COLUMN – In this column, three lean and technology experts respond to the same pressing question shaping today’s tech/AI debate. This month, we ask what a lean approach to AI looks like.

Words: Sandrine Olivencia, Fabrice Bernhard and Eivind Reke

THE QUESTION

This Guardian article argues that workers need a voice in how AI is introduced in their workplace, and that involving them actually leads to better outcomes for both people and organizations. Lean has always said the same thing about any change. So why, in most AI rollouts, are workers still the last to be consulted? And what would a genuinely lean approach to AI implementation look like?

THE ANSWERS

Lean practitioners would likely agree with the core message of this article by The Guardian, but maybe for a different reason.

The article frames the issue primarily as one of participation and worker agency, but Lean sees worker involvement as something even more fundamental: the key to learning where value can actually be created.

At the Lean Summit in Paris last week, Thomas Métivier, CEO of French e-commerce giant Cdiscount, described AI as a form of "digital karakuri." In Toyota, karakuri are simple devices designed by teams themselves to eliminate tedious work. Their purpose is not to replace people, but to help them succeed by removing frustration, improving flow, and freeing people up so they can focus on work that requires judgment, creativity, and problem-solving. This is a useful way to think about AI.

Most AI initiatives begin with the technology and then look for an application for it, whereas Lean teaches us to start with a problem and look for a solution. Who knows where the waste is? Who spends hours searching for information, re-entering data, producing reports nobody reads, or navigating cumbersome procedures? The people doing the work. They know best because they deal with waste every day.

Yet, in many organizations, AI projects are selected by executives, consultants, or software vendors and then deployed onto teams that had little input into the decision. This is not just a people issue; it’s a learning issue.

Without the people closest to the work, organizations cut themselves off from their richest source of knowledge about where AI can generate value. As a result, many AI projects end up automating tasks that were never the real problem in the first place.

A genuinely lean approach would begin at the gemba: leaders would study how work is performed, identify the frustrations and obstacles preventing people from creating value, and run small experiments with them to discover where AI can help. Some ideas would fail, and others would spread naturally because teams find them useful. The objective would not be to automate people out of the process, but to free people from a (wasteful) process.

This question is interesting because it is not really about AI: it is about how organizations introduce change. Lean has wrestled with this for decades, and the lesson it provides us remains remarkably consistent: if you want better solutions, start by learning from the people who face the problems every day.

Artificial Intelligence has rapidly become one of the defining investment priorities of the global economy with Gartner estimating that AI-related spending could reach $2.5 trillion in 2026.

Indeed, 80% of organizations report using generative AI. Given that only 6% of the organizations recently surveyed by McKinsey report that their use of AI has generated significant value, however, the problem is how businesses are deploying this technology. Under pressure to prove they are not missing out on the next “Industrial Revolution,” leadership teams are launching many AI initiatives from the top down. At the same time, they need to justify the scale of investment. This combination creates a dangerous temptation: using AI not to improve the work, but to short-circuit management and more easily monitor, schedule, score, and control workers.

In The Guardian article mentioned above, Nazrul Islam refers to it as “bossware”—something that is not only ethically problematic, but also bad management and bad economics. It measures activity instead of creating value. And by reducing trust, it disengages knowledge workers—the very people whose judgement, creativity and problem solving are needed to turn AI into real business value.

The mistake organizations are making is deploying AI without the input of the teams on the ground who understand the work.

This is not a new issue. Manufacturing faced a similar illusion in the robotic automation wave of the 1980s: the belief that replacing humans with robots would automatically produce superior productivity. One lesson from that time is that some parts of the work can, indeed, be fully automated by robots (in car manufacturing, for example, it happened to body welding and painting). Yet, surprisingly for outsiders, 50 years later many steps are still very much human-powered, such as assembly. Even Elon Musk, who originally wanted to automate Tesla’s production much more than his competitors, ended up admitting he was wrong and that “humans are underrated.”

Toyota understood this early on, staying true to its "Jidoka" principle — automation with a human touch. Not because humans should do every task, but because humans are needed to interpret problems, understand customer value, and improve the system.

This lesson can easily be translated to technology and AI automation. In some use cases, AI can fully automate processes. But in most high-value use cases, it alone cannot enable the highest levels of quality without human supervision. More importantly, it cannot lead its own continuous improvement, taking into account ever-changing customer preferences.

AI used to manage people leads to both inhumane workplaces and worse quality. The right approach is to empower teams on the ground to experiment with AI. And where useful automations are identified and built, put humans in charge of continuous improvement of the system. Doing this sustainably and profitably at scale requires a learning system that supports continuous improvement driven by customer value in the whole organization.

This is the Lean Tech opportunity. By studying and translating the system that Toyota built over decades to create such a culture of human-powered excellence in manufacturing, we can leverage their lessons to better navigate the AI transformation. The effort to adopt and deploy such a culture in a company is real and explains why few organizations across the world have reached Toyota levels of lean culture. But the prize is more than worth it: better quality to outperform competitors and more human-centric AI adoption to contribute to a more humane society.

The uncomfortable truth is that most AI rollouts will start from the same assumption that has weakened many digital transformation initiatives. Technology is the change, and people are the adoption problem. In that logic, workers are consulted late because the key decisions have already been framed as technical ones: which system to buy, what data to use, what tasks to automate, what efficiency gains to expect. By the time employees are involved, the main question is often how to secure compliance, rather than how to improve the work.

Lean Thinking starts from a very different premise, in that it sees work as a socio-technical system: you cannot improve its technical side without understanding the human and organizational side. Kaizen is built on the idea that the people closest to the work are essential to improvement because they see abnormalities, variation, friction, and unintended consequences before anyone else. If AI is introduced without that knowledge, organizations risk automating assumptions rather than improving processes.

A recent SINTEF study on Digital Kaizen (to be published in APMS Conference Proceedings in late 2026) has shown this clearly. In the case study of a digital transformation of a medium-sized manufacturing company, we found that digitalization only created value when it was connected to everyday problem solving on the shop floor. Sensors, dashboards, machine vision, and data platforms did not create improvement by themselves. They became useful when operators, team leaders, maintenance staff, engineers, and managers used the data together to understand problems, test countermeasures, and learn. When leadership attention and competence development weakened, the same digital systems became passive information screens. The technology was still there, but the improvement capability regressed.

This is why many AI implementations will struggle. They are treated as linear maturity journeys: first implement the technology, then train users, then scale. But in practice, technology adoption will be fragile, uneven, and reversible. Some workers will become champions, others will only follow once the tools are embedded into routines, and others will resist because the relevance is unclear or because the technology threatens existing expertise and identity.

A genuinely lean approach to AI implementation would therefore begin at the gemba. It would ask: what problem are we trying to solve? What is the current condition? What do workers already know about the causes? Where does the process lack stability? What decisions could AI support, and what decisions should remain human? How will we know whether the change improves quality, flow, safety, learning, and engagement—not just cost? Such an approach would introduce AI through small PDCA cycles rather than large-scale deployment. Workers would be involved in defining the problem, interpreting the data, testing the tool, evaluating unintended consequences, and improving the standard. Leaders would frame AI not as a replacement for human judgment, but as an enabler of better problem solving. Training would be continuous and tied to real work, not a one-off system introduction.

In short, a lean AI implementation would not ask workers to adapt to technology already chosen for them. It would use their knowledge to shape what the technology should do, where it should be applied, and how it could improve the work. That is not only better for workers; it also makes for a better digital transformation.