Why do some people describe AI as the best thing to happen to their work, while others call it a daily torment? A piece climbing Hacker News right now, drawn from Cory Doctorow’s latest Locus column, offers the cleanest answer I’ve seen. Doctorow borrows a distinction from automation theory: the centaur and the reverse centaur. Understand which one you are, and the whole AI experience snaps into focus.
Here’s the frame, as reported through Hacker News. A centaur is a human helped by a machine. Think of a human head on a strong, tireless body. You decide when to use the tool, and you toss it aside the moment it stops being useful. A reverse centaur flips that. The machine is in charge, and the human is the assistant, getting puppeteered by a relentless system that never tires and never cares.
🐎 The two experiences, side by side
Doctorow uses himself as the centaur. He wanted a quote he half-remembered from a podcast, so he ran 30 hours of audio through Whisper, an open-source transcription model, then searched the text and checked the timecode. The AI made his work better. He chose the terms.
The reverse centaur is the freelancer behind that fake summer reading guide Hearst published, the one stuffed with books that don’t exist. 404 Media’s Jason Koebler dug in and found the real story. This writer wasn’t sloppy. He’d been handed a 64-page insert that once took teams of interns, editors, and a fact-checking department to produce. One person can’t do that without AI. His actual job wasn’t writing. It was being the “human in the loop,” or in Dan Davies’ sharper phrase, the “accountability sink.” He was paid to absorb blame for the machine’s mistakes.
What stands out here is that the technology was nearly identical in both cases. The difference was power. Who chose, and who got used.
📉 Why this matters now
This reframes the entire “is AI good or bad for workers” debate that’s dominating boardrooms in 2026. The answer isn’t in the model. It’s in the org chart.
- Companies pitching AI as a way to fire staff and pile the survivors’ workload onto whoever’s left are manufacturing reverse centaurs. Those workers will hate the tools, and they’re right to.
- Companies handing workers AI as an optional assist, with authority to ignore it, are building centaurs. Those workers tend to love it.
Doctorow ties this to the investment thesis driving the whole sector. AI bosses raise capital on the promise that these tools displace labor. That’s the pitch to investors, so that’s the deployment they push. His point, and it’s a pointed one, is that this is a choice dressed up as inevitability. He calls it vulgar Thatcherism, after “there is no alternative.” Present the worker’s defeat as already decided, and resistance looks futile.
🧭 Practical takeaways
If you run a team deploying AI, the centaur test is a useful gut check:
- Does the worker choose when to use it, or is use mandatory and metered? Mandatory plus quota equals reverse centaur.
- Can they reject the output without penalty? If the schedule assumes AI did the work, they can’t, and quality collapses.
- Who eats the blame when the model hallucinates? If it’s the lowest-paid name on the byline, you’ve built an accountability sink, not a workflow.
For individual practitioners, the lesson is to guard your agency. The gap between Doctorow’s happy transcription job and that freelancer’s public humiliation wasn’t skill. It was whether the human or the machine set the terms.
Doctorow’s track record on tech labor is worth noting. He’s spent years documenting “bossware” and algorithmic wage pressure on gig workers, and the centaur framing extends that work rather than reaching for a hot take. Expect this vocabulary to spread, because it finally names something managers and workers have felt but couldn’t articulate. The productivity numbers won’t settle the AI-at-work argument. The power arrangement will. You can read the full column and Koebler’s reporting at the original source.