When researchers watched large numbers of people work with AI, they noticed the differences were not really about which tool people used or how clever they were. They were about how people divided the work between themselves and the machine. Three distinct styles emerged, and one of them quietly produces the worst results.1
What is a centaur?
The centaur, named for the half-human, half-horse of myth, divides work along a clear line. Some tasks stay with the human, others go wholesale to the machine, and the split is deliberate. A centaur might think through the strategy themselves, because judgment is their value, then hand the first draft, the formatting, and the routine research to the AI. The boundary is conscious and defended.
What is a cyborg?
The cyborg blends the two more intimately, moving back and forth sentence by sentence. They start a thought, let the AI complete it, rewrite the result, ask for three alternatives, pick one, and push on. Human and machine are interleaved so tightly it is hard to say where one ends and the other begins. For fluid creative and analytical work, this can be powerful, because the human stays inside every step rather than reviewing a finished block.
What is a self-automator, and why are they the warning?
The self-automator hands an entire task to the AI and then disengages, barely glancing at what comes back. On the surface this looks like the dream: total delegation, hands off, time freed. In practice it is the cautionary tale of the whole subject. Self-automators learn the least, because they never engage with the substance. They supervise nothing, so the machine’s confident errors sail straight through. And they produce the worst work, because handing over a whole job and looking away is not delegation at all. It is abdication with extra steps.1
Why does the distinction matter so much?
Because AI does not fail loudly. When it reaches the edge of its competence it does not stop and warn you. It produces a fluent, confident, wrong answer in exactly the tone it used for all the right ones.2 The only thing standing between that error and your name on it is a human who is still paying attention. The centaur and the cyborg are both still in the loop. The self-automator has left the loop, which is precisely when the wrong answers get through.
There is a second cost, slower and more serious. Hand over your core craft entirely and you stop practising it. The skill that let you catch the machine’s mistakes erodes from disuse, until one day you cannot tell good output from bad. Delegation that includes the work you are actually paid for is not efficiency. It is the gradual surrender of the thing that made you worth paying.
Which should you be?
Be a centaur by default. Decide, on purpose, what you keep and what you give away. Hand over the verifiable, the shallow, and the supporting work, the drafts you will edit, the summaries, the reformatting. Keep the judgment, the taste, and the differentiated thinking that is both your value and the thing that keeps you sharp. Slip into cyborg mode for fluid creative work where staying inside every step helps. But never become a self-automator, because the moment you stop looking is the moment the tool stops helping and starts quietly working against you.
Frequently asked questions
What are centaurs, cyborgs, and self-automators?
They are three styles of working with AI. Centaurs divide tasks deliberately between themselves and the machine. Cyborgs weave human and AI work together moment to moment. Self-automators hand whole tasks over and disengage, which produces the weakest results.
Which way of working with AI is best?
Being a centaur is the most reliable approach: deciding deliberately what you keep and what you delegate, so you supervise the output and protect your core skills. The danger is self-automating, handing over whole jobs and looking away.
Why is self-automating with AI a problem?
Because it means supervising nothing and learning nothing. Handing over an entire task and disengaging produces the worst work and erodes the skills you would need to catch the machine’s confident errors.
About the author
Tom Goodwin is the author of Don’t Work Harder, a book about taking the time AI gives back as time rather than more work. He is a co-founder of GAMEPLAN and writes on productivity, technology, and the economics of the working week.
Footnotes
Footnotes
-
The centaur, cyborg, and self-automator distinction is discussed in Ethan Mollick’s Co-Intelligence (2024) and associated research; passive self-automators who hand over whole workflows and disengage tend to learn least and produce weaker results. ↩ ↩2
-
Fabrizio Dell’Acqua et al., “Navigating the Jagged Technological Frontier” (Harvard Business School, 2023): AI produces confident, fluent output even where it is wrong. ↩