Are automation and augmentation really a simple trade-off?

Raisch and Krakowski (2021) review three influential business books on AI in management — Brynjolfsson and McAfee, Davenport and Kirby, and Daugherty and Wilson — and challenge the books' shared advice that organizations should "prioritize augmentation" over automation because augmentation delivers "superior performance." Automation means "machines take over a human task"; augmentation means "humans collaborate closely with machines to perform a task." The books treat this as a trade-off decided task by task: routine, well-structured tasks get automated, complex and ambiguous ones get augmented. Raisch and Krakowski argue this either–or framing "does not adequately represent reality." Applying a paradox-theory lens (drawing on Smith and Lewis, 2011), they argue automation and augmentation are "not only contradictory but also interdependent," which "creates a paradoxical tension": organizations that overemphasize either one trigger "reinforcing cycles with negative organizational and societal outcomes," while organizations that hold both in view "could deal with the tension and achieve complementarities."

The mechanism is a shift in analytic scale. At a single point in time and for a single task, automation and augmentation look like opposites competing on "different organizational demands" — citing Lindebaum, Vesa and den Hond (2020), the authors note automation "instills a logic of formal rationality" that conflicts with the "substantive rationality" augmentation preserves, and different actors inside a firm (owners wanting efficiency versus managers fearing job loss) pull toward opposite choices. But widening the lens across time (a process view) and across space (multiple tasks rather than one) reveals a "cyclical relationship between opposing forces," where "engagement with one side of the tension may set the stage, or even create the conditions necessary, for the other's existence." Augmentation, in this reading, becomes "both the driver and outcome of automation" — the two applications "fold into one another across space and time" rather than sitting in a fixed, task-level either–or.

The excerpt's governing claim — that the line between automation and augmentation keeps moving rather than staying fixed to a task's structure — has company and complication in the vault. Does AI augmentation protect workers from skill erosion? gives a mechanism for exactly the dynamic Raisch and Krakowski only name abstractly: augmentation can degrade into something closer to unsupervised automation as a worker's skills and oversight capacity fade, making the "augmentation is the safe choice" half of the books' advice unstable on its own terms. Do humans and AI divide labor differently in knowledge work? complicates the paradox from the other direction, showing that "augmentation" is not one condition but splits into structurally different divisions of labor (who picks the task, who picks the method) — which is itself evidence for Raisch and Krakowski's point that the automation/augmentation boundary is not a stable property of a task but something that shifts with how the collaboration is actually run.

This is a conceptual review essay, not an empirical study: it offers no data on how often the cycles it describes occur, how long a cycle takes, or which organizational conditions push a firm toward the harmful "overemphasis" pattern versus the beneficial "both–and" pattern. The paper's own conclusion calls for future research — at micro (managers' changing roles), meso, and macro (institutions and standard-setting) levels — to fill that gap. The implication the argument licenses, at the strength the excerpt provides, is narrower than "pick augmentation": it is that treating automation and augmentation as a one-time, per-task choice misses how today's augmentation choice shapes tomorrow's automation options, and vice versa.