Rewiring Work vs. Automating It

The two ways to apply a new technology

Confronted with a capable new technology, an organization has two distinct moves available. It can automate: take the existing sequence of tasks and have the machine do them faster or cheaper, leaving the structure of the work intact. Or it can redesign: rethink what the work is for, recombine the tasks, and rebuild the process around what the technology now makes possible. The two look superficially similar — both involve “adopting AI” — but they sit at opposite ends of the value distribution. The central claim of this paper, supported across a primary management literature, a macroeconomics literature, and a recent body of generative-AI field evidence, is that the first move captures a small, bounded gain and the second captures the large, compounding one.

The failure mode is older than AI. Bill Gates compressed it into two rules: “automation applied to an efficient operation will magnify the efficiency,” and “automation applied to an inefficient operation will magnify the inefficiency”¹⁶. The mechanism is unforgiving. A process you automate without first improving is a process whose flaws you have just hard-coded and scaled. Hammer & Champy put the same point more bluntly in their 1993 sequel: “Automating a mess yields an automated mess”¹. This is the credible articulation of “automate your existing workflow and you scale your existing problem” — and it is attributable to named authors and a real published literature, not to anonymous folklore. (See the verification note below on the precise book-and-page provenance of the Gates formulation.)

Hammer 1990: stop paving the cow paths

The foundational text is Michael Hammer’s 1990 Harvard Business Review article, “Reengineering Work: Don’t Automate, Obliterate.” Hammer’s diagnosis was that companies had spent a decade using information technology to mechanize old ways of doing business — leaving inefficient processes intact and using computers simply to speed them up¹. His prescription was the inverse: “use the power of modern information technology to radically redesign our business processes in order to achieve dramatic improvements in their performance”¹. The metaphor that carried the argument is the one still quoted today:

Hammer was explicit that this was not incrementalism. “Reengineering cannot be planned meticulously and accomplished in small and cautious steps. It’s an all-or-nothing proposition with an uncertain result”¹. The canonical illustration from the article — Ford’s accounts-payable function — made the magnitude concrete: by redesigning the procurement process around a shared database and “invoiceless” payment rather than automating the existing invoice-matching task, headcount in that function was cut dramatically.

Davenport & Short published the complementary founding text the same year. “The New Industrial Engineering” argued that “business process design and information technology are natural partners,” a relationship industrial engineers “have never fully exploited”⁹. Drawing on field studies of nineteen companies, they observed that organizations which “used IT to redesign boundary-crossing, customer-driven processes have benefited enormously,” while IT implementation often failed precisely because it automated flawed processes rather than reimagining them⁹. Davenport’s later work, Process Innovation (1993), extended the discipline; thirty-two years on, he returned to it for the AI era — more on that below.

The dynamo precedent: why the payoff lags by decades

The reason redesign beats automation is not a management aphorism; it is an economic regularity observed across general-purpose technologies. Paul David’s 1990 study “The Dynamo and the Computer” is the canonical precedent. Factory electrification produced almost no measured productivity gain for roughly two to three decades after Edison’s first generating stations (1881) — by 1900, motors powered less than about 5% of factory mechanical drive¹⁰. The surge arrived in the 1920s, and only once factories abandoned the centralized steam-era layout of shafts and belts and adopted the “unit drive” — an individual electric motor on each machine — which let plants be laid out around the workflow rather than around a central power source. David’s explanation for the lag is precisely the cost of replacing capital and work organization “adapted to the old regime”¹⁰. Swapping a steam engine for a big electric one (automating) bought little. Redesigning the factory around what electricity now permitted bought the revolution.

Brynjolfsson, Rock & Syverson formalized this for the AI era. Their 2017 paper on the “modern productivity paradox” — AI capabilities soaring while measured productivity growth had halved — weighed four explanations (false hopes, mismeasurement, redistribution, implementation lags) and concluded that lags are “likely the biggest contributor”³. The mechanism: like other general-purpose technologies, AI’s “full effects won’t be realized until waves of complementary innovations are developed and implemented,” and “the required adjustment costs, organizational changes, and new skills can be modeled as a kind of intangible capital”³. Their 2018 follow-up gave the pattern its name and shape — the Productivity J-Curve: because the complementary intangible investment (reorganized processes, retrained people, new business models) is poorly captured in the accounts, measured productivity is understated early, while firms are investing, and overstated later, when those investments are harvested⁷. The dip is the redesign cost. The upswing is the redesign payoff.

Generative AI reproduces the pattern — at speed

The most striking feature of the current moment is that the generative-AI evidence is reproducing the same finding the management and economics literatures established, only on a compressed timescale. Adoption has moved faster than any prior enterprise technology: Stanford’s 2025 AI Index reports organizational AI use jumped to 78% in 2024 from 55% in 2023, and the share of organizations using generative AI in at least one business function more than doubled, from 33% to 71%¹⁴. Yet bottom-line impact has badly lagged adoption — McKinsey finds just 39% of organizations report EBIT impact at the enterprise level⁵, and only about 11% of companies worldwide are using generative AI at scale¹³. The adoption-to-value gap is the J-curve, live.

The single most-cited industry datapoint is BCG’s 10-20-70 rule: of the effort required to capture value from AI, roughly 10% goes to algorithms, 20% to the underlying technology and data, and 70% to people and processes⁸. The model — the thing most executives instinctively treat as the project — is explicitly the smallest slice. BCG’s complementary framing sorts AI plays into three tiers: Deploy (off-the-shelf tools that streamline everyday tasks, “boost workforce productivity by 10–15%”); Reshape (“re-imagination of functions through workflow re-engineering,” driving “30–50% improvements in efficiency, effectiveness across affected functions”); and Invent (AI-native offerings and new business models — a “new revenue play”)⁸. The gap between Deploy and Reshape — between inserting the tool and redesigning the work — is roughly a three-to-five-fold difference in function-level impact. BCG states the conclusion plainly: “this is a people transformation not a tech transformation”⁸.

McKinsey’s survey evidence triangulates the same conclusion from the demand side. In its March 2025 “State of AI: How organizations are rewiring to capture value,” McKinsey reports that “out of 25 attributes tested … the redesign of workflows has the biggest effect on an organization’s ability to see EBIT impact from its use of gen AI”⁵. The roughly 6% of respondents who qualify as AI high performers — attributing at least 5% of EBIT to AI and reporting significant value — are “nearly 3x more likely to have fundamentally redesigned workflows as part of their AI efforts”⁵. The word in McKinsey’s own title — rewiring — is the thesis. And the value-sizing report that anchored the whole investment wave carried the same caveat in its fine print: McKinsey Global Institute’s estimate that generative AI could add $2.6–4.4 trillion annually across 63 use cases, with about 75% of that value concentrated in four functions, is explicitly contingent on leaders “re-examining and redesigning core business processes” — not on deploying tools¹².

Augment, don’t just automate: Brynjolfsson’s Turing Trap

If Hammer supplies the “redesign, don’t automate” half of the argument, Erik Brynjolfsson supplies the “augment, don’t imitate” half. His 2022 essay “The Turing Trap” draws the distinction precisely: automation systems “substitute for human labor,” while augmentation systems are “focused on augmenting humans rather than mimicking them”². The trap is that the entire field has, since Turing, implicitly aimed at human-imitating AI — and that this target carries two costs. First, a distributional one: “as machines become better substitutes for human labor, workers lose economic and political bargaining power and become increasingly dependent on those who control the technology,” whereas augmentation lets “humans retain the power to insist on a share of the value created”². Second, and decisive for the strategy question, a value one:

The augmentation thesis is not merely normative — it has field-experimental support. Brynjolfsson, Li & Raymond’s study of 5,179 customer-support agents at a Fortune 500 firm found a generative-AI assistant raised productivity (issues resolved per hour) by 14% on average — but with sharp heterogeneity: a 34% gain for novice and low-skilled workers and “minimal impact on experienced and highly skilled workers”⁴. The mechanism is the redesign of the work, not the tool itself: the AI “disseminates the best practices of more able workers and helps newer workers move down the experience curve”⁴. In other words, the value came from codifying the tacit knowledge of high performers into the workflow available to everyone — a reorganization of how expertise flows through the team. The same study found improved customer sentiment and increased employee retention⁴.

The jagged frontier: redesign the human–AI division of labor

The augmentation argument has a sharp operational corollary: AI capability is “jagged,” strong on some tasks and abruptly weak on adjacent ones, so the gains depend on matching AI to the right tasks and redesigning how work is allocated — not on blanket insertion. The largest field experiment on this — Dell’Acqua, Mollick and colleagues’ study with 758 BCG consultants — quantifies both edges of the frontier. On tasks inside the frontier (work AI is good at), consultants using AI completed 12.2% more tasks, roughly 25% faster, at roughly 40% higher quality than the control group⁶. On tasks outside the frontier (beyond current AI capability), consultants using AI were 19 percentage points more likely to produce incorrect answers⁶. The same tool, the same people — opposite effects, depending entirely on whether the work was matched to the capability.

What separated effective users was not the tool but the pattern of work they designed around it. The study identifies two successful patterns: the Centaur (a human-led, strategic division of labor, switching between AI and one’s own work by task) and the Cyborg (deep, fine-grained integration of human and AI effort throughout the task)⁶. Both are redesigned divisions of labor. Neither is “give everyone the tool and hope.” The lesson generalizes the Hammer point into the AI era: value is a function of how you re-architect the work, and naive tool insertion can make output strictly worse on the tasks where the tool is weak.

The OECD’s vacancy-level evidence shows what the reorganization looks like inside a job rather than across a project. Its 2024 study of online vacancies across ten OECD countries gives a concrete mechanism: an insurance company adopts an AI tool that flags which customers are likely to escalate a service issue, and “the job of a sales agent changed to emphasise greater customer interaction with less time needed to analyse customer files”¹⁹. The work was reorganized toward the human-strong activity, not merely accelerated. On the broader question of where skill demand is heading, the OECD’s evidence is genuinely mixed and should not be over-read. Across AI-exposed occupations, demand for business and management skills rose about 8% and for emotional, digital and social skills about 15% over the period studied — but the OECD is explicit that this rise is not isolable to AI (“factors other than just AI may be driving these changing skills demands, such as the general trend towards increased digitalisation”), since the same increases appear in less-exposed occupations¹⁹. When the OECD instead isolates the AI effect at the workplace level, it finds the opposite for those groups: demand for management, business and digital skills fell by over three percentage points in workplaces becoming more exposed to AI — a decline the authors call “relatively small” but worth monitoring as adoption grows¹⁹. The reorganization mechanism in the insurance example is the cleaner support for the augmentation reading; the aggregate skill-demand direction is contested in the source itself.

Source: Dell’Acqua, McFowland, Mollick et al., “Navigating the Jagged Technological Frontier,” HBS/BCG WP 24-013, 2023.

The sequence: experiment → adopt → absorb / redesign

If redesign is where the value sits, why do so many organizations stop at automation? Because redesign is the last and hardest stage of a sequence, and most efforts stall before reaching it. The economics make the stages explicit. Adopting a general-purpose technology is cheap and fast; the productive reorganization around it is expensive, slow, and intangible — the J-curve dip⁷. The field data show the stall directly: 78% of organizations now use AI¹⁴, but only about 11% use generative AI at scale¹³ and only 39% report enterprise-level EBIT impact⁵. Adoption is nearly universal; absorption is rare.

What gets the few across is process discipline, not more tooling. McKinsey’s operations research finds that “a disciplined, stage-gated review process with clear go/no-go criteria separates the merely promising deployments from the ones most likely to be productive,” and that two-thirds of respondents set a three-to-five-year timeline for realizing full value¹³ — the lag, again. The figure below renders the sequence as a sourced framework: each stage’s cost, payoff, and the share of organizations observed to reach it.

The moving bottleneck: where redesign should aim

Redesign is not redecoration; it needs a target. The discipline that supplies the targeting logic is Eliyahu Goldratt’s Theory of Constraints: a system’s throughput is governed by its single binding constraint, so improvement effort anywhere other than the constraint yields little, and once a constraint is broken the bottleneck moves. The implication for AI is exact. Accelerating a task that is not the constraint — automating a step that was never the bottleneck — produces no system-level gain; it merely builds inventory in front of the next bottleneck. This is the structural reason tool insertion so often disappoints: a generative-AI assistant can make one step 40% faster and leave end-to-end throughput unchanged, because the binding constraint sat elsewhere. Redesign works because it re-architects the whole flow and relocates effort to the constraint, rather than optimizing a local step.

The counterargument: “so-so automation” and why redesign matters more, not less

The case for redesign is sharpened, not weakened, by its strongest economic counterargument. Acemoglu & Restrepo’s 2019 work decomposes automation’s effect on labor demand into a productivity effect and a displacement effect, and shows the displacement effect “always reduces the labor share”¹¹. Their counter-intuitive conclusion: the real danger to employment and wages is not “brilliant” automation but “so-so technologies that generate small productivity improvements” — automation just good enough to be adopted, but not good enough for its productivity gain to offset the labor it displaces¹¹. Their named example is “automated customer service, which has displaced human service representatives but is generally deemed to be low quality”¹¹ — and they note this category “may also include several of the applications of artificial intelligence technology to tasks that are currently challenging for machines”¹¹.

This is the precise economic statement of “automate the existing workflow and you scale the existing problem.” So-so automation is what you get when you take the automation path without redesign: you displace the labor, hard-code the flawed process, and capture too little productivity to justify it. Acemoglu & Restrepo’s warning is that “further automation, especially when it is induced by … excessive enthusiasm about automating everything, would take the form of such so-so technologies and would not bring much in productivity gains”¹¹. The escape is the same one Brynjolfsson prescribes — augmentation and the reinstatement of new, higher-value tasks — which is to say, redesign. The counterargument therefore does not undercut the thesis; it raises the stakes on it.

Synthesis: a decision frame for leaders

The literatures align on a single instruction with two axes. One axis is Hammer’s: are you embedding the existing process in software (paving the cow paths) or rebuilding it? The other is Brynjolfsson’s: is the AI substituting for people (automation) or extending what they can do (augmentation)? The high-value quadrant — redesign plus augmentation — is where BCG’s “Reshape,” McKinsey’s “rewiring,” Davenport’s process management, and the Centaur/Cyborg patterns all live. The low-value quadrant — pave-the-cow-paths plus pure substitution — is Acemoglu & Restrepo’s “so-so automation.”

The operational reading of every source in this paper is the same: the technology is the cheap, fast, commoditized part, and the value lives in the expensive, slow, organization-specific work of redesigning how the job gets done. Davenport — who helped found the discipline in 1990 and returned to it in 2025 — frames the current moment exactly this way: process management “is experiencing a renaissance, thanks to AI,” because “AI helps firms significantly scale up improved processes, and well-managed processes make it easier to obtain the high-quality data needed to train AI”¹⁸. Layer AI onto a poorly-understood process and you amplify its flaws; pair it with disciplined process redesign and the two reinforce each other. Wilson & Daugherty make the complementary point that natural-language interfaces now let non-technical employees participate in redesign directly — “kaizen 2.0,” an evolution of Toyota’s continuous-improvement model¹⁷. Thirty-five years after Hammer told managers to stop paving the cow paths, the instruction is unchanged; only the speed at which the choice between automating and rewiring now compounds has changed.

References