A significant shift is underway in the corporate push for artificial intelligence, as tech firms begin urging companies to pay for AI agents out of their labor and services budgets rather than their software accounts. Earlier this month, Workday Inc. signaled this transition to investors, with its leadership suggesting that the company itself is looking to replace human new hires with autonomous AI systems.
While the promise of replacing expensive, variable human effort with cheap, uniform machine output is attractive to the bottom line, management historians and researchers warn that corporate America has run this experiment before with disappointing long-term results. The move toward total algorithmic uniformity risks reviving the “productivity dilemma,” a concept championed by the late William Abernathy of Harvard Business School, which posits that extreme gains in efficiency often come at the direct expense of innovation.
The current AI boom bears a striking resemblance to the Six Sigma movement that dominated the 1990s and early 2000s. Originally developed at Motorola Inc. in 1986 and later popularized by General Electric, Six Sigma focused on reducing variance in business routines to near zero. While highly effective for factory floors and repeatable manufacturing, management scholars Mary Benner and Michael Tushman found that the methodology had a chilling effect on creativity. Their study of the paint and photography industries over two decades revealed that as companies leaned harder into process management, their innovation became increasingly incremental—building only on what they already knew rather than exploring radical new approaches.
This trend is best illustrated by the historical decline of innovation at 3M Co. Known for breakthrough inventions like the Post-it note, 3M adopted Six Sigma under a former GE executive in 2000. Historically, 3M had a goal that at least one-third of its sales should come from products less than five years old. By 2007, under the weight of efficiency metrics that viewed speculative experimentation as waste, that share had dropped to just one-quarter. The co-inventor of the Post-it note famously noted that innovation is a numbers game requiring thousands of raw ideas to find one success—a process that modern AI agents are often optimized to eliminate as inefficient.
The technical nature of AI introduces new risks that Six Sigma did not. AI models, including the latest reasoning systems like OpenAI’s o1, operate based on probability distributions found in their training data. A 2024 study from Princeton University highlighted this limitation: when researchers asked GPT-4 to decode a simple cipher, it succeeded 51% of the time for likely sentences but only 13% of the time for unlikely ones, despite the decoding rule remaining identical. This suggests that AI is fundamentally biased toward the probable and the average, whereas human innovation often stems from eccentric backgrounds, stubbornness, or even productive mistakes.
Some industries are already integrating AI in ways that emphasize this trade-off. Danaher Corp. has developed an AI-powered laboratory that designs and tests antibodies, feeding results back into the model to refine the next batch. While this “test-and-verify” discipline is highly effective when the target is clearly defined, experts warn that applying the same rigid logic to broader business problems could trap companies in a cycle of repetition.
There is evidence that AI can succeed through sheer persistence. In May, OpenAI reported that one of its models disproved a long-standing geometry conjecture from 1946 by exploring paths a human mathematician might have dismissed. However, in the world of pure mathematics, a failed path costs only computing time. In a corporate environment where failures represent lost customers and wasted capital, companies are likely to tune their AI systems to only pursue the safest, most predictable outcomes.
The challenge for modern leadership is to protect the space for human eccentricity while leveraging the speed of automation. 3M eventually recognized this imbalance, releasing its research scientists from Six Sigma requirements to restore their creative freedom. As companies like Workday look to automate the “rote work,” the risk remains that they may inadvertently eliminate the human variance that leads to the next industry-defining breakthrough. If software replaces the human capacity for the implausible, businesses may find themselves optimized for a peak that is already behind them.
