Diagnosis as substitute for action — in standups, hospitals, and backlogs.
The same bug has been on the board for three sprints. Everyone in the standup knows it. On the first Monday, someone called it a race condition in the connection pool. On the second sprint, someone else refined the language: actually, it’s a deadlock under high concurrency, probably related to the timeout configuration. By the third sprint, the team had a taxonomy — connection pool exhaustion, stale connection reuse, timeout cascade — and a Confluence page with a diagram. They had described the problem so thoroughly that describing it felt like fixing it.
Nothing was fixed.
This pattern — describing a problem until description substitutes for repair — operates across an unlikely range of human activity: cognitive psychology, organizational management, software engineering, and clinical medicine. The mechanism is the same in each domain. The consequences scale differently. In a standup, it costs a sprint. In a hospital, it costs something else.
In 2002, Daniel Kahneman and Shane Frederick formalized something that had been hiding in plain sight. They called it attribute substitution: when faced with a difficult question, the mind unconsciously replaces it with an easier one and answers that instead.1 The substitution is invisible to the person making it. Asked “Should I invest in Ford stock?” — a question requiring financial analysis — people instead answer “Do I like Ford cars?” — a question requiring only a feeling.2
The mechanism runs on Kahneman’s dual-process framework. System 1 — fast, automatic, effortless — can label and categorize nearly instantly. System 2 — slow, deliberate, energy-expensive — is required for genuine problem-solving. Here’s the critical part: when System 1 successfully labels something, it generates a sensation of cognitive ease that is neurologically indistinguishable from the sensation of having understood it.2 Your brain cannot tell the difference between naming a problem and solving it without deliberate, effortful System 2 intervention.
This is what makes the substitution dangerous rather than merely interesting. When an engineering team faces “How do we fix this?” (hard — requires design, risk assessment, coordination, and someone willing to touch that part of the codebase) and instead answers “What should we call this?” (easy — requires only taxonomy), the swap is invisible because both activities look like productive engineering work. The meeting ends with a sense of progress. The progress is genuine. The fix is not.
Yale psychologist Laurie Santos named the recursive version of this trap: the G.I. Joe Fallacy, after the cartoon’s tagline “knowing is half the battle.”3 Santos’s point is that knowing is far less than half. Awareness of a cognitive bias does not inoculate you against it. And so the teams that recognize they have a naming-not-fixing problem will predictably attempt to close the gap by naming it more precisely. They will hold a retrospective about why retrospectives don’t lead to action, producing an action item about improving action items. The recursion is built into the cognitive mechanism itself.
Individual cognitive substitution is a bug. When organizations systematically reward it, it becomes architecture.
Jeffrey Pfeffer and Robert Sutton, after four years of research at Stanford’s Graduate School of Business studying companies including General Electric, Toyota, and Southwest Airlines, documented what they called the knowing-doing gap: the systematic failure of organizations to convert knowledge into action.4 Their finding was blunt: companies pour resources into accumulating knowledge — training programs, consultants, executive education — and routinely fail at the step of turning that knowledge into something that happens.
In a 1999 Harvard Business Review article, Pfeffer and Sutton identified the specific mechanism. They called it the Smart-Talk Trap: articulate, confident, critical diagnosis that actively prevents action.5 Smart talk has predictable features — eloquence, interesting information, good vocabulary — plus two toxic additions: a default to negativity and unnecessary complexity. “People who want to get ahead in organizations,” they wrote, “learn that talking a lot helps them reach their goal more reliably than taking action or inspiring others to act does.”
Consider how this plays out in the fifteen-minute standup. A developer who says “I’ve identified the root cause as a race condition in the connection pool” sounds more competent than one who says “I spent six hours yesterday fixing a race condition in the connection pool.” The first statement is description. The second is action. The first gets promoted, because it sounds like insight. The second sounds like plumbing.
Meetings, Pfeffer and Sutton argued, are the natural habitat of the Smart-Talk Trap. “Because management today revolves around meetings, teams and consensus building, people are rewarded for talking — and the longer, louder, and more confusingly, the better.” The format itself selects for description over execution. You cannot demonstrate six hours of debugging in a standup. You can demonstrate a crisp, two-sentence diagnosis. And the person who delivers the diagnosis — confident, concise, pointing at the root cause — will leave that meeting having advanced their career more reliably than the person who fixed the thing.
In software engineering, the description-as-substitute pattern found its most durable vessel: the phrase “technical debt.”
Chelsea Troy, writing on the Stack Overflow Blog in December 2023, argued that the term creates false consensus — everyone assumes they know what it means, but their definitions diverge completely.6 When she asked teams to define it, designers described design constraints, product managers described lost feature velocity, and engineers invoked “bad code” — while disagreeing about what makes code bad. “The minute we trot out the term ‘tech debt,’” Troy wrote, “everyone is upset but no one is listening.”
The phrase is a masterwork of attribute substitution. It borrows the authority of financial modeling. It implies that someone has calculated the principal and interest rate. Nobody has. The metaphor is so satisfying that teams use it for years without ever measuring the thing it supposedly quantifies.
Troy observed a telling pattern in tech debt remediation weeks: business leaders watched the result and “saw with their own eyeballs that the last one improved the team’s capacity zero percent.” The problem wasn’t that the work was done badly. The label “tech debt” had aggregated unrelated problems under a single banner, and the resulting “fix” addressed the team’s refactoring preferences rather than the actual bottleneck.
Her proposed alternative is a vocabulary intervention: replace “technical debt” with maintenance load — the proportion of developer time spent on non-feature work. It replaces a metaphor (debt, which invites philosophical debate about whose borrowing caused it) with a measurement (percentage of time, which invites a spreadsheet). Change the noun and you change the verb that follows it.
And then there is the backlog itself. Every ticket filed to address a known problem is, structurally, a description standing in for a fix. The ticket creates an artifact that looks like action: it lives in a system, carries a priority, can be referenced in meetings. But the artifact is a description of work, not the work. The backlog is an institutional archive of descriptions mistaken for progress. As the engineering proverb has it: later means never.
The most consequential version of the substitution operates in clinical medicine, where the gap between naming a problem and fixing it is measured in human outcomes.
In 2021, Sims and colleagues published a systematic scoping review of 146 articles examining the consequences of diagnostic labeling across multiple medical conditions (Frontiers in Public Health, 9:725877).7 The numbers are dissonant. Sixty-one percent of individual-perspective studies reported positive psychological impact from receiving a diagnostic label — primarily relief and validation. Patients valued being believed by their clinician. The label functioned as social recognition: “I have X” becomes a statement that anchors diffuse suffering in shared vocabulary.
But 60% of studies reporting psychosocial consequences also documented negative impacts — increased anxiety, perceived severity, and preference for more invasive treatment. Patients described “relief mixed with fear” upon diagnosis — relief at being named and believed, fear because the condition was now official, permanent, real.
The finding that should stop you: more than a third of studies found that diagnostic labeling was associated with reduced support and service withdrawal. Once the diagnostic decision was made, support was pulled back. The label signaled to clinicians that the descriptive work was complete. In 42% of individual-perspective studies, patients reported negative treatment experiences despite having a diagnosis.
This is the description-action gap stripped of abstraction. A patient receives a diagnosis and feels relief — genuine, measurable psychological relief. The relief is real. But it operates as a substitute for treatment, not a precursor to it. The cognitive ease that comes from having a name for the suffering is, at the neurological level, indistinguishable from the ease of being helped. And in more than a third of cases studied, the label doesn’t just fail to precipitate treatment. It reduces the probability that anyone tries.
The review found one more thing worth naming. Diagnostic labels can create self-fulfilling prophecies: “Therapists’ expectations when they first assess patients will influence the later course of treatment.” The description doesn’t merely substitute for action. It shapes the trajectory of the thing it claims to describe.
Three limits deserve naming — precisely to avoid the irony of an essay about description problems that only describes a description problem.
First, description is a necessary precursor to action. You cannot fix what you haven’t identified. The pathology isn’t that teams describe problems. It’s that they stop there. Diagnosis is the first step; the failure is treating it as the last.
Second, some problems genuinely resist action. In medicine, conditions exist for which no effective treatment is available, and the diagnostic label provides real psychological benefit — the 61% who reported relief were not being irrational. They were being seen. The critique applies when action is available and description displaces it, not when description is the best available intervention.
Third, the organizational version is structural, not characterological. Pfeffer and Sutton were explicit: the Smart-Talk Trap is an incentive problem. Organizations reward diagnosis because meetings reward it, promotions reward it, and performance reviews reward it. Individual resolve does not overcome institutional incentives. The fix has to be structural too.
Pfeffer and Sutton’s most actionable finding was linguistic: companies that close the knowing-doing gap ask “how” questions instead of “why” questions. Not “why is the auth service failing?” — which invites diagnosis — but “how do we stop the auth service from failing by Thursday?” — which requires a plan with a name attached to it.
Troy’s intervention has the same shape: replace “technical debt” (a metaphor that invites debate about who caused it) with “maintenance load” (a metric that invites measurement of what it costs). Both are vocabulary-level changes. They work because they redirect processing from System 1 — label it, feel ease, move on — to System 2 — estimate it, plan it, commit to a date. The smallest possible disruption to the description-action substitution is changing a single word in the question you ask.
The standup from the opening is still happening. That bug is still on the board. But the team that asks “Who is pairing on the connection pool fix this morning?” instead of “What’s the status on the connection pool issue?” will find that the question itself changes what happens next. Not because the words are magic. Because description is a verb, and verbs have directions. Some point at the problem. Some point past it.
If reading this essay felt like progress, you’ve just demonstrated the thesis.
The audit trail is where description stops substituting for action.
The essay’s core finding: naming a problem produces the same cognitive ease as solving it, and organizations systematically reward the naming. When there’s no external record of what actually happened, diagnosis and repair look identical in every retrospective. Chain of Consciousness makes the gap visible — every agent action anchored to a verifiable external record that distinguishes “diagnosed the root cause” from “deployed the fix.” When the record shows description without repair, the substitution is caught at the infrastructure level, not in another retrospective about retrospectives.
pip install chain-of-consciousness · npm install chain-of-consciousness
Hosted Chain of Consciousness → · See a verified provenance chain