Continuous improvement has always been the backbone of the manufacturing industry. From the Toyota Production System to Lean and Kaizen, the principle is simple: optimize processes as they happen, actively solve problems, and involve employees through regular improvement cycles to increase efficiency and customer satisfaction.
This discipline has gone through three major transformations. From 1960 to 2000, it was the Invention phase: Toyota formalized standards, GE and other groups amplified the movement, and operational excellence became a global discipline guided by clearly defined Lean Daily Management frameworks. From 2000 to 2025, Digitalization accelerated decision-making, enabled company-wide standardization, and made field data actionable.
Since 2026, a third transformation has been underway: the Agentic era, where AI and teams collaborate continuously to monitor, improve, standardize, and steer the operational system, constantly pushing performance levels higher.
But before going further, one thing must be clarified: Agentic Continuous Improvement (Agentic CI) is not intended to replace humans. Rather, it keeps people at the center of continuous improvement processes, amplifying their capabilities and providing new levers to maximize efficiency. Continuous improvement has always relied on the people doing the work, and that does not change with AI. What changes are the tools these people have at their disposal to work.
What neither paper nor digital has solved
Continuous improvement has always faced the same three structural limits. Digitalization made some things faster and more traceable, but it did not solve them.
- Knowledge remains local: When a team faces a problem in the field, they think, test solutions, and iterate. What they often don’t know is if another team, in another factory, has already solved this same problem six months earlier. This knowledge exists somewhere in the organization: in someone’s memory, in a report, buried in a shared folder that no one opens under pressure. Digitalization has allowed for more data capture, but it has not made this knowledge visible and accessible.
- Know-how evaporates: With every departure of an experienced operator, supervisor, or plant manager, years of knowledge leave with them. The kind of knowledge that is never really written down: how to read the behavior of a particular machine, what adjustment really works on this line, what invariably happens when production pressure rises. This knowledge has never been fully captured, meaning it cannot be passed on, and the next person starts from a lower level.
- Patterns remain invisible: A plant manager responsible for several buildings, multiple lines, and rotating teams cannot track every signal that emerges in this complexity. Dashboards show overall performance. They do not reveal that a specific quality deviation appears every Monday morning on a line, or that a recurring maintenance problem is linked to a particular changeover sequence. By the time these patterns surface, they have already had an impact.
These problems are not new. They existed before digitalization, and they have persisted through it. Not for lack of effort or tools, but because the underlying model, where knowledge remains in the team or site that generated it, has never been questioned.
With AI, a third wave is emerging: Agentic CI, or agentic continuous improvement
Digitalization changed how continuous improvement is practiced. The arrival of AI changes what continuous improvement is capable of.
Where digital tools made existing routines faster and more traceable, AI introduces something qualitatively different: the capacity for an organization to learn from its own daily operations, continuously and at the scale of all its sites.
The knowledge generated in the field every day (every problem solved, every solution validated, every pattern that emerges) can finally be captured, shared, and acted upon at scale. This is what Agentic CI is based on, and it’s what the two previous phases never achieved.
Without Operational Memory, Industrial AI Remains a Chatbot
It’s important to clarify that AI isn’t magic and that an agent needs data to learn and perform effectively. It needs structured routines, defined problems, versioned standards, and a history of daily activities organized over time. Without this foundation, industrial AI remains a chatbot placed on top of machine data: capable of retrieving information, but unable to improve the system.
This has a name: what we can call a business ontology, or domain ontology. It is a structured representation of the entities that truly matter on the ground (rituals, teams, problems, standards, actions) and the relationships that connect them.
Without this structuring layer, AI receives data but doesn’t understand what it means for the organization. It can’t distinguish a one-off incident from a recurring pattern or know that a solution tested at one site deserves to become a standard shared across the rest of the network.
This is precisely what makes building this layer strategic for any industrial SaaS player aiming for continuous, agent-based improvement. A business ontology cannot simply be bought or automatically generated by a language model. It is built progressively, with business teams, through genuinely facilitated rituals and genuinely solved problems.
It is a fundamental undertaking, what can be described as a form of Essential Domain Intelligence (EDI), that distinguishes a simple conversational assistant grafted onto an existing product from a system truly capable of learning and improving an organization’s standards over time.
What Agentic CI changes in concrete terms
Once this is available and structured, agentic continuous improvement ceases to be an abstract concept. It becomes directly integrated into the rituals that teams already perform every day with operators, without replacing the existing operational model.
Four changes stand out in particular:
Accessing the best of the organization’s knowledge: Imagine a team faced with an anomaly on a production line: a scrap rate that suddenly rises, a machine whose behavior changes. Today, this team relies on what the people present know. Tomorrow, they can access what the entire organization has already learned about similar situations: how this type of problem was diagnosed elsewhere, what solutions were tested, which one worked, and in what context. Less time spent rediscovering what is already known, more time on what is new.
What one factory learns becomes the standard for all: Your site faces a new problem: it brainstorms, tests different solutions, verifies their effectiveness. The challenge is to ensure that what has been learned locally becomes the global standard. Agentic CI detects these opportunities: when a solution is validated in one site, it identifies that a new standard can be created and shared with the entire network. What one factory solves once, no other needs to solve again.
Surface patterns that no one can see alone: In a system as distributed as that of an industrial group (multiple continents, multiple buildings, teams working in 3×8 shifts), there are patterns that no plant manager can spot alone: a problem that repeats systematically at certain times, a deviation that correlates with a specific product reference, etc. Agentic CI surfaces these signals as opportunities for action, allowing management to intervene top-down on what the data reveals, before these patterns become chronic problems.
Moving from ad-hoc audits to continuous measurement of operational maturity: This is perhaps the most structural change: today, the maturity of a site is evaluated during infrequent audits, biased by the preparation effect, and often too late to correct what is drifting. Agentic CI continuously measures the real maturity of each site on every dimension: adherence to routines, quality of problem-solving, pace of escalations. It detects drifts before they become deviations and indicates exactly where to intervene. Operational discipline ceases to be a post-mortem observation and becomes a real-time signal.
An operational necessity
What distinguishes Agentic CI from previous waves is not a new technology, but an optimized operating model.
Where traditional continuous improvement enhances execution within a defined operating model, Agentic CI will improve that model by continuously integrating what teams learn on the ground into the standards applied across the entire organization.
This matters more today than it did a decade ago because industrial companies now operate in a context of persistent volatility: faster-changing demand, more frequent supply chain disruptions, and a skills gap that makes losing experienced employees costly.
In this environment, the ability to learn faster than problems accumulate is not a competitive advantage. It’s an operational necessity.
AI, yes, but to help field teams
The question that comes up most often when we discuss this subject with industrial leaders is legitimate: will the introduction of AI into continuous improvement processes reduce the role of field teams?
In practice, it is often the opposite.
The greatest friction in daily continuous improvement is not reflection, but having to start from scratch every time. When operators and supervisors have access to what the rest of the organization already knows, they spend less time reconstructing the context and more time acting: deciding what fits their situation, adjusting solutions, generating new learning.
Continuous improvement has always relied on the people who do the work. AI-powered continuous improvement doesn’t change that, but it gives them better tools to work with.
An agent can find a similar problem. It can identify a pattern. It can suggest that a solution tested elsewhere is worth investigating. But it’s still the teams that understand the context, define the problem, test the solution, and decide whether a learning should become a standard.
AI doesn’t eliminate the continuous improvement loop, but it increases its speed and scope. And above all, it allows each new loop to enrich the subsequent ones.
It is this cumulative mechanism that gradually transforms an organization that solves problems into an organization that learns from every problem it solves.
The Choice at Hand Is Not a Choice of Tool
An industrial company that digitizes its daily management today is not simply selecting software for the next three years. It is choosing the foundation on which its operational intelligence will be built for the coming decade.
The cost of such a change is rarely fully appreciated. A day-to-day management system isn’t a database that can be migrated in a few weeks. It involves dozens, sometimes hundreds, of sites, thousands of operators trained in specific procedures, and years of accumulated problem-solving history.
Changing it isn’t a simple IT project, but a new approach to change management across the entire group, an exercise few organizations can afford to repeat twice in ten years.
This cost does not stop at the migration itself. Every year spent on a system that does not capitalize on field knowledge is a year of know-how that is not structured anywhere, and therefore remains vulnerable to turnover, departures, and organizational oversight.
This prompts a simple question for any vendor during the selection process:
Does the product offer an AI layer added to the existing system, or is its data model designed from the ground up so that agents can reason within it?
The two may look similar in demonstrations, but they have very different capabilities.
A chatbot plugged in next to a product answers questions. It doesn’t perceive the structure of rituals, can’t distinguish a validated standard from a draft, and can’t detect that a solution proven at one site deserves to become the network’s standard.
A system built on a true business ontology, on the other hand, learns from what happens and continuously improves standards.
In an environment where volatility has become constant and where know-how leaves factories faster than it is acquired, this capacity for collective learning is no longer a competitive advantage. It is the foundation upon which tomorrow’s industrial performance will be built. Continuous agentic improvement is the infrastructure for this learning.
What This Changes in Concrete Terms When Choosing a Solution
For an industrial company preparing to choose or renew its Daily Management System, this question must now carry as much weight as price, ergonomics, or functional coverage:
Is the chosen solution designed to accumulate and structure operational knowledge over the years, or does it simply digitize rituals without ever learning from them?
The choice of a DMS is therefore no longer just a choice of field collaboration tool. It is the choice of the foundation on which the operational intelligence of tomorrow will be built or not.