Business
Centaur vs Cyborg
Also known as: human-AI collaboration styles
Two styles of working with AI, "centaur" divides tasks cleanly between human and AI, "cyborg" blends them tightly, switching back and forth moment to moment.
Centaur and cyborg are two ways of teaming up with AI. A centaur splits the work: "I'll do this part, AI does that part", clear division of labour. A cyborg blends with the AI, weaving back and forth constantly, letting it draft a sentence, tweaking it, asking again, in a tight loop. Both can work well; the best approach depends on the task and on knowing where AI is strong and where it is not.
The centaur/cyborg distinction (named after the mythical half-human figures) describes two modes of human-AI collaboration: - Centaur: strategic division, hand clearly separable subtasks to whoever (human or AI) does them best. - Cyborg: deep integration, interleave human and AI continuously within a task, in a rapid back-and-forth.
Why it is useful
It gives practical shape to co-intelligence. The key skill in both is judging AI's uneven abilities, the "jagged frontier" where it is unexpectedly strong on some tasks and weak on adjacent ones, and routing work accordingly.
The takeaway
Neither mode is universally best. Effective collaborators develop the AI literacy to switch fluidly, keeping human judgement and verification in the loop.
"Centaurs" and "cyborgs" name two human-AI teaming strategies popularised in studies of knowledge work: centaurs partition tasks along a human/AI boundary based on comparative advantage, while cyborgs interleave contributions at fine granularity within tasks. The framing operationalises co-intelligence and connects to empirical findings on the "jagged frontier": AI performance varies sharply and non-intuitively across nearby tasks, so naive reliance can reduce quality on tasks beyond the frontier while boosting it within (Dell'Acqua et al. 2023).
The practical implication is that value depends less on the mode label than on calibrated task routing and verification, i.e. knowing where the model is reliable, which requires AI literacy and oversight. It informs workflow design in the augmentation-vs-automation space: both centaur and cyborg patterns are augmentation strategies, distinguished by integration granularity, and effective teams adapt the pattern to task structure and to real-time judgements about model competence.