AI adoption depends on more than model progress: insights from Art Smalley

Art Smalley and the article’s context

Art Smalley is a Lean expert who spent ten years living, studying and working in Japan and was among the first foreign nationals to work for Toyota. He later helped transfer Toyota’s production, engineering and management systems to overseas facilities and became McKinsey & Company’s leading expert in Lean manufacturing.

In an article published by the Lean Enterprise Institute on 5 October 2026, he examines what the history of work improvement can teach us about AI adoption. The following sections review his insights; the final section presents LeanAI’s perspective on applying them.

Who should think about improving work?

Smalley compares two organisational choices. In Frederick Taylor’s scientific management system, planning and analysis largely belonged to management and specialists, while workers followed the prescribed method. Toyota sought to give the people doing the work greater influence over its planning and improvement.

AI adoption presents a similar choice. An internal expert team can build solutions for other employees or also develop their ability to analyse and improve their own work. In Smalley’s view, long-term progress requires technical expertise without removing improvement responsibility from the people who understand day-to-day operations.

What sits beneath a convenient AI interface?

Chat interfaces, prompts, document summaries and simple agents are the most visible part of AI. Reliable organisational solutions also require data architecture, software engineering, security, permissions, integrations, output evaluation and knowledge of the specific domain.

Smalley points to a similar experience with Lean: copying visible tools and practices is easier than building the engineering and organisational capabilities behind them. Access to an AI tool can be a useful starting point, but it does not by itself establish a company’s ability to apply AI reliably.

Why do individual gains not add up automatically?

AI can be adopted quickly for individual tasks: drafting a text, analysing material or automating a bounded action. Moving to an entire process introduces dependencies between systems, functions, permissions and responsibilities.

A local improvement can shift work, cost or risk to another team. Smalley therefore treats coordination as part of the AI implementation challenge itself. Organisations must agree not only how a tool works, but also who accepts its output, handles exceptions and remains accountable for the next step.

Measure operational results as well as AI activity

Counts of licences, prompts, demonstrations or agents do not show whether work has improved. Smalley recommends examining the quality of the work itself: whether problems are solved more effectively, the system operates more smoothly and people develop their capabilities.

His message for leaders is that technological capability and an organisation’s ability to use it develop at different speeds. Learning, work redesign and aligned responsibilities are essential parts of the change.

What do these insights mean from a LeanAI perspective?

LeanAI’s approach is to start AI implementation with a significant business challenge and the process constraining its resolution. Lean helps clarify workflows and unnecessary activities; we assess AI opportunities alongside employee knowledge, data and solution reliability.

For example, AI may speed up order entry while the order still waits for approval, leaving the overall fulfilment time largely unchanged. The approval flow and accountability then also need attention. This is an illustrative example, not a client result.

A practical first step is to select one business-critical process, involve its participants and agree on the outcome to improve. Value is demonstrated through changes in quality, delivery times, costs or other agreed measures.