Annotated Reading List

Ranked by how much they would change what somebody does, not by importance to the field. Several canonical works are far down this list because their central idea transfers in a paragraph and the rest is context. A few obscure ones are near the top because there is no substitute for reading them.

Where a work is cited more often than read, including by me, that is noted.

1. E. L. Trist and K. W. Bamforth, "Some Social and Psychological Consequences of the Longwall Method of Coal-Getting" (Human Relations, 1951).

The most useful thing on this list and the one almost nobody in operations has read. A mechanized method that was better on every technical measure produced worse results, and the paper works out exactly why: the old arrangement was carrying social load nobody had put in the drawing. If you build systems and read one item here, read this. The thirteenth rule in this book exists because of it, and everything in Part I that looked like an isolated unmeasured loss turned out to be this paper's finding arriving one symptom at a time.

2. Charles Perrow, Normal Accidents: Living with High-Risk Technologies (1984).

Perrow's two dimensions, interactive complexity and coupling, are the most useful diagnostic I know for deciding whether a system you are about to make more efficient is a system that can afford it. The contradiction he identifies — tight coupling requires centralized control, complex interaction requires decentralized control, and both are true at once — has no solution and knowing it exists changes what you build. Long, and the length is the argument: the case studies are where the pattern becomes undeniable.

3. Lisanne Bainbridge, "Ironies of Automation" (Automatica, 1983).

Under ten pages, forty years old, and still the best single thing written about what automation does to the person left behind. Every business case for automation should be required to answer it. The observation that the operator is left with precisely the tasks the designer could not think how to automate, and is therefore doing only hard cases with nothing easy in between, is worth the reading time on its own.

4. William M. Grove, David H. Zald, Boyd S. Lebow, Beth E. Snitz and Chad Nelson, "Clinical Versus Mechanical Prediction: A Meta-Analysis" (Psychological Assessment, 2000).

The most uncomfortable finding in this book for anybody who trusts their own judgment, including me. 136 studies; mechanical prediction substantially better in 47 percent, tied in 48, worse in 6. The detail that rescues human judgment is buried and does more work than the headline: in seven of the eight studies where clinicians won, the clinicians had more data. Read the results section; the discussion is for specialists.

5. James C. Scott, Seeing Like a State (1998).

The scientific-forestry chapter alone justifies the book, and it is the first chapter, so the cost of finding out whether you want the rest is low. Scott's concept of metis is the clearest available name for what an abstraction cannot hold, and his four conditions for catastrophe are a genuinely useful check on how much power sits behind any simplification you are proposing. Long, occasionally polemical, and right about the thing it is right about.

6. David J. Snowden and Mary E. Boone, "A Leader's Framework for Decision Making" (Harvard Business Review, 2007).

A magazine article, which is the correct length for this idea. The distinction between complicated and complex is the single most expensive thing to get wrong in applying anything in my creed, and this is where to get it. Skip the leadership advice; the domains are the content.

7. Michael Polanyi, The Tacit Dimension (1966).

Short, and harder than its length suggests. Worth reading rather than absorbing secondhand, because the secondhand version — some knowledge is hard to write down — is not what he argues, and the actual argument is considerably more damaging to any knowledge-management project. The face-recognition example does most of the work in the first twenty pages.

8. W. Edwards Deming, Out of the Crisis (1982).

Enormous, repetitive, and containing several chapters that will change how you read a variable result for the rest of your life. Read the material on operational definitions and on common versus special cause. The funnel experiment is the most useful demonstration in operations and takes twenty minutes to run with actual objects, which is worth doing rather than reading about.

9. Walter A. Shewhart, Economic Control of Quality of Manufactured Product (1931).

The origin of everything in the previous entry, and heavier going. Most readers are better served by Deming. Read Shewhart if you want the postulates in their original form, which are stated with a modesty that later writers dropped: a state of control is described as an economic limit rather than as a goal, and that framing is more honest than most of what followed.

10. Steven J. Spear and H. Kent Bowen, "Decoding the DNA of the Toyota Production System" (Harvard Business Review, 1999).

The best short answer to why copying Toyota's practices does not reproduce Toyota's results. The paradox they open with — rigid specification producing flexibility, because a rigidly specified operation is a series of controlled experiments — is the mechanism most standardization efforts are missing.

11. Donald A. Norman, The Design of Everyday Things (1988, revised 2013).

Knowledge in the head versus knowledge in the world is the concept worth taking, and it arrives early. The rest is excellent and mostly about physical objects, which readers building information systems will have to translate. Translate it; the translation is where the value is.

12. James Reason, Human Error (1990).

The taxonomy is the thing: slips and lapses are execution failures that design can remove, mistakes and violations are planning failures that design cannot. Knowing which kind you are looking at determines whether removing a step will help or hurt, and most people arguing about human error have not made the distinction.

13. Eric M. Eisenberg, "Ambiguity as Strategy in Organizational Communication" (Communication Monographs, 1984).

A useful corrective for anyone, like me, whose instinct is that clarity is always the goal. Eisenberg's four functions of strategic ambiguity are recognizable in any organization, and his line that clarity is only a measure of competence if being clear is the goal is the sort of sentence that rearranges a position.

Fifteen pages, written before the problem existed at scale, and it got the shape exactly right. The poverty-of-attention passage is quoted everywhere and the surrounding argument is not, and the surrounding argument is the useful part: that the design problem in an information-rich environment is filtering rather than distribution.

15. Friedrich A. Hayek, "The Use of Knowledge in Society" (American Economic Review, 1945).

Short, elegant, and generalizes well past economics. The knowledge of the particular circumstances of time and place is the same object Scott calls metis and Polanyi approaches from a different direction, and having three vocabularies for it is more useful than having one.

16. Alfred North Whitehead, An Introduction to Mathematics (1911).

One paragraph on page 61 is what anybody needs, and it is quoted in full in this book. The rest is a mathematics primer from 1911.

17. William James, The Principles of Psychology (1890), Chapter IV, "Habit."

The habit chapter is beautiful and still largely correct on the mechanism. Worth reading for the prose as much as the psychology. Note that the folded-paper and water-channel images everybody attributes to James are James quoting Léon Dumont, which is a small lesson in how attributions drift.

Twelve pages, and the most instructive failed prediction I know. He got the productivity right and the hours wrong, and the passages where he worries about what people will do with the leisure are better than the famous ones about the fifteen-hour week.

19. Percy W. Bridgman, The Logic of Modern Physics (1927).

The operational definition idea transfers in two sentences, which are in Chapter 2 of this book. Read Bridgman if you want to see a physicist working out in real time that his field's vocabulary had been quietly wrong, which is an unusual document.

20. E. F. Codd, "A Relational Model of Data for Large Shared Data Banks" (Communications of the ACM, 1970).

Foundational and technical. Most readers want the idea of atomicity and normalization, which is available anywhere, plus the knowledge that Date and Darwen later showed the concept has no absolute meaning, which is what anybody applying it outside a database has to know.

The division between control, planning, and improvement, and the distinction between sporadic spikes and chronic waste, is the cleanest resolution available for the standardize-versus-improve problem. The primary sources are handbooks and the idea is better encountered in summary.

22. Ikujiro Nonaka and Hirotaka Takeuchi, The Knowledge-Creating Company (1995).

Listed because it is the foundation of thirty years of corporate knowledge management and because reading it alongside the next entry is more instructive than reading either alone.

23. Stephen Gourlay, "Conceptualizing Knowledge Creation: A Critique of Nonaka's Theory" (Journal of Management Studies, 2006).

The critique that made me rewrite the title ism of this book. Three of the four conversion modes appear plausible and none are supported by evidence that cannot be explained more simply. If you have ever run or funded a knowledge-management initiative, this is uncomfortable and worth the discomfort.

Goodhart's original sentence is about statistical regularities collapsing under control pressure, which is a stronger and stranger claim than the popular version. Strathern's compression — when a measure becomes a target it ceases to be a good measure — is the one everybody quotes and is not Goodhart's wording. Both are one sentence each; there is no reading to do, only the correction of an attribution.

A joke that turned out to be a finding. Nobody needs to read the essay.

26. David Graeber, Bullshit Jobs (2018), read against Magdalena Soffia, Alex J. Wood and Brendan Burchell, "Alienation Is Not 'Bullshit'" (Work, Employment and Society, 2022).

Included as a pair and in this order. Graeber's thesis is vivid and widely believed; the empirical test finds 4.8 percent rather than 20 to 60, and declining rather than rising. Reading the second without the first is unfair to Graeber, and reading the first without the second is how the figure keeps circulating.

The core claim transfers immediately and the literature is large and contested in its details. Take the claim, skip the argument, unless you are designing instruction.

For somebody who builds systems and has an afternoon: Trist and Bamforth, Bainbridge, and the results section of Grove. Those three cover the social cost of a technical improvement, what automation does to the person left behind, and how badly calibrated expert judgment is about its own accuracy. Between them they contain most of what I got wrong.

For somebody who manages people who build systems: Snowden and Boone, Perrow, and Eisenberg. Domain, structural risk, and the recognition that clarity is a goal rather than a virtue. Those are the three places where a good decision looks like a bad one from a distance.

For somebody who is skeptical that any of this applies to them: Gourlay and Soffia. Both are demolitions of things widely believed, one of them a thing I believed and built on. If you want to know whether this book's method is honest, the test is whether it reports findings that damage it, and those two are where it does.

There is no book on this list about habit formation for a general audience, no book about deep work, and none of the large popular syntheses that cover this territory. That is not a judgment about their quality. It is that every one of them draws on primary sources that are also on this list, and the primary sources are usually shorter, more careful about what they claim, and franker about what they do not know.

There is also nothing here about artificial intelligence, which some readers will find strange in a book about automation written in 2026. The reason is that the arguments in this book are about the relationship between a person, a rule, and the thing the rule is about, and that relationship does not change based on what is executing the rule. Everything Bainbridge says about the operator left monitoring an automated process applies without modification, and reading her is better preparation than reading anything written in the last three years.