It’s barely an exaggeration to say that the advent of artificial intelligence (AI) hit the literary world like a once-a-century earthquake—and that it continues to set off aftershocks of dread as new issues emerge. Writers have been justifiably alarmed that their work is being fed, generally without their knowledge or consent, into large language models (LLMs) to train AI algorithms to emulate human prose. They’ve been aghast that their pay for this unwitting participation in the AI revolution is often…zilch. (The AI industry was valued at $279.22 billion in 2024, according to Grand View Research.) And many writers are grappling with a more fundamental fear: that AI-generated books could someday crowd out those written by humans or, at the very least, hugely devalue them.
But what do agents think about all of this? What worries them most about AI’s steady encroachment? How are they seeing it show up in their daily work? Could it ever, dare we say it, play a positive role in the industry? I spoke to six agents about how they view AI in the context of their profession. Notably, several agents to whom I reached out opted not to comment. This reticence speaks, I think, to just how new and unnerving this technology is. Most cited a lack of knowledge on the matter; some agents, it seemed, would prefer not to have to engage such questions at all.
“I think there are two ways of thinking about it,” says agent Sarah Burnes, who’s been with the Gernert Company since 2005 and represents me. “One is the legal issues and who owns what and how, because there’s this question of how much are [the rights to train AI with your work] worth. They’re not worth very much for a single piece of work, but in aggregate they’re worth billions. And the second way of thinking about it is that it’s existential—these companies are using this work to create systems that will totally change the world.” Not necessarily for the better, Burnes makes clear, and not solely in the realm of publishing: “The idea that whatever will happen with AI won’t create systems that destabilize our democracy, I think is just foolish.”
Burnes also believes that the forward march of AI echoes other paradigm-shifting tech rollouts from years past, which have demanded new ways of being and new ways of publishing. “We’ve been through it with Google, we went through it with Twitter, we went through it with Meta,” she says. “In terms of contracts it’s similar to when e-books started becoming an issue in the mid-nineties, and publishers had to work out new legal language.” But Burnes indicates that with AI, the threat feels deeper and more direct. “It’s a neat tool,” she says. “But we as a culture need to move beyond thinking that something is a neat tool, and we can’t just stumble into this the way we’ve stumbled into everything we’ve stumbled into in the digital age.”
I ask Burnes if she would be put off by the knowledge that one of her writers was using, say, ChatGPT to organize their thinking or conduct some of their research. “It’s a slippery slope,” she says—particularly since any work produced using generative AI is not copyrightable. Blur the line too much between “bouncing ideas around” and “straight up copy-and-pasting” and a writer risks endangering their rights to their own work.
This is echoed by agent Hannah Bowman, speaking on behalf of the Association of American Literary Agents’ AI Special Committee. “When writers submit their work for publication, they are generally warranting—promising—that the work is their own intellectual property and can be copyrighted,” Bowman says. “For this reason alone, they need to be cautious about any use of generative AI. Some publishing contracts are now also requiring the author to warrant explicitly that AI was not used in producing the work.”
A good deal of the mental labor of writing happens not just in the generating of prose, but in the thinking it through. Burnes mentions a Bluesky exchange among academics about AI in the classroom setting: “Using AI in education,” one person had written, “is like using a forklift at the gym.” That principle is just as applicable to the crafting of prose.
“On the one hand,” Burnes says, summing up her position, “I want to understand as much about this as I can and advise my clients to the best of my ability; but on the other I want to never lose the sense of the existential threat. We can use AI to do machine things. We don’t need it to do human things.”
Some disagree with that statement: They think we shouldn’t use AI for anything. “I wish I could be more sanguine about the opportunities—there will be opportunities,” says Pamela Malpas, an agent of more than two decades who is now with Jennifer Lyons Literary. “The best I can say is that it is an opportunity to learn something.” Ian Bonaparte of Janklow & Nesbit Associates says AI “wreaks huge ecological effects on our fragile planet and seeks to transfer power to those who deserve it least,” calling AI and other technologies like it “morally bankrupt.”
But other agents see a path forward—albeit one studded with confounding roadblocks. Michelle Brower, cofounder of Trellis Literary Management, says her “biggest concern” is the need for remediation for authors whose works were used to feed LLMs “without permission or compensation.” As she puts it, “The legal challenges against that theft will take years to resolve. All the while, AI tools are being built atop that data and are becoming more sophisticated, and there is no clear way to disentangle the stolen work retroactively.” That’s both an ethical and a financial problem, Brower believes. “So much money has been put into AI,” she says, “that it’s going to be impossible to claw back the true value of those original stolen works.”
Burnes told me that a client was recently approached by a Big Five publisher on behalf of a tech giant building its own LLM. The author was offered a onetime sum in the mid five figures to allow their book to be fed into that model. (The author declined.) Brower says she has been similarly approached. “A publisher wanted to license a book for inclusion in a dataset for LLM training, but my current recommendation is to not license that right yet,” she says. “There is so much we still don’t know about how the underlying work will be used and what recourse the author has should they change their mind.” Ultimately, however, Brower predicts that these are conversations that will need to be had. “I think there will likely be a business model that develops around licensing, and that will come up in contracts discussions,” she says. “But we have no idea what that will look like yet.”
However, the AALA’s AI Special Committee does have an idea about this. Hannah Bowman shares that the organization’s position is that “unless explicitly granted to a publisher in a publishing agreement, the right to license a work for the training of generative AI rests with the creator or copyright holder and should not be used without consent and compensation.” Additionally the “AALA favors the establishment of licensing mechanisms and marketplaces to the extent that they facilitate an opt-in model for copyright holders.”
Clare Mao of Sanford J. Greenburger Associates is just as rankled by published works being fed to AI without compensation and has begun proactively changing how her deals are structured as a result. “On a contractual level,” she says, “I strive to include language that limits how publishers can use an author’s work or make it accessible to AI training or prohibits the use of AI in an author’s design, marketing, and editorial process, though we don’t always succeed.”
I asked the agents whether they had received any query letters that they suspected were written by ChatGPT—or if they had ever worried that an actual manuscript might bear the influence of AI. Most said no. But Bonaparte shared a curious anecdote: “What I find very strange is that having represented a book called The AI Con: How to Fight Big Tech’s Hype and Create the Future We Want that virtuosically dissects the paper tiger of AI, I continue to receive AI-generated queries and ‘AI-collaborated’ queries about the wonders and unlimited creative promise of AI.”
Burnes mentions that while she has happily never received any queries bearing the telltale quirks of AI when her agency was hiring an assistant two years ago, she noticed that many of the cover letters were oddly similar. “I was talking to my brother about it, and he said, ‘Sarah, they’re using ChatGPT.’ Which I thought was completely wild. You’re applying for a job where a good portion of it is reading and analyzing what you’re reading and then writing about it, and you’re using AI to get in the door?”
However, Malpas makes the point that being able to discern the presence of AI in a text “isn’t really the point.” To her the root issue is that AI has so thoroughly muddied the waters for all of us—agents and readers alike. “The problem,” she says, “is that we must now question everything we receive to consider whether it might be a product of AI.”
If you’re looking for a glimmer of hope on the subject, you may be glad to hear that a few agents were willing to consider potential upsides. “The main benefit I can see,” says Brower, “is that AI does seem really useful for parsing data for trends, and that could potentially help publishers find and reach readers on a smaller and more diffuse scale.” Mao was less concerned than others by the prospect of AI ever fully usurping writers or rendering their work obsolete. “I don’t think AI is at a place yet where it can effectively capture that most elusive quality that I and editors are always looking for: an author’s voice,” she says. “I’m not sure an LLM ever can, no matter how well trained it is.” Fingers crossed.
Katie Arnold-Ratliff, author of the novel Bright Before Us (Tin House, 2011), is a freelance journalist and editor.
Thumbnail credit: Dori Flynn