Professional gossip fuels many good lunches. In the past year, it’s always been about the same thing. This is mostly my fault: these days, I have a one-track mind. I sit down, I start asking questions.

Broadly speaking, there seems to be an industry-wide understanding that generative A.I. threatens nearly every stage of the publishing process. These threats are legal, contractual, editorial, creative, moral, existential: we fight for language prohibiting A.I. use in contracts all the while knowing how impossible it is to police; we watch the pool of work for human translators and designers shrink; we read about the literacy rate plummeting; we click through The New York Times quiz to see if these A.I. passages can fool us; we wonder for how long we will all have our jobs. Every day, a new horror pops up, another issue you hadn’t considered, and every other issue that’s come before has still not been solved. Everybody is angry and everybody is panicking. 

And yet under that broad umbrella of resistance, opinions begin to fracture. Often I find myself sitting down across from a fellow hater who rejects the technology entirely, who doesn’t even use the walled garden LLMs set up by their companies. I learn which corporate executives are chugging the Kool-Aid and asking subordinates to do the same, which editorial directors are protecting their imprints. We trade stories — did any of your authors get an Anthropic payout? Did you see what Olga Tokarczuk said? Did you hear about the publisher who “accidentally” circumvented agents to send authors vaguely-worded contract amendments about licensing their work to A.I. programs? Round and round we go. 

And then, occasionally, I’m caught off guard. An older mentor casually mentions they don’t see the problem with letting Claude organize notes into an editorial letter. An agent bemoans an inbox overstuffed with A.I.-written queries and a minute later admits that it’s a great tool for drafting a rejection. These have always been tasks we have needed support for. They were also the first tasks I was entrusted with as an intern and then an assistant. They’re how you learn the job. 

I reject the idea that this technology’s proliferation is inevitable, that we must bend the knee. In my optimistic moments, I like to think that this crisis will lead us to value human-made art even more, and that publishing and other creative industries help safeguard that. At the same time, this additional responsibility — being the authenticators — adds to the plate of an already deeply overstretched workforce.

The high-profile scandals we’ve seen so far, those books pulled for being fully or majority A.I.-generated, are concerning. But there have always been those who want to be writers without having to write. It’s simply easier to get away with now. I do not care if I can be fooled — if I haven’t already, I’m sure I will be eventually. When I learn something has been written or aided with the help of A.I., my interest evaporates no matter the quality — though more often than not the quality is bad. The magic is gone. Are we not drawn to the art as much for the process as the product? 

Instead, I worry about the gray areas, about anyone in the life cycle of the book — from author to editor to publicist and beyond — growing reliant on using generative A.I. tools as a crutch for editing or revising or brainstorming. I now ask writers to disclose their A.I. use before signing and will not sign anyone who uses LLMs for any of those gray areas, even if only occasionally. The slope is too slippery. I reach toward writers who feel as strongly as I do, and abstention, at least for now, feels far safer than moderation.

I mentioned this policy on a panel recently, after which several concerned writers came up to me to argue their various use cases. You don’t want me to be your agent, I told them. The relationship, like any, is built on trust. Don’t you want to be able to trust each other?