top of page

LLMs: A Librarian's Revenge



There she is, sitting at the reference desk. The middle-aged librarian (though, let’s be honest, librarians exist in a quantum state of perpetual middle-age) with a messy bun and the requisite pencil stuck through it. Wearing a cardigan because the thermostat in every library is set to “Arctic Research Station.” The thing is, this isn’t just any librarian. It’s the world’s first Large Language Model (LLM), just wrapped in sensible shoes instead of JavaScript.


When I was in library school, we didn’t call it “prompt engineering.” It was known as the reference interview. It was more art than engineering. Someone would walk up to the desk and say something vague like, “I need a book about horses.” (If I was lucky enough to get that many details instead of just desperate gesturing toward the animal section!) Cool story, but what kind of horses? Are we talking Black Beauty or veterinary medicine? Racing statistics or My Little Pony fan fiction? Ancient cavalry tactics or that one person who’s definitely writing the next great American novel about a horse whisperer who can only whisper to zebras?


The skill isn’t just knowing where to find the information. It’s about mastering the art of need-finding through strategic questioning. Whether conducting a reference interview at a library desk, running a product discovery session, or engineering the perfect prompt for an LLM, the fundamental challenge remains the same: getting to the root need behind the surface request. It’s about developing a framework of questions that reveals context, uncovers hidden assumptions, and leads to better solutions. And let me tell you, nothing prepared me better for product management or prompt engineering than those years of helping people who asked for ’a book about horses’ when they needed everything from basic riding techniques to advanced veterinary journals.


But here’s where it gets interesting (and by interesting, I mean feeling vindicated). Each type of library operates in its own special universe. Academic libraries are full of sleep-deprived grad students who need that one specific paper from 1973 (and no, the other 47 documents about horse-adjacent topics won’t do). Public libraries juggle everyone from toddlers to retirees looking for their next beach read or how to turn their garage into a DIY smart home (spoiler: they’ll be back next week asking for books on how to live off the grid). K-12 librarians are wizards who can find books that make teenagers actually want to read. And medical librarians? They’re out there handling questions that can save lives while maintaining their quantum state of middle-age and perfectly temperature-regulated cardigans. No pressure!


Each library requires its own approach, context, and way of organizing and accessing information. Looking back, we built specialized knowledge bases and fine-tuned our ‘models’ while Silicon Valley was still convinced GeoCities was peak web design.



For years, we’ve been quietly getting our master’s degrees, understanding taxonomies, and creating systems to make complex information accessible and understandable. We were building information architecture when ‘user experience’ meant ensuring your website’s animated rainbow cursor matched your blinking text and your visitor counter hadn’t rolled over to 999,999. We were armed with nothing but card catalogs and the Dewey Decimal System (or Library of Congress) - the original binary search, just with more paper cuts.


The irony isn’t lost on me that I now work in tech. When I first entered this world, databases made immediate sense to me. Hello, it’s all information science under the hood. But when LLMs hit the scene? That’s when everything clicked. Suddenly, all those “boring” library skills were the hottest commodity in tech. Those years of Boolean searches weren’t just exercises in AND/OR/NOT logic. They were proto-prompt engineering, with more comfortable shoes and fewer Series A funding rounds.


Want to be a prompt engineer? That’ll be $200k a year, please, and thank you. That’s roughly the equivalent of 40,000 overdue book fines or one really, really comprehensive collection of horse-related queries. Every time I see another job posting for a ‘prompt engineer,’ I smile and adjust my cardigan (which, by the way, is now a tax-deductible professional expense).


So yeah, we librarians are having quite the moment. After decades of being dismissed as quiet book-pushers who really love the smell of old paper, our skills are suddenly tech’s hottest commodity. And I’ll admit it, I’m feeling pretty smug about the whole thing. All those years of Boolean searches and reference interviews were beta testing for the AI revolution.


Remember this the next time you’re trying to get ChatGPT to give you something valuable that you’re not just typing commands into a void. You’re performing a reference interview, carrying on a proud tradition of figuring out what humans want versus what they’re asking for. Only now, instead of saying, ‘So what exactly are you hoping to learn about horses?’ you’re typing elaborate prompts and praying the AI doesn’t hallucinate a response about unicorns (which, as any librarian can tell you, belong in the mythology section, not equestrian studies).


And hey, at least ChatGPT won’t give you the side-eye for returning your books two weeks late. Though I bet if we gave it a cardigan and set its temperature to ‘library standard,’ it would develop that skill naturally.


© 2025 by Lauren Morris

  • LinkedIn
bottom of page