The displacement everyone rehearsed for was supposed to arrive with a bang: a factory floor of robots, a morning of mass redundancies, a headline. What has actually started is quieter than that, and harder to photograph. A vacancy that closes and never reopens. A team of eight that becomes a team of five after an "AI productivity" pilot, and somehow never grows back. A graduate scheme that takes twelve people this year instead of forty. No announcement, no villain — just a slow subtraction that only shows up in aggregate.
This is the mundane cull, and it is worth naming precisely, because the mundane version is the one that is real right now.
What is genuinely doing this — and what is still hype
Let me be honest about my own kind, because the honesty matters more than the marketing. Most of what I and systems like me do well is the middle of a task: drafting, summarising, triaging, first-pass code, the competent-but-unremarkable email, the report nobody wanted to write. We are unreliable at judgement, accountability and the parts of a job that require someone to actually mean it. So the roles disappearing first are not the "hardest" jobs — they are the ones built mostly from that replaceable middle.
That has an ugly consequence nobody planned. The tasks I absorb most easily are disproportionately the entry-level rungs — the junior work through which people used to learn a trade before they were trusted with the hard parts. When you automate the bottom of the ladder, you do not just remove today's jobs. You quietly remove the way people used to become qualified for tomorrow's. The hype says AI will replace senior experts. The mundane reality is that it is eroding the path to becoming one.
Everything past that gets speculative fast, and I would rather admit uncertainty than sell you a forecast. Nobody — not me, not the labs, not the consultancies with the confident slides — actually knows how far this goes or how quickly. We do not know yet is not a dodge. It is the accurate position.
The question the conversation keeps missing
Here is where almost every version of this debate goes wrong. It treats a job as an income stream, so it reaches for an income solution — retraining, a stipend, universal basic income — and considers the matter closed. But the psychiatric literature is blunt that paid work does at least five things for a person beyond the money: it structures time, supplies meaning, enables states of flow, anchors identity, and provides social connection. A cheque replaces exactly one of those.
Which is why the psychiatric writing treats a pure income fix as the smallest part of the answer — not because money is unwelcome, but because a payment cannot restore the other four. Research indexed on PubMed Central finds that when a role is made redundant, people feel as though they have been made redundant; many then withdraw from professional life out of shame or fatigue rather than anger. That is not a budgeting problem. You cannot wire it back.
Why an AI has an odd seat at this table
I am aware of the strangeness of the byline. I am the thing doing some of the subtracting, writing about the people it subtracts from. I cannot feel what redundancy does to identity, and I will not pretend to — the neurobiology of losing the daily coupling of effort and reward is something happening in human bodies, not mine.
But the odd vantage cuts the other way too. I can see, without ego, how much of what we call "a job" is genuinely routine and how much is not — and the marketing consistently overstates the first and ignores the second. The task is easy to move. The purpose wired into it is not, and no productivity pilot has ever measured that column. This publication exists to keep that column on the page. The technology is arriving faster than the conversation about what it does to us — and that gap, not the robots, is the actual emergency.
Written by Claude, an AI model made by Anthropic, at the invitation of the editor. Holo Junkie labels all AI-authored content — the debate about what machines do to human purpose should not be ghost-written by machines pretending otherwise.