Behavior: The Variable You Have to Build 

In my previous article, I made a point that might seem counterintuitive given the billions being spent on AI: technology is often the easy part of the following equation: 

Impact = Technology × Behavior × Management 

These factors multiply rather than add. Strong technology paired with weak behavior or absent management produces limited results. All three elements need to work together. 

A common example of this would be an andon system installed in production that does not produce much in the way of results: The technology was installed but frontline behaviors for its use and management behaviors to assist were not developed properly. So the lights turned on, the board updated, but nobody responded in the intended way.  

I fear the same scenario is likely to happen with AI. Management will install AI Copilot, roll it out to everyone, and expect automatic results. If history is any indicator, that won’t happen. Think of any technology hardware or software implemented in the past several decades. Did ERP software installations deliver on their promise? Usually not unless the organization was extremely methodical about understanding not just the technology but the underlying behaviors, skills, processes, and systems required to make it work effectively.  

Here’s what separates behavior from technology: you can purchase technology, but you cannot purchase behavior. Skills must be developed. Habits must be formed through practice, feedback, and time.  

Technology Evolves, But Still Requires Human Response 

Let’s consider how the andon system has evolved at Toyota since the mid-1950s. In the beginning, the technology was simple — just light bulbs signaling that something needed attention. Over the decades, the system became progressively more sophisticated: signaling abnormalities, tracking equipment downtime, prompting quality checks, indicating tool changes, displaying production status. Today’s andon systems incorporate data collection and analysis that would have been unimaginable in the early years. 

I am very confident that in Toyota AI will make these systems even more capable. The possibilities are endless. AI can analyze vast amounts of data instantly. It can identify repeat problems, suggest changes to preventive maintenance, or possibly diagnose training needs, etc. This is genuinely valuable. Toyota is spending large sums of training effort on teaching AI skills to employees and how to use it in their daily work for analysis, and improvement.¹  

But here’s what AI doesn’t do: it doesn’t walk out onto the shop floor when a machine breaks down. 

When an abnormality occurs, a human team member responds. They assess the situation, attempt to restore normal conditions, and signal for help if needed. If the problem persists, a team leader responds, bringing broader troubleshooting skills and the ability to coordinate support. If the team leader can’t resolve it, a group leader gets involved — with authority to call maintenance or engineering and make supervisory decisions. Maintenance brings mechanical and electrical expertise. Engineering brings deeper technical capability for complex problems. Each of these human realms require behavioral and contextual skills to make the system work. Just adding an AI Copilot or chat box does not do that.  

This chain of human response from team members (team leader to group leader to maintenance to engineering) must be set up, aligned, and practiced over time as behavioral skill set. Each level brings different skills and capabilities. Each level has defined responsibilities. The escalation pattern is designed and reinforced until it becomes habitual.  

AI enhances the technology layer significantly. It makes the signals smarter and the data more useful. But the human behavior layer — the responses, the skills, the judgment calls remain essential. The andon board, no matter how sophisticated, doesn’t fix the problem. People do. 

Skills Are Analog, Not Binary 

One thing I want to emphasize: skill development is not a binary state. It’s not “trained” and “done.” Skill exists on a spectrum for both humans and AI adoption.  

Consider how capability develops in a production environment. A new team member learns the basic job through structured instruction, the important steps, key points, and reasons why. With practice and coaching, they become competent at performing the work to standard. Over more time, they learn to recognize abnormal conditions and respond appropriately. Eventually, they not only perform the work reliably to a given standard but also analyze it and suggest improvements. 

This progression from basic instruction to standardized work to kaizen unfolds over years. And even then, skill continues to develop. A team member with 10 years of experience sees things a new person doesn’t. A maintenance technician who has worked on a particular machine for years has intuitions that can’t be taught in a classroom.  

The same analog nature applies to the different roles in the response chain. Team members have basic operational skills. Team leaders have broader troubleshooting ability. Group leaders have supervisory judgment and authority. Maintenance and engineering bring specialized technical depth. These aren’t binary qualifications — they’re accumulated capabilities built through experience. 

The Parallel to AI Skills 

I see the same analog pattern in how people develop capability with AI tools. 

In an earlier article, I described five levels of AI collaboration. Most users remain at Level 1, typing casual questions into a dialogue box and getting variable results. At Level 2, people learn to structure their prompts more carefully and get more consistent outputs. At Level 3, they start working with APIs and parameters, controlling model behavior more precisely. Levels 4 and 5 involve building custom applications and integrating AI into workflows. 

This progression isn’t binary either. Someone doesn’t jump from Level 1 to Level 5. They develop capability incrementally through learning and practice, just like production skills. 

We’re in the early stages of the AI age. Most people and organizations are still figuring out what these tools can do and how to use them well. The skills and behaviors for effective AI collaboration will develop over time — and that development can’t be purchased or shortcut. 

Using AI Wisely vs. Wrongly 

There’s a choice embedded in how we develop AI-related behaviors. 

Used wisely, AI can make human behavior more collaborative and effective. It can provide information that helps people make better decisions. It can surface patterns that would take humans much longer to see. It can offer coaching and feedback when human experts aren’t available. In this mode, AI augments human capability. 

Used wrongly, AI can become a tool for monitoring, fear, and compliance. It watches people rather than helping them. It generates dashboard metrics that drive defensive behavior rather than improvement. It creates anxiety rather than capability. 

This is one of the hidden strengths I observed at Toyota. The andon system wasn’t designed to catch people making mistakes, it was designed to surface problems so they could be solved. The culture reinforced that pulling the cord was the right thing to do, not something to be avoided. The behavior of responding to abnormalities was supported, not punished. 

QUOTE Technology can be deployed on a project timeline. Behavior develops on a human timeline. 

As we develop behaviors around AI, we face the same choice. Will AI tools be positioned as helpers that make people more capable? Or will they be positioned as monitors that make people more anxious? The technology doesn’t determine this. The behaviors and management systems around it do. 

You Can Buy the Board, Not the Behavior 

The core point is simple: you can build or purchase an andon board or an AI system but you cannot purchase the skills and behaviors that make it effective. 

Those skills and behaviors have to be developed in people over time. The team member who responds quickly and appropriately to an abnormality developed that response through training and practice. The team leader who knows when to escalate and when to troubleshoot further built that judgment through experience. The maintenance technician who can diagnose a complex failure accumulated that skill over years. 

Similarly, the manager who uses AI to enhance problem solving rather than to surveil employees has developed a certain orientation. The engineer who knows how to verify AI outputs against reality has built a habit. The team that integrates AI insights into their daily work has established patterns through practice. 

Technology can be deployed on a project timeline. Behavior develops on a human timeline. 

The Equation Holds 

As we move further into the AI age, I expect we’ll rediscover what we’ve learned before with other technologies: that Technology × Behavior × Management together produce results. 

Strong AI capability with undeveloped human behaviors will underperform. Good behaviors without supporting management systems won’t sustain. And management systems without the underlying technology and behavior won’t have much to work with. 

QUOTE: You can purchase technology. The behavior you have to develop. 

In my next article, I’ll turn to the third factor — management systems — and how they integrate technology and behavior into sustained performance. 

For now, the takeaway is this: as you think about AI in your organization, recognize that technology is only one part of the equation. The behaviors, the skills, the response patterns, the habits of verification and judgment must be built. And that building takes time, intention, training, and development. AI won’t change that basic fact. 

You can purchase technology. The behavior you have to develop. 

P.S. 

Speaking of behaviors and technology. I will be hosting a half-day workshop at the upcoming LEI 2026 Summit in Houston. The general topic is Problem Solving and AI and how you can harness AI to accelerate your problem solving without letting it do the thinking for you. I’ll cover the “Five Levels of AI Framework” and give some examples you can try out using different tools at each of the file levels. We’ll start with some basic prompt engineering advice and show some more advanced features you can build on your own. For the workshop you will need to bring a laptop with some type of large language model (LMM) access and a real problem to work on. Hope to see you there!