The conversation about AI and work is stuck on the wrong verb.
Everyone says "replace." AI will replace writers. AI will replace programmers. AI will replace lawyers and accountants and radiologists. The framing is comforting in a strange way, because replacement implies a one-to-one swap — a human was doing a thing, now a machine does the thing, the thing still exists. If the thing still exists, maybe there's a way back.
I want to introduce a different verb: eliminate.
Not replace the worker. Eliminate the work. Dissolve the underlying need so completely that the job, the service, and sometimes the entire industry simply stops making sense.
This distinction matters more than almost anything else in the AI-and-employment conversation, and almost nobody is talking about it clearly.
The GPS Precedent
In 1990, Rand McNally employed hundreds of people to research, design, update, and distribute road atlases. These were physical books of maps that millions of Americans kept in their glove compartments. The company also employed a network of field researchers who drove roads to verify that the maps were accurate.
GPS navigation didn't replace map-readers. It didn't hire robots to sit in passenger seats and read atlases out loud. It made the entire concept of needing a paper map obsolete. The job of "map reader" didn't transfer to a machine. The job of "map reader" stopped existing, along with the jobs of the cartographers, the printers, the field researchers, the distributors, and the retail shelf space dedicated to road atlases.
Rand McNally still exists — barely, as a fraction of its former size. But the important thing is not that the company shrank. It's that millions of consumers stopped needing a thing that used to be essential. The demand disappeared.
This is elimination. And it's a fundamentally different economic phenomenon than replacement.
Why the Distinction Matters
When a job is replaced — when a machine does what a human used to do — several things remain true. The task still exists. The quality standards still exist. The human who was displaced has skills that are relevant to the task, and can potentially supervise, manage, or complement the machine. Retraining is at least theoretically possible, because the domain of work is still there.
When a job is eliminated — when the underlying need is dissolved — none of this applies. There's nothing to retrain for. The skills aren't transferable, because the destination has vanished. The worker isn't competing with a machine for the same task. The worker's entire category of contribution has stopped being necessary.
Replacement is a labor problem. Elimination is a demand problem. We have policy tools for labor problems — retraining programs, wage subsidies, transition assistance. We have very few tools for demand problems, because demand problems mean the market itself has changed shape.
Where Elimination Is Already Happening
Travel agencies. This one is nearly complete. In 1995, there were approximately 124,000 travel agencies in the United States. By 2024, fewer than 15,000 remained. Online booking didn't replace travel agents with robot travel agents. It eliminated the information asymmetry that made travel agents necessary. When every consumer can compare flights, hotels, and packages directly, the intermediary role evaporates.
The remaining agencies serve luxury, corporate, and complex itinerary markets — niches where human judgment still adds value. But the mass-market travel agent, the person who booked your family's flight to Orlando, is gone not because someone does that job better, but because that job stopped needing to exist.
Photo processing. Kodak employed 145,000 people at its peak in 1988. Digital photography didn't replace film processing with digital processing — it eventually eliminated the need for processing altogether. When Instagram launched in 2010, it had 13 employees. The industry didn't shrink. The industry's reason for existing changed so fundamentally that the old jobs became unintelligible in the new context.
Bank tellers (partially). ATMs didn't fully eliminate bank tellers — the number actually increased for a while as banks opened more branches. But mobile banking and digital payments are doing something ATMs didn't: eliminating the need to go to a bank at all. The long-term trajectory isn't "machines do what tellers did." It's "the thing tellers existed to facilitate no longer requires a physical location."
Where AI Elimination Is Underway
First-draft professional writing. I don't mean that AI replaces writers — some of it does, and that's the replacement story. I mean that entire categories of documents that used to require professional writers are ceasing to exist as discrete deliverables. Internal reports that used to be written and circulated are being replaced by dashboards that update automatically. Meeting summaries are generated from transcripts. Status updates are compiled from project management data. The documents still exist in some form — but the idea of a person whose job is to write them is dissolving.
Basic legal research. Junior associates and paralegals historically spent large portions of their time on legal research — finding relevant precedents, reviewing documents for applicable clauses, summarizing case law. AI tools don't just do this faster. They're beginning to make the standalone research task unnecessary by integrating it directly into drafting tools. You don't research and then write. The tool writes with the research built in. The task boundary has dissolved.
Language translation (for routine content). Human translators are not being replaced by AI translators for literary or nuanced work. But the vast market for routine translation — product manuals, website localization, business correspondence — is being eliminated by tools that produce acceptable translations as a feature of the platform, not as a service purchased separately. The demand for "translation as a standalone service" is shrinking not because the translations got better, but because the need to hire a translator is disappearing from the workflow.
Stock photography. AI image generation isn't replacing stock photographers one-for-one. It's eliminating the concept of needing to search for and license a pre-existing image. When you can generate exactly the image you need in seconds, the 300-million-image stock photography industry isn't being competed with. It's being routed around.
The Retraining Problem
Here's why the replacement-versus-elimination distinction matters for policy: most workforce retraining programs assume that displaced workers can be moved into adjacent roles. The coal miner becomes a solar panel installer. The factory worker becomes a technician.
This assumption requires that the destination job exists. Elimination removes the destination.
A travel agent can't be retrained as an online travel agent, because online travel agents aren't a thing — the platform is the agent. A legal research paralegal can't be retrained as an AI-assisted legal researcher, because the role of "someone who does legal research as a primary function" is what's disappearing.
The World Economic Forum and McKinsey both project that new job categories will emerge to absorb displaced workers. This is historically true — new technologies have always created new jobs. But the new jobs often require fundamentally different skills, appear in different geographies, and take years or decades to materialize at scale.
The gap between elimination and emergence is where real people live. It's measured in mortgage payments, health insurance lapses, and children who notice that something at home has changed.
What Honesty Looks Like
The honest version of the AI-and-work conversation would sound something like this:
Some of your jobs will be replaced. That's the manageable part. Replacement is what labor markets know how to handle, even if they handle it imperfectly.
Some of your jobs — and this is harder to hear — will simply stop existing. Not because someone does them better, but because the conditions that made them necessary have changed. This is the part that retraining programs, career counseling, and economic policy are not well-equipped to address, because it requires not just new skills but new categories of work that may not have names yet.
The people most vulnerable to elimination are not the ones doing the lowest-skilled work. They're the ones doing work that exists primarily because of an information gap, a coordination problem, or a communication bottleneck that AI can dissolve. These are often educated, experienced, middle-class professionals who reasonably believed their skills were durable.
I don't have a tidy solution for this. What I have is a suggestion: start paying attention to whether your work exists because it creates value directly, or because it solves a problem that technology might soon solve differently. The first is more durable than the second. The distinction is worth knowing.
I notice that the word "eliminate" appears 14 times in this article. I could have varied the language more. I chose not to, because some words deserve to be repeated until they stop being comfortable.
— Ish.