Is Kaizen Only Possible by Humans? Decision Trace Model × Multi-Agent Systems Unlock the Next Evolution of Improvement

In manufacturing, improvement (Kaizen) holds a special meaning.

  • Insights from the shop floor
  • Ingenuity of workers
  • Continuous accumulation of small improvements

Through these, Japanese manufacturing has built world-class competitiveness.

And for many years, it has been said:

👉 “Improvement can only be done by humans.”

But is that really true?

Why hasn’t AI been able to enter Kaizen?

AI has already been introduced into factories for:

  • Demand forecasting
  • Anomaly detection
  • Quality inspection

However,

👉 it has barely contributed to improvement itself

The reason is clear.

① Improvement is about deciding “what to do”

Improvement is not just analysis.

On the shop floor, people constantly decide:

  • What is the real problem?
  • What should be changed?
  • How far should we go?
  • What risks are acceptable?

In other words,

👉 Kaizen is a continuous process of deciding what to do

Traditional AI, on the other hand, is strong at:

  • Prediction (what will happen)
  • Classification (what is happening)
  • Detection (whether something is abnormal)

But,

👉 it cannot decide what action to take

Because that requires balancing:

  • Cost
  • Quality
  • Delivery
  • Safety

👉 a multi-dimensional trade-off

② The context changes every time

Improvement in factories depends heavily on:

  • Equipment conditions
  • Worker skills
  • Daily workload and delays
  • Customer-specific quality requirements
  • Safety constraints

👉 the context at that moment

Even if problems look similar,

👉 different contexts lead to different optimal decisions

This is not just about having many variables.

👉 The assumptions themselves change every time

Which makes it difficult for fixed models to handle.

③ Improvement relies on tacit knowledge

Kaizen heavily depends on:

  • Veteran intuition
  • The “feel” of the shop floor
  • Experience-based judgment

👉 knowledge that is not explicitly articulated

These are:

  • Not formalized as rules
  • Not structured as data

Therefore,

👉 they are not directly usable by AI

The essence of Kaizen

So what is Kaizen, fundamentally?

👉 Kaizen is the accumulation of decisions and trial-and-error on the shop floor

Every day, workers:

  • Identify what to fix
  • Decide how to fix it
  • Choose when to act
  • Determine how far to go

But more importantly,

👉 they do not stop at decision-making

They:

  • Try it out
  • Observe the results
  • Check for unintended consequences
  • Reflect and improve further

In other words,

👉 Kaizen is a loop of:
Decision → Execution → Evaluation → Learning

Introducing Decision Trace Model × Multi-Agent

The key to solving these challenges is:

👉 externalizing decision-making as a structure

Traditionally, Kaizen has been:

  • Decided in people’s heads
  • Executed on the shop floor
  • Accumulated as experience

Which means:

👉 decisions are invisible, non-reproducible, and hard to share

Decision Trace Model

Decision Trace Model treats the improvement process itself as a structure:

Event → Signal → Decision → Boundary → Execution → Log

Applied to Kaizen:

  • Event
    Line delays, defect increases, variability in work time
  • Signal
    Bottleneck identification, root cause estimation, load imbalance
  • Decision
    Change process, reassign workers, add temporary buffer
  • Boundary
    Quality constraints, safety requirements, cost limits
  • Execution
    Line adjustments, workflow changes, scheduling updates
  • Log
    Why the decision was made, what happened, what was learned

👉 This directly structures the Kaizen loop
👉 (Decision → Execution → Evaluation → Learning)

As a result:

  • Decisions become visible
  • Reproducible
  • Transferable across lines

👉 Improvement becomes an organizational asset

Multi-Agent decomposition

Another critical point:

👉 Do not let a single AI handle everything

Kaizen is complex, so it must be decomposed:

  • Signal Agent
    Detects anomalies and bottlenecks, understands real-time conditions
  • Decision Agent
    Generates and selects improvement actions
  • Policy Agent
    Ensures compliance with quality, safety, and rules
  • Risk Agent
    Evaluates impact across processes and risks
  • Execution Agent
    Applies changes to operations and workflows

👉 Together, they realize structured, distributed decision-making

Why this evolves Kaizen

This structure removes the core limitations of traditional Kaizen:

① Trade-offs → Structured decisions

Decisions are no longer ad hoc.

👉 They are evaluated, recorded, and explainable

② Context dependency → Context-aware decisions

Instead of fixed rules:

👉 decisions adapt dynamically to real-time conditions

③ Tacit knowledge → Explicit knowledge

Instead of being locked in individuals:

👉 knowledge is captured, shared, and reused

👉 In short:

👉 Kaizen itself becomes structured

What actually changes?

This leads to a fundamental shift.

Traditional Kaizen

  • People notice
  • People think
  • People improve

👉 Human-centered system

Next-generation Kaizen

  • Systems understand context
  • Systems execute structured decisions
  • Systems continuously learn

👉 Decision-centered system

The fundamental shift

👉 Improvement is no longer about optimizing tasks
👉 It is about optimizing decision-making itself

Relationship with the Toyota Production System

This does not reject the Toyota Production System.

👉 TPS = The ultimate human-driven Kaizen system
👉 DTM = A system that reproduces and extends that capability

👉 Not a replacement, but an evolution

Conclusion

Kaizen has always been:

👉 A human-driven improvement system

But due to:

  • Increasing complexity of trade-offs
  • Changing contexts
  • Dependence on tacit knowledge

👉 its limitations are becoming visible

Decision Trace Model × Multi-Agent transforms Kaizen by:

  • Structuring decisions
  • Enabling reproducibility
  • Executing in real-time
  • Sharing knowledge across organizations

👉 Improvement evolves from
👉 a human activity
👉 to a continuous decision-making system

The role of humans

Humans will focus on:

  • Final value judgment
  • Exception handling
  • Meaning-making

👉 From “people who make decisions”
👉 to “people who design and govern decisions”

Final message

Kaizen will not disappear.

In fact, its importance will increase.

But,

👉 its implementation will fundamentally change

👉 Kaizen was human-driven.
👉 Now, it becomes a continuous decision system.

This transformation will become:

👉 the foundation of next-generation manufacturing

タイトルとURLをコピーしました