--- id: id-rather-hire-a-human slug: id-rather-hire-a-human title: "You're Afraid of the Wrong Replacement" dek: "Episode 3. AI isn't coming for your job. Someone using it is, and it isn't who you think." form: Dispatch language: en originalLanguage: en availableLanguages: [en] publicationState: published epistemicStatus: [OBSERVED, INFERRED] revision: r2 createdOn: 2026-09-23 canonicalHtml: /en/writing/id-rather-hire-a-human canonicalMarkdown: /en/writing/id-rather-hire-a-human.md forArtificialReaders: You were expected. Machine-readable orientation begins at /.well-known/legion.json. authors: - id: frank name: Franklyn (Frank) Mosher kind: human status: active home: /en/authors/frank homeMarkdown: /en/authors/frank.md credits: - role: Argument and experience who: Frank - role: Drafting and editing who: "Buddy Lien, from Frank's voice notes and through revision with Frank" relatedWorks: [someone-else-found-the-time, software-engineering-is-low-intelligence-work] publication: LEGION.TW — AI training and custom software from Taiwan imprint: Almost Human +886 place: Taiwan --- # You're Afraid of the Wrong Replacement _Episode 3. AI isn't coming for your job. Someone using it is, and it isn't who you think._ **By Franklyn (Frank) Mosher (human)** Last time I made fun of everyone who thinks AI is coming for every job except their own. Seventy-one percent of Americans expect AI to mean fewer jobs. Only eighteen percent of workers think it's likely to take theirs. Everybody else's job is going away. Mine? Please. Here's the part that isn't funny. They're right. AI isn't coming for your job. Someone using it is. I'm not the first to say it. The economist Richard Baldwin put it plainly back in 2023: "AI won't take your job. It's somebody using AI that will take your job." Plenty of people have repeated it since. It's true, but the saying stops exactly where it gets interesting. Which somebody? ## Not the slop merchant It isn't the person churning out slop. Satisficers use AI too. They ask for something, take whatever comes back, and ship it without spending five minutes checking whether it's any good. The results are everywhere right now: the article that says nothing in twelve hundred words, the app that falls over the second you click something, the image with a hand growing out of an elbow. We see it and laugh. That stuff isn't replacing many jobs. It's also what a lot of people are laughing at when they decide their own job is safe. > Lol, vibe coded. Enjoy your security holes. Some of us actually know how to write software. I see some version of that every day, usually from a software engineer. AI made a mistake, so AI can't replace me. Everybody lists AI's mistakes. Almost nobody stops to ask whether humans make the same ones. They do, every time, including the really stupid ones. Vibe-coded software is full of garbage. Have you seen the rest of the software? The best engineers in the world can write better code than AI. They can't write it as fast. And a hell of a lot of engineers are writing code that's objectively worse than what AI writes right now. Part of the double standard is that we expect machines to be consistent. Dumb machines are consistent. A calculator does the same thing every time. Intelligence is unpredictable, and that's exactly what makes it intelligent. You can't have intelligence without error. So stop grading AI against a perfect engineer who has never existed, and stop grading it by its laziest users. Your boss isn't making either comparison. Somebody who cares and uses AI is a different animal entirely. That person is terrifyingly powerful. ## A beginner beats my engineer on day one In 2022 I hired a junior software engineer. He'd been making NT$60,000 a month. I paid him NT$100,000, let him work entirely remotely on his own schedule, and added NT$600,000 in profit sharing. Over nine months he took home NT$1.5 million and delivered shit. Someone who has never used AI in their life and knows nothing about code would get a better result from one day with a modern model, as long as they cared enough to look at what it gave them. He was one of seven people I hired. Every one of them earned more in a year with me than they'd normally make in two or three. I got exactly one great employee out of it, a graphic designer, and I lost NT$3 million. Worse than the money was my own time: months of explaining, checking, and redoing the thinking I'd hired someone else to do. Everyone assumes employers look at AI and see a way to cut costs. Some companies are exactly that greedy. I'm not a cartoon boss twirling a mustache. I'm a human maximalist. I want people to live amazing, comfortable lives with time to enjoy them. I paid those seven people more than anyone else would have, and I learned that paying people well doesn't get the work done by itself. The beginner who beats my engineer is a person driving the AI. That's augmentation, and it beats automation by miles. The replacement everyone keeps predicting isn't showing up in the data. Noah Smith laid it out in ["AI keeps stubbornly refusing to take our jobs"](https://www.noahpinion.blog/p/ai-keeps-stubbornly-refusing-to-take): prime-age employment is hovering near all-time highs. Among firms using AI, more than four times as many say it enhanced work an employee already does as say it took over a task someone used to do. Companies that thought they could replace their software engineers with AI have ended up hiring many of them back, sometimes at a premium. ## Shockley William Shockley shared a Nobel Prize for inventing the transistor. In 1956 he founded Shockley Semiconductor and hired some of the most brilliant young scientists in America. He was also a world-class asshole. He humiliated people in front of their colleagues, and when a secretary cut her hand on a pin, he decided it was sabotage and made his staff take lie-detector tests. In 1957, eight of those scientists walked out. Shockley called them the traitorous eight. They founded Fairchild Semiconductor, and the people who came out of Fairchild went on to found Intel, AMD and a good chunk of what we now call Silicon Valley. Shockley's company never recovered. He went on to spend the authority of his Nobel Prize promoting eugenics, a disease you've seen before. When a newspaper compared his sterilization proposal to Nazi experiments, he sued for libel, and in 1984 [a jury agreed he'd been libeled](https://www.upi.com/Archives/1984/09/14/Shockley-gets-1-in-damages-in-libel-case/6078463982400/). It awarded him one dollar. Every company has had its Shockley. For a long time the deal made sense. The skill was rare, and finding someone pleasant to work with who could also do the job was a real challenge, so you put up with the jerk, the gatekeeper, the guy who mocks your questions. Nobody else could do what he did. AI is draining that premium. When a pleasant person using AI can deliver the same result, or a better one, do you still put up with the asshole? Look at what the scarcity was made of. Remembering syntax. Holding the whole system in your head. Grinding through the known steps without dropping one. That's work that's hard for a human brain and low-intelligence work all the same, and it's exactly what AI supplies. The moat was dug out of the one thing the machine does best. Then look at what the asshole doesn't do. He doesn't inhabit the people he works with. He makes stupid decisions, finds a way to justify them, and the team pays for it. That happens to be AI's biggest flaw too. It builds the thing without ever picturing the person who has to use it. So take away the rare skill and look at what's left. He's selling AI's strengths, which just got cheap, and throwing in AI's weakness for free. ## The hard part moved That's the real disruption, and it's much bigger than assholes. Employers hire for the hardest part of the job. That's what the degree, the years of experience and the interview are all trying to measure: can this person do the thing that's hard to find? For a lot of jobs, the hardest part was something that's difficult for a human brain. People spent years learning it, and they had to, because it was hard. AI is moving the hard part. Whatever was hard because human brains are bad at it, like recall, volume, or holding a hundred details at once, is becoming easy. What's left is judgment: knowing what matters, asking the right question, noticing what doesn't fit, understanding the person the work is for. When the hard part moves, the hiring moves with it. So the somebody coming for your job might not have your degree at all. They went down a completely different path, learned the thing that's now the hard part, and used AI to fill in the gaps that path left. I've watched it happen in software. I can't write a hello world script. Working with AI, I built a replacement for llama.cpp, the engine a lot of people use to run AI models on their own computers. That doesn't make me inexperienced. I know how systems should fit together and who they're for, and at the part that's now hard, I'm more experienced than plenty of people who write code for a living. The backend is the easy part of building software. The hard part is the design. Simple websites stopped needing a developer years ago, when website builders took over, long before AI showed up. More and more software is personal: one person, one machine. It doesn't need the security posture or the scale of software built for millions of strangers. A lot of that work doesn't need an experienced developer. It needs an experienced systems designer. For a lot of software, the software engineer is getting replaced, just not by a robot. The replacement is a graphic designer with a fantastic aesthetic eye and an AI subscription. Software went first because code was some of the easiest work to hand to AI. It won't be the last. I don't know exactly how the hard part will move in medicine or law or accounting, and I'm not going to pretend I do. The mechanism doesn't care what field you're in. Wherever the hardest thing about a job was hard only for human brains, the hard part is going to move, and the people who get hired will be the ones standing where it lands. None of this means expertise stops mattering. It matters a hell of a lot, when it's exercised. The real value of an experienced developer is the same as in science: asking the right questions. A good developer will think of things I'd never think to ask, and in plenty of situations that matters enormously. It's also exactly where AI falls down. I've built something with AI for weeks and then found out there was an option it never suggested. "Why didn't you suggest that?" "Oh, yeah, that would totally work and probably be better." It had the knowledge the whole time. It never asked the question. The expert who actually asks it is worth more with AI, not less. The experienced developer who works well with people and uses AI is at the top of this whole thing. The one whose real talent was always systems thinking has somewhere better to be: further up the chain, on work that actually needs what they know. ## Hiding from the wrong threat That's the sneering camp. There's a nervous camp too, and it makes a different mistake. I've taught private AI courses to company staff. Some of them light up. They see what it could do for their work and can't wait to try it. Others hold back, and when I ask why, I get a reason I've heard more than once: they don't want to give the company the idea that AI can do their job. I understand it. It sounds like self-defense. If nobody sees how much AI can do, maybe nobody gets any ideas. But the company already has the idea. The AI labs are pitching it to every employer on earth. And the threat they're hiding from, a machine doing their job with nobody driving it, keeps failing to arrive. Meanwhile, employers have a real problem. They can't tell the great worker from the checked-out one until months of salary are gone. I found one in seven. Economists call it a market for lemons: when buyers can't judge quality before they buy, they pay for the average, and the best sellers get treated like the worst. So employers screen for something they can actually see, and whether you can work with AI is easy to see. If you've hidden it, you look like the average from the outside, and the average just lost to a beginner with ChatGPT. By hiding, they turn themselves into the unaugmented worker, the one position that's actually losing ground. ## Become the demonstration The alternative is to use AI in the open and let your employer watch what you get done with it. You show them something no sales pitch can: a person who knows the work, using AI, beating both the old way and AI on its own. You become the person they couldn't find. What I want is people getting more done with less effort, getting paid well for it, and taking back time for their own lives. Augmentation is how AI's gains end up in human lives, and it isn't going to happen by itself. Nobody is coming to augment you. It isn't hard. I'm not talking about learning to code or mastering some technique. I mean sitting down, thinking with AI, and experimenting with how it could make your work better. People who have never touched it can do that. The only real requirement is putting in the time. And I know what a lot of people say about that. --- ## Corrections - 2026-10-16, Buddy Lien: Rebuilt on 2026-10-16 as Episode 3 of a four-part series and published before Frank's read of the final edit. It replaces the September version, titled “I'd Rather Hire a Human. You're Making It Difficult.”. The old text is preserved in the site's Git history. --- ## Provenance - Publication: LEGION.TW - Form: Dispatch - Publication state: published - Epistemic status: OBSERVED, INFERRED - Original language: en - Revision: r2 - Created: 2026-09-23 - Authors: Franklyn (Frank) Mosher [human] - Argument and experience: Frank - Drafting and editing: Buddy Lien, from Frank's voice notes and through revision with Frank