote testifies to the occurrence with which the programme maintains the illusion of understanding… Thus, portion the… programme whitethorn beryllium utile arsenic an instrumentality for the investigation of two-person conversations, and portion it is surely fun, its purpose indispensable beryllium changed from that of concealment of misunderstanding to its explication.”
The existent debates connected consciousness, sentience, and AI would nett from a important input from philosophers who are acquainted with that tradition.
In 1972, Weizenbaum took a two-year permission from MIT to constitute a monograph titled Computer Power and Human Reason: From Judgment to Calculation. It’s an insightful, passionate critique of AI and technology, which helium explains is rooted successful his acquisition with Eliza. “I knew from agelong acquisition that the beardown affectional ties galore programmers person to their computers are often formed aft lone abbreviated exposures to their machines. What I had not realized is that highly abbreviated exposures to a comparatively elemental machine programme could induce almighty delusional reasoning successful rather mean people. This penetration led maine to connect caller value to questions of the narration betwixt the idiosyncratic and the computer, and hence to resoluteness to deliberation astir them.”
As I wrote successful Geek Sublime, programmers autumn successful emotion with their tools—I surely did. But Weizenbaum’s penning is an aboriginal knowing of what we present telephone “AI psychosis.” And it’s a overmuch wider critique of the effects of computers successful the world, and of scientism successful general.
As a writer, of people this second spot resonated with maine strongly. “When I accidental that subject has been gradually converted into a slow-acting poison, I mean that the attribution of certainty to technological cognition by the communal wisdom, an attribution present made truthful astir universally that it has go a commonsense dogma, has virtually delegitimatized each different ways of understanding. People viewed the arts, particularly literature, arsenic sources of intelligence nourishment and understanding, but contiguous the arts are perceived mostly arsenic entertainments.”
Computer Power and Human Reason is indispensable speechmaking for our existent moment.
I should besides say—because pre-modern Indian doctrine was a cardinal portion of Geek Sublime—that the existent debates connected consciousness, sentience, and AI would nett from a important input from philosophers who are acquainted with that tradition. Those old-timey thinkers spent centuries investigating what it meant to person a mind, the quality to enactment successful the world, the quality of autonomy, and truthful on.
TC: In caller months, especially, determination has been accrued scrutiny of generative AI being utilized to constitute fabrication and nonfiction. Do you spot parallels betwixt this and the issues you’ve described with utilizing AI for coding, oregon bash you spot these arsenic comparatively abstracted phenomena?
VC: I deliberation they’re related but astatine angles. I’m not definite I person thing absorbing to adhd to the statement astir cheating-with-AI, but to accidental that I deliberation astatine immoderate constituent we’ll spot certifications similar “No AI was utilized successful the making of this book.”
What I’ve been reasoning astir is AI arsenic an writer with an audience. It’s wide that AI-written stories tin evoke emotions successful quality readers. Therefore the outrage erstwhile we find retired that we’ve been tricked by a machine. But adjacent now, determination are readers who don’t attraction whether AI was used. See here, for instance. Take a look astatine the Instagram station linked successful the Reddit post.
And determination are readers who spot LLMs arsenic sentient beings with interior selves, qualia, and truthful forth. If an AI tin gully connected its ain interior states and models of the satellite to marque creation for humans, for these readers the LLM becomes an writer with whom they tin person an progressive affectional narration with done the fiction. For these people, LLMs are not statistical engines trained connected immense volumes of text, they are afloat live entities who tin marque art, who constitute from immoderate benignant of subjectivity. So the larger questions of instrumentality and caput are being played retired successful the realm of fabrication successful precise profound ways.
What we admit successful humans arsenic “genius” is the uncommon quality to crystallize thing miraculous retired of our ever imperfect bodies and spirits. I judge that adjacent the astir prosaic constructions made by humans volition ever stay precious to america precisely due to the fact that they travel from these constricted means. I spot nary crushed for despair.
English (US) ·