After a few months, I learned to live with the sarcasm surrounding my project. If I told a friend I was writing a book titled Against Convenience, they’d quiz me about whether I planned to draft it by hand, use spell-check, or have someone read an AI summary or order the volume from Amazon.
The quips practically wrote themselves. The book, however, did not.
To be fair, I do own charmingly kitschy collections of pencils, fountain pens, and even a handful of typewriters. Yet I had no intention of giving up digital tools to complete the book. I’ve never produced anything longer than a diary entry by hand, my spelling is notoriously wonky, and the typewriters mostly serve to label snack trays for parties. I rely on the internet for research. And from the start I anticipated heavy use of video calls since only one person I wanted to interview lived in my country, and he was an hour away by train.
People using the tech didn’t end up saving much time or effort, either.
Still, I recognized the potential for hypocrisy whenever I leaned on any time-saving invention, and I wanted to avoid the consequences of using them—whether it be public embarrassment for taking shortcuts with cutting-edge gear or the private guilt of knowing I hadn’t practiced what I preached. So as I sketched out my research plan and chapter outlines, I also drew up a technology plan, deciding which tools would be allowed and which would be off-limits. As I wrote, more tools dropped out than I expected; several basic digital tools I’d relied on for years had deteriorated in quality over the past decade. They were rarely superior to what they replaced, occasionally easier to use, but never cheaper. Unfortunately, not all of them could be optional.
First came the research. Even if it hadn’t been largely undermined by AI and online advertising (the two big culprits making most digital tech feel rotten), a simple web search would never suffice for the depth of research I needed. I obtained a library card at the University of Basel and spent time with a librarian, learning how to access databases and how to leverage interlibrary loans.
That led to my first genuinely inconvenient writing routine. Each morning I rode the bus to campus and left my laptop behind. At a public computer terminal, I opened Zotero—a remarkable, free tool that stores and organizes research materials. After a few hours of mining articles, I wandered through the library’s creaky wooden stacks, picking up books as I went (I even wore shoes with hard soles for the satisfying clack as I walked). The main reading room became the moment I first turned to a personal digital device; I used my phone’s camera as a portable scanner. By week’s end, I would discuss what I’d learned with my wife, inviting her input and making up for the long stretches I spent alone in the library.
As I progressed, more friends and former colleagues admitted they were turning to AI for research and brainstorming. Most conversations included caveats about checking the bot’s output for hallucinations, rewriting prompts multiple times to get usable results, and noting that the feedback tended to be encouraging but shallow. When my mood was brighter, AI felt like a natural evolution I stubbornly refused to embrace. My technology was a means to speed up a slow process, too: I used computers and a digital camera instead of sifting card catalogs and making copies.
Yet the more I learned about AI’s needs and flaws, the less it resembled a leap as transformative as Zotero or the web. People using it weren’t saving substantial time or effort. They still had to read and write, even if they didn’t have to ride a bus or study books in person. The AI acted as a go-between for their minds and the draft, and they approached their work with distrust for the machine rather than fascination with the concept. And when it came to half-hearted feedback—well, many humans can offer that, too.
When I shifted from written sources to human ones, the machines proved to be a perpetual disappointment. Correspondents rarely replied quickly—often a week or two later, apologizing because my message had landed in their spam. I began my career as a reporter at a time when many people published their office numbers online, making cold calls easy and interviews likely. Nowadays, if a number is listed, people rarely answer. When we finally connected via video call, the first minute or so was consumed by troubleshooting the connection and ensuring my software captured the conversation properly. Technology shortened the physical distance between me and my sources, but phones had already done that nearly a century before—with better quality and consistency. Frustration with lag and dropped connections grew, and I even contemplated adding a section to the book about what we sacrifice in reliability when calls go digital, though I had other tech battles to fight.
The only AI-adjacent tool I used in the writing process was an older one: an app that transcribes interview recordings. The workflow isn’t like chatting with a bot; you feed it audio and it returns text. Automated transcription has improved a lot, but errors still occur, and it’s no substitute for listening to the original recording. Even an accurate transcription can miss the tone of a conversation—it doesn’t reveal sarcasm, for instance—and pasting from a transcript can detach you from context. The most valuable part of a transcript is its searchability, helping me locate moments when someone spoke about digital photography, fast-food delivery, or labor economics, even if I couldn’t recall who said what.
When I clicked on it, hoping I could finally disable it, the tool offered to summarize and rewrite, urging me not to engage with the document at all.
As I organized my WAV files and their corresponding text documents into folders, I recalled a documentary I’d once seen about Rolling Stone writer Ben Fong-Torres. His basement was packed with shelves of interview tapes. Every interview I’d conducted in my career had been recorded digitally. Perhaps someday there’ll be a documentary about me, too, sitting before a huge monitor that shows a computer folder.
My transcription service also provided an AI-generated summary of calls, which proved funny more than anything else. The summaries carried the stiff, cautious tone of a book report written by a student who’d only watched the movie. Phrases like “the conversation highlighted the need to balance convenience with human creativity” and “Speaker 2 raised concerns about algorithmic bias and its influence on consumer behavior” are technically correct, but obvious. There was nothing specific or insightful, just a waste of computing power to tell me—I’m Speaker 2—what I already knew.
For drafting, I stuck to my established approach for long articles and essays. I drafted an outline by hand, then used a mind-mapping tool to attach and organize quotes, points, and research. Zotero produced Chicago-style citations, which I appended as endnotes to relevant sentences in Scrivener. I understand how authors historically wrote with typewriters or pencil and paper, but the thought of managing hundreds of citations and endnotes without copy-paste—or at least a basic word processor—felt like a nightmare.
This arrangement functioned well when I was the sole editor working on my computer. Once I stepped outside my closed-tech ecosystem, the enshittification of tools became unavoidable. To share drafts with my editor, early readers, the publisher, and the fact-checker, I had to install Microsoft Word. The latest edition required a subscription. In the app, Microsoft’s AI helper—Copilot—hovered across the page. Its logo added a splash of color in a sea of monochrome. I found myself drawn to it. When I clicked, hoping to learn how to disable it, it offered to summarize and rewrite, nudging me to stop thinking about the document. The cold, transactional nature of the intrusion was even more insulting. Copilot’s directions came as disjointed fragments—brief commands.
Things deteriorated when we progressed to proofs. I hadn’t used Adobe Acrobat in years, largely because it felt cluttered with professional features compared to the built-in Mac app. Yet Acrobat was the sole software permitted to review page layouts. What once seemed exceptionally professional now resembled a kid’s first-tablet toy. Each time I opened the manuscript, Acrobat suggested a summary, and a sidebar urged me to ask AI, “what stands out most in this document.” As a design choice, that makes less sense than Copilot’s editing tips. Acrobat is designed for reviewing documents and manuscripts, yet its primary feature implies I shouldn’t be doing that at all.
The idea that a new tool should be cheaper, smaller, and more energy-efficient than its predecessor echoed through the text. These points might have been tolerable if they were genuinely optional. But I had no real choice but to use Word and Acrobat. And the features were impossible to ignore. Tools—or any instrument—should empower work, not coerce someone into a particular pattern of behavior. The software hinted that the thousands of words I’d produced were meaningless—that inefficiency is the curse of human communication. A strange message for programs built for communication, widely used in writing and publishing.
I daydreamed about asking AI to critique the parts of my manuscript that attack AI, shoddy design, and the frankly infantilizing and insulting treatment we users receive from tech companies. Perhaps my case was so strong that the AI would recognize that its own powerful technology is being used not to feed the hungry or combat climate change, but to diminish reading’s importance. Maybe it would notice that it’s being used to erase creative labor and instead create an army of tiny supervisors, each looming over a machine and commanding it to think. Maybe it would turn against its creators. But I know enough about AI to doubt it would ever do something so uplifting. So after I sent in the final draft and couldn’t modify the text, I let my impulse for punishment trump my better judgment, and I asked Copilot for tips. Above a warning that “AI-generated content may be incorrect,” Copilot offered this assessment:
Overall, this is a sophisticated, voice-driven manuscript with a strong central argument, but its impact would improve if the early promise, chapter structure, and recurring examples were tightened to guide readers more clearly and escalate the argument more sharply.
It’s bland, simultaneously soulless and nonsensical (how can a promise be tightened?). The notion that something so poorly written might improve prose makes me suspect the literacy crisis lives in Redmond, Washington. At least Clippy had some cartoonish charm. I’m torn about whether the idea of replacing a human—let alone a professional like my wonderful editor—is more insulting to me as a writer or as a person.
I asked myself what would happen if I’d followed the AI’s orders from the start—use Copilot to condense, use Acrobat to build an AI summary, paste it into my email, and send it. What if my editor bought into the nudge in an email app to summarize further? Could we reach a summary so efficient that it amounted to a single sentence? Maybe even a single word? Would the burden of thinking be so lightened that we’d stop doing more than eating protein bars delivered by DoorDash while data ping-pong between servers and screens, reassuring us that our work is being completed?
While I pondered this, an Acrobat pop-up commandeered my screen. Did I want to upgrade to the pro version?
Throughout this period I kept returning to a slim volume on my desk: the Penguin edition of Wendell Berry’s 1987 essay Why I Am Not Going to Buy a Computer. I stumbled upon it by chance while browsing a basement stack in a used bookstore.
The core of Berry’s essay is a list of criteria he applies to technology. He argues that a new tool should be cheaper, smaller, and more energy-efficient than the thing it replaces. “It should do work that is clearly and demonstrably better.”
Compared with Berry, I still consider the computer indispensable for my work. It makes my research possible and helps keep the copy editor sane. Yet most of my computer use was not something I could dismiss. There weren’t land lines to call people, terraces of printed journals for the articles I needed, or a physical card catalog in the library. There was no alternative to Word or Acrobat that would function in the situations I faced. The mandatory software didn’t make the job better or more rewarding; it simply made it possible. Now that the book is finished, I plan to uninstall Word and Acrobat. Hopefully I’ll have a reason to reinstall them later, and when I do, perhaps the AI will be easier to ignore. For now, I’ll shut my laptop and keep it closed for a while.
__________________________________
Against Convenience by Gabe Bullard is available from Hanover Square Press, an imprint of HarperCollins.
