5 Pillars of Composability
Bottom-Up, Agility, Democratization, Human-Centric, Compliance. The load-bearing test for whether a system is genuinely composable.
The Field Guide is deliberately built to be used, not just read. Here is what already exists, what is coming with the book, and where to go deeper today.
Each of these gets a full working treatment in the chapter that owns it — definition, mechanics, and a real deployment where it was applied.
Bottom-Up, Agility, Democratization, Human-Centric, Compliance. The load-bearing test for whether a system is genuinely composable.
Picking a first use-case by marrying business value, operator activity, physical requirements, and business process — starting at the gemba.
A deliberately throwaway app used to test whether a use-case is real before anyone commits to building the solution.
The smallest deployable thing that changes what happens at the station — plus the Diagnostic Test for spotting agile-flavored waterfall.
The four modes of engagement a first project team has to cover, mapped against the five roles a first use-case needs.
The core design artifact — three perspectives on the same operation, rendered so a team can argue about it productively.
The design test a proposed solution has to pass before it gets built.
Builder agents and staff agents as composable building blocks — scoped, governed, human in the loop.
Explore, Scale, and Transform — and how to tell when a deployment no longer needs central pushing to keep moving.
Naming a failure mode is what lets a team point at it in a meeting without accusing anyone. These recur across the book.
One system built on the premise that it can serve every process the same way, so the process adapts to the system. Implemented that way on composable technology, it becomes JAM: Just Another MES.
Lean’s visible artifacts adopted without the routines, relationships, and decision rights that make them work. Digital tools make the same mistake faster and more expensive.
The sprint cadence survives on the calendar while the actual sequence of work quietly returns to big-bang delivery.
Everything still gets built by one team, so the queue that democratization was supposed to remove simply moved.
The governance framework arrives before anything has shipped — so the program is policed before it has proved anything.
One brilliant, unsupervised builder becomes a single point of failure. The difference between citizen development and shadow IT is governance.
AI is the foundation of the digital technologies Augmented Lean brings to operations, and it amplifies the quality of the operation it runs on. Applied to an operation with daily frontline problem-solving and data captured at the source, AI makes that operation learn faster. Applied to incomplete, context-poor data, it produces recommendations that conflict with constraints only the operators know, and costs the operation its trust in both the AI and the Kaizen practice it was meant to accelerate.
The book treats agentic AI as an extension of composable architecture. A composable solution already works like a multi-agent system: each app has a defined goal, operates within its own scope, and collaborates with the others through shared data. Agents add a cognitive layer to that architecture. Chapter 1 traces the design logic back to the Toyota Production System, and Chapter 9 develops the agentic architecture of an augmented lean project. Every agent the book describes is scoped, governed, and keeps a human in the loop.
The agent patterns in the book are where the Exchange starts. The Exchange is a curated catalog of operational agents and the multi-agent solutions they compose into, organized by the Composable Agentic Framework. Each agent is placed by where in the value stream it does its work:
The Exchange is deliberately deep rather than broad. Its weight falls on classification, composition, oversight, and validation, because in physical operations a flawed instruction produces a defective part. The book gives the method; the Exchange will give the full definition of each agent and how agents combine into a solution.
Four rules govern every agent in the book. Design for the operator who didn't ask for it. Keep a human in the loop. Start with the data you have, not the data you wish you had. Measure the outcome, not the usage.
Open resources you can use before the book publishes.
Everything here assumes the foundation Chapter One describes. It is the right place to begin — and it is open.