Skip to content
Published: Built with: book · zenn · software-design · architecture · organization · team-topologies · ai

Cognitive Feedback Design (Book, 2nd ed.)

A technical book on Zenn (in Japanese) arguing that system design is the design of the paths, bandwidth, latency, and encoding of feedback. From that thesis, it reads design practices such as types, CI, ADRs, modularization, the inverse Conway maneuver, Team Topologies, and the use of AI. Five parts, 17 chapters, and appendices.

A book published on Zenn (paid, 500 yen; written in Japanese). The full title is 認知フィードバックデザイン 第2版——経路・帯域・遅延・符号化で読むソフトウェアと組織 (Cognitive Feedback Design, 2nd edition: Software and Organizations Read Through Path, Bandwidth, Latency, and Encoding).

Zenn: 認知フィードバックデザイン 第2版(opens in a new tab) (the introduction through Chapter 8 and the appendices are free to read)

What the book is about

Modularization, types, tests, CI, code review, ADRs, team splits, the inverse Conway maneuver. Software development has countless design principles and practices, each with its own books. The book starts from one question: are these really separate things?

Its answer fits in a single thesis.

System design is the design of the paths, bandwidth, latency, and encoding of feedback.

"System" here means the whole network whose nodes are people, AI, code, tests, documents, and the running product, with "signals about the consequences of actions" flowing between them.

  • Path: Does the signal reach the person who needs to know?
  • Latency: How long between an action and knowing its consequence?
  • Bandwidth: Can the receiver process the signals that arrive?
  • Encoding: Is the signal in a form from which the receiver can recover the intent?

For example, types create a path from the compiler to the developer and shorten its latency. ADRs encode the reasons for a decision so that future readers can decode them. Practices that look unrelated are reread as operations on one or more of these four axes.

As of November 2026, it hasn't collected many likes on Zenn, but 10 copies have been bought (probably mostly by people at the projects I've worked with).

Who it is for

Engineers with hands-on experience who know individual design principles and want a systematic understanding of the structure behind them — especially tech leads, architects, and engineering managers who design both code and organizations. On AI, it is for people who want to think not about how to use the tools, but about "where in the network AI sits, and how that changes each axis."

Structure

Five parts, 17 chapters, and appendices (glossary and references).

  • Part 1, Model (Chapters 1–3): Defines the four axes and the four constraints behind them, and sets up models of the receiver (a holder with a schema) and of artifacts, which store signals and send them out again
  • Part 2, The Four Dimensions (Chapters 4–8): One chapter each on path, latency, bandwidth, and encoding, then the coupling and trade-offs between axes
  • Part 3, Synthesis (Chapters 9–12): Shows the four axes recurring at every scale from functions to product portfolios, organizations as a communication network, and the three Team Topologies interaction modes as three configurations of the channel at a boundary, then turns this into a way to diagnose the axis behind a symptom
  • Part 4, AI as a Node (Chapters 13–15): When AI joins the network, what gets cheaper, what becomes the new bottleneck, and what is lost
  • Part 5, Eras and Limits (Chapters 16–17): Rereads the history of software development as a shift in which axis is the bottleneck, and asks what remains when each constraint is lifted

How the structure was designed

  • One case followed through the whole book. A change to a subscription billing system that "went unnoticed for over a month" is introduced in Chapter 1, examined one axis per chapter, and summed up in Chapter 12. The abstract discussion of axes keeps coming back to the same situation
  • "Diagnostic questions" in every chapter. Each chapter ends with three questions for observing that chapter's axis in your own workplace, so the ideas can be applied to your own team
  • Stating up front what the book does not claim. It does not claim the four axes explain everything. Chapter 1 shows how precisely each axis corresponds to communication theory and where it becomes a metaphor. Where the book is reinterpreting existing practices through the four axes, it says so rather than calling it a derivation
  • Turning diagnosis into action. "This team communicates badly" leads to no fix. "The signal from accounting passed through four human relays, arrived more than a month late, and was phrased in words outside the developers' vocabulary" tells you where to fix. What the book offers is a coordinate system for breaking design decisions down into observable variables

What changed in the 2nd edition

The first edition was published in February 2026. The 2nd edition (September 2026) is a full rewrite, with a new title and structure.

  • No longer derived from a single axiom. The first edition derived design principles from one axiom: the finiteness of cognitive holding load. The 2nd edition notes that even with unlimited capacity, the time for a signal to arrive never reaches zero, and loss in representation happens regardless of capacity. So it sets out four independent constraints and the four design axes that correspond to them. The capacity constraint becomes the one behind a single axis, bandwidth
  • Rewritten into 17 chapters. The chapter structure was reorganized from the first edition's 15 chapters, and a single case is now followed through every chapter
  • The subtitle changed from "the unified principle of software design derived from finite holding load" to "software and organizations read through path, bandwidth, latency, and encoding," and the subtitle on the cover was updated too