Why the service desk holds this job together
Library assistant work splits two ways. One half is records and routine: checking materials in and out, registering new borrowers, sending overdue notices, keeping circulation records tidy. Software has been chipping away at that half since barcode scanners arrived, and self-checkout kiosks and automated sorters carried it further. The other half is people and place: helping a patron who cannot name the thing they are looking for, settling a dispute over a computer slot, getting a story hour ready, finding a book a child describes only by its cover.
Then there is the physical work. Returns get sorted and shelved by call number. Damaged spines get taped and repaired. Interlibrary loan boxes arrive and have to be unpacked, checked and routed. Our robotics read puts most of that in the dexterous humanoid tier, which means a machine would need hands and judgment good enough to move through a public room full of people and shelves. Nothing like that is working in branch libraries at any scale.
The pressure on this job is budget-shaped rather than robot-shaped. The Bureau of Labor Statistics counts about 85,520 library assistants, clerical, in the United States, with median pay of $36,910 (BLS, 2025), and projects employment falling 6.4% between 2025 and 2035. That looks less like work disappearing and more like fewer desks funded, longer kiosk lines and fewer openings for someone starting out. Our wider read on that pattern is in the guide to AI and entry-level jobs.
What software does, what it assists, and what stays with staff
Start with the tasks our split puts in the machine’s column: 0% of task time. These are the closed-loop clerical pieces. Issuing and receiving materials through a circulation system, generating overdue and hold notices, and keeping borrower records current are jobs a system can run end to end once the rules are set. The method behind that call is on the coverage scoring page.
Next, the assisted column: 33% of task time. Here a person stays in the loop and the tool speeds the step up. Copy cataloging and metadata cleanup, searching catalogs and databases for a patron’s request, and drafting routine replies to common questions all go faster with language models, but someone has to check the result against the actual collection on the actual shelf.
The rest sits with people: 67% of task time. Reference and reader’s advisory at the desk, shelving and shelf-reading, running children’s and community programs, handling conduct problems and privacy questions, and supervising volunteers or student workers all need a person in the room. Much of this is judgment about a specific patron in a specific moment, which is exactly what our scoring method treats as hard to hand over.
What the evidence actually shows
There is no published head-to-head test of AI against library assistants on their own tasks. Our evidence grade for quality parity reflects that: D. A grade at that level means the question of whether a system matches a trained assistant has not been measured, so we publish no parity number for this job rather than guessing one.
What would settle it is specific. A timed study of reference interviews, where patrons bring real, half-formed requests and raters compare a chatbot’s answers with a trained assistant’s, scored on whether the patron left with the right item. A field trial of automated sorting and shelving in a working branch, counting misshelved items and staff hours saved. Audit data on copy-cataloging accuracy when a model generates records that a technician then checks. Until work like that exists, the honest answer is that the task mix tells us more than any benchmark does. How we grade evidence is explained on the quality parity page.
When this could shift
Most likely after 2036 (8 in 10 of our scenarios). What that window measures is set out on the replacement year method page.
Two things could pull it earlier. The first is money: library software and kiosks are cheap next to funded staff hours, and a squeezed budget cycle can cut a desk shift before any technology is proven. The second is consolidation, where a system centralizes cataloging, notices and call handling across branches, so one hub does clerical work that used to sit in each building.
Two things hold it back. Shelving, repair and handling physical deliveries need dexterity and mobility that current robots do not have at a price a public library would pay. And libraries are public spaces with duty-of-care and patron privacy obligations: someone has to be accountable for a minor at a terminal, a confused visitor, or a records request. Those obligations are written for people.
How to stay needed
Lean into the work the split already leaves with people. Get good at the reference interview, where you turn a vague request into the right item. Own programming: story hours, job-help sessions, digital literacy drop-ins. Take on the floor work that holds a branch together, including conduct, privacy and access questions that need a named person.
Two skills raise your floor. First, library systems literacy: knowing the integrated library system well enough to fix bad records, run reports and train others. Second, reviewing AI output, which means checking a generated catalog record or chatbot answer against the real collection and spotting what is wrong.
What to do: ask to be the person who tests any new kiosk, chatbot or sorter in your branch, and keep notes on what it gets wrong.
Nearby work is worth a look. Library technicians handle more cataloging and systems duties, librarians and media collections specialists move into collection and program decisions, and file clerks show how a records-heavy mix reads differently. You can put any two side by side in the job comparison tool, see the rest of the information and record clerks family, check the wider education sector page, or scan the list of jobs expected to shrink for the same pattern elsewhere.