Right Kind of Wrong

In the AI Story

The book opens with the argument that most organizational responses to failure are pathological. Blame-all cultures suppress the experimentation that generates learning. Blameless cultures forfeit the discipline that keeps experiments from becoming excuses. Edmondson's alternative is a structured vocabulary that allows organizations to hold rigor and exploration together — celebrating the engineer who tried a novel approach and documented what did not work, while holding accountable the one who skipped the checklist on a known procedure.

The three failure categories are developed with case-level precision. Preventable failures — the dropped pass, the forgotten step, the shortcut taken — are addressed through training, checklists, and accountability. Complex failures — the ones that emerge from multiple interacting factors in uncertain systems — are addressed through systemic redesign and organizational learning. Intelligent failures — genuine experiments in uncertain territory — are addressed through celebration, documentation, and the infrastructure for extracting lessons. Treating them identically is the central error the book corrects.

The relevance to AI is direct. Every organization deploying AI is, whether it recognizes it or not, running experiments. The tools are new, the use cases uncertain, the best practices unwritten. Whether those experiments produce learning depends on whether the failures they generate are categorized correctly and processed accordingly. Organizations that treat AI-related failures as preventable will suppress the exploration the transition demands. Organizations that treat them all as intelligent will lose the discipline that keeps experimentation rigorous. The distinction must be made, and making it requires the vocabulary the book provides.

The book also introduces the concept of failure literacy — the collective skill of recognizing which kind of failure one is facing and responding appropriately. Failure literacy is not natural. Organizations develop it through deliberate practice: structured retrospectives, explicit categorization exercises, leadership modeling of honest self-assessment. The AI transition, by compressing failure cycles from months to hours, makes failure literacy not optional but urgent. The organizations that develop it will adapt. The ones that do not will find themselves either paralyzed by fear of failure or undisciplined by indifference to it.

Origin

The book synthesized work Edmondson had published in scattered form since her 2011 HBR article 'Strategies for Learning from Failure,' which first popularized the three-category taxonomy. Its 2023 publication was explicitly positioned relative to the AI transition, with Edmondson noting in promotional materials that the capability to fail intelligently had become foundational to organizational survival in a period of rapid technological change.

Key Ideas

Three categories, three responses. Preventable, complex, and intelligent failures require different organizational responses; treating them identically is the central pathology.

Failure literacy. The collective skill of recognizing and responding to different failure types; developed through deliberate practice, not natural to organizations.

Rigor and exploration together. The framework rejects both blame culture and blameless culture in favor of structured differentiation.

Timing is not accidental. The book's publication was positioned to address the organizational capabilities the AI transition demands.