Errors, Not Hallucinations

It can feel pretty isolating to reject generative AI in 2026, but author Cory Doctorow’s intent is more fundamental than that. “If you want to get the good without the bad,” he writes in the book’s first section, “you need to switch from fighting technologies to fighting bosses.”

This is far from a fringe sentiment, but for writers interested in actually convincing rather than choir-preaching, politically-charged writing needs to be not just rational but relatable. Even recognizing my bias in his favor, I really do think Doctorow’s success owes largely to his accessibility as a writer. His prose is crisp yet colloquial, his humor is cutting yet goofy, and his anecdotes are imaginative yet grounded.

The foundational arguments are nevertheless uncompromising. Doctorow makes no apology for his insistence on workers’ rights, and he demonstrates solidarity by choosing examples beyond the ones that directly impact him as a writer. In darkly fascinating case-studies throughout the book, he explains how LLMs are being used to exploit long-haul truckers, ride-share contractors, and creative workers of all stripes. Not even nurses are safe from this grotesque scheming, for Pete’s sake. I found his empathy for tech workers (namely the ones at large and dubiously-motivated companies) to be particularly compelling, especially because I sometimes fail to give them the same grace.

Doctorow’s underlying thesis is that technology should be evaluated based not on what it does, but who it does it to and who it does it for. By this metric (and many others¹), “AI” fails so spectacularly that Doctorow almost makes critique look easy. And he’s not satisfied simply with throwing shade–²he is explicitly advising readers on how to engage in the discourse.

Surprisingly, though, Doctorow pays almost no mind to the technology’s externalities. He recognizes the underlying human labor simply to deflate the perception of “AI’s” capabilities (not to highlight the horrors of that work), and he makes only passing reference to the environmental impacts of training models. These omissions don’t just weaken the argument; they make Doctorow’s limited tolerance for LLMs seem acceptable. In the book’s introduction, he writes, “Though I don’t brainstorm with AI or other forms of automated or random text generation, I don’t have a moral objection to these practices.” He also describes his own use of open-source local models for audio transcription. This all seems a lot more reasonable if you ignore the harm caused in the production of the models, and I’m disappointed that Doctor has done so.

That said, if there’s anything to be learned from the current political moment, it’s that infighting doesn’t get us anywhere. Although my stance is a little more hard-line than Doctorow’s, I still see tremendous value in the work he continues to do, and I still think you ought to read his latest.