Cargo Cult Everything: When Mimicry Replaces Mechanism

In 2025, U.S. businesses spent up to $40 billion building artificial intelligence systems and got almost nothing back. An MIT-affiliated analysis circulated by Cogent Executive in mid-2025 estimated that roughly 95% of those AI initiatives produced no measurable return. The same year, 78% of organizations reported using AI in at least one business function — up from 55% in 2023 — while 42% admitted they had abandoned most of their AI projects, more than double the 17% abandonment rate from a year earlier (industry surveys aggregated by Fullview.io, 2025). Adoption surged. Returns flatlined. The two curves diverged in real time.

This is a familiar shape. The dashboards exist. The AI teams exist. The "AI-first" branding is on every careers page. What's missing is everything those things were supposed to make possible — clean data, integrated workflows, problems worth automating. The form is meticulously reproduced. The mechanism that was supposed to give it meaning is absent.

We have a name for this. It's old, awkward, contested, and stubbornly useful: cargo cult.

Tanna, 1942

The literal cargo cults arose in Melanesia in the early twentieth century and intensified when Allied and Japanese forces airlifted manufactured goods — clothing, medicine, weapons, canned food — to islands whose communities had often never seen outsiders. When the war ended and the planes stopped landing, some communities built bamboo runways and wooden control towers, lit fires along improvised landing strips, and waited. The most famous is the John Frum movement on Tanna in Vanuatu, which still observes February 15 as John Frum Day decades on (Smithsonian Magazine; the term itself first appeared in print in Pacific Islands Monthly, November 1945).

The pop-cultural version of this story — primitive natives mistaking form for function — has been getting harder to defend. In a January 2025 essay at righto.com, Ken Shirriff argued that much of the canonical cargo-cult imagery was likely fabricated or staged, citing the 1962 "shockumentary" Mondo Cane as a probable source for both the airfield iconography and Feynman's 1974 framing. A peer-reviewed paper by Andrew Gelman and Megan Higgs in Theory and Society the same year ("Interrogating the 'cargo cult science' metaphor," Springer 2025) made the more careful version of the argument: the metaphor smuggles in colonial assumptions about who counts as rational. Gelman's proposal is to drop "cargo cult" in favor of "ritual science."

He is making a fair point. But the metaphor's force, when used honestly, comes from reversal — Western institutions repeating the very pattern they once condescendingly pinned on Pacific islanders. The colonial gloss is the part to drop. The mechanism the metaphor describes is durable.

Feynman's 1974 Caltech Address

Richard Feynman's commencement address at Caltech extended the metaphor to scientific practice and fixed its modern meaning. His passage is worth quoting because it does the work in five sentences:

"They've arranged to make things like runways, to put fires along the sides of the runways, to make a wooden hut for a man to sit in, with two wooden pieces on his head like headphones and bars of bamboo sticking out like antennas — he's the controller — and they wait for the airplanes to land. They're doing everything right. The form is perfect. It looks exactly the way it looked before. But it doesn't work." (Caltech Library, "CargoCult.htm")

What was missing, in Feynman's reading, was "a kind of utter honesty" — specifically, the discipline of looking for the ways your own approach might be wrong. The replication crisis that erupted in psychology after 2011 turned out to be his case study, delivered four decades after his speech: studies that passed every formal test of scientific rigor and didn't replicate. P-hacking. HARKing. Selective reporting. The form was perfect. It didn't work. More than 140 psychology journals have since adopted result-blind peer review — accepting studies based on methods before any data is collected — a structural attempt to reinstate the mechanism that the form had quietly displaced.

Why It Keeps Happening

The cognitive science here is unflattering: form-without-function copying is not a failure of intelligence. It's the default. Developmental psychologists call it overimitation — the tendency to copy all of an adult's actions, including the obviously unnecessary ones. The counterintuitive finding is that overimitation increases with age. Five-year-olds overimitate more than three-year-olds. Adults still do it under most conditions tested (PNAS 2007, "The hidden structure of overimitation"; Hogrefe, Social Psychology Vol. 43, No. 4).

A 2025 paper by Dragon and Poulin-Dubois in Frontiers in Developmental Psychology refined the picture. At 16-21 months, infants overimitate at low rates; the behavior strengthens with age and becomes coupled to in-group preference around age 4½. We start by copying because we don't know what else to do, and only later does the copying become a way to belong. The "copy-all/refine-later" hypothesis explains the evolutionary logic: in most learning environments, mechanisms are genuinely opaque, and high-fidelity imitation is a wildly efficient strategy. You can learn to make a working fishing trap without first learning fluid dynamics.

Two findings make this directly applicable to organizations. First, adults overimitate prestigious models far more than peer models — when the demonstrator is believed to be a fellow participant rather than an authority, fewer than half overimitate (Hogrefe). Second, mimicry from outsiders to insiders can backfire — being mimicked by a member of an outgroup actually makes you like that outgroup member less (PMC, "Where is the love? The social aspects of mimicry"). The corollary: hired-gun consultants and FAANG transplants who arrive with a playbook and start performing the rituals often damage trust before they damage outcomes.

There is at least one piece of hopeful evidence. A 2024 paper in the Journal of Experimental Child Psychology found that Montessori-schooled children show a lower propensity to overimitate than traditionally-schooled peers. Pedagogy moves the needle. Whatever Feynman meant by "utter honesty" can apparently be taught.

The Pattern, Across Domains

Once you have named the shape, it is hard to un-see.

Software engineering. Fischer et al. (IEEE Symposium on Security and Privacy, 2017) analyzed Android applications on Google Play and found that 30% of cryptographic code examples on Stack Overflow were insecure — and that those insecure patterns had propagated to over 190,000 production applications, including apps in security-sensitive categories. The form was a code snippet that compiled and ran. The mechanism — what the code was actually supposed to defend against — was nowhere visible in the snippet, and the developer copying it almost never went looking.

Management. The General Motors–Toyota NUMMI joint venture (1984-2010) is the canonical case. Toyota turned GM's worst plant, with the same union workforce, into one of its best, using the Toyota Production System. GM's response, as recounted by a former employee on This American Life's 2015 NUMMI episode, was to send people in with cameras: "I want you to go there with cameras and take a picture of every square inch. And whatever you take a picture of, I want it to look like that in our plant." Fifteen years after NUMMI opened, GM had still not successfully implemented lean manufacturing across its other U.S. plants (Harvard Business School, "Knowledge Transfer: Toyota, NUMMI, and GM"). The factory layout, the andon cords, the visual management boards — all photographable. The culture of mutual respect and continuous improvement that made any of it work — invisible to cameras.

Agile. Scrum's industrial-scale failure mode now has its own academic literature. A 2025 paper in Information and Software Technology developed a diagnostic instrument specifically for detecting cargo-cult patterns in Agile adoption. You know the symptoms: standups that have become status reports, Jira boards that perform transparency without enabling it, project managers renamed product owners without changing what they do. The most damaging consequence is the second-order one — failed Agile teaches the organization that "we tried agility and it didn't work," which inoculates against the real thing.

Cybersecurity. Phil Venables, formerly CISO at Goldman Sachs and Google Cloud, catalogued twelve categories of "ceremonial security" — controls that started as effective practices and ossified into rituals performed because "we keep doing these things." Forced periodic password changes, which NIST itself reversed in SP 800-63B. Mandatory annual security awareness training that satisfies compliance without measurably reducing incidents. Post-mortems that produce reports rather than root causes. When the form fails, organizations blame execution. The mechanism never gets sought.

Governance. Lant Pritchett, Matt Andrews, and Michael Woolcock formalized the development-economics version in 2013, calling it "isomorphic mimicry": developing nations adopt the visible forms of successful Western institutions — constitutions, regulatory agencies, anti-corruption commissions — without the underlying capacity to make them work. Donors measure form (does the agency exist?) not function (does it prosecute?). Schools get built but children don't learn. IT systems are introduced but not used. Plans are written but not implemented.

Leadership. James McElroy's American Affairs essay "Cargo Cults and the Disorganization of America" (Spring 2025) catalogues the more recent corporate version. Boeing's 737 MAX cycled CEOs through P&G, McKinsey, and GE — process-optimizers rather than engineers — while the stock hit all-time highs and the planes had a fatal design flaw. Ernst & Young promoted 1,033 associates to "partner" while stripping the equity that made the title mean something. Job title inflation across early-career tech roles has roughly tripled the use of "lead" since 2019 and increased "principal" by about 57%, while "junior" titles have dropped by half. The form of seniority. Without the mechanism.

A Hierarchy of Mimicry

Not all mimicry is the same kind of failure. A useful hierarchy — from least to most pathological — looks roughly like this:

The critical threshold is between Levels 2 and 3 — the moment the imitator's goal flips from understanding the mechanism to appearing to possess it. Levels 4 and 5 are self-reinforcing: the surface legitimacy that mimicry produces removes the very pressure that would otherwise force genuine capability development. This is the ratchet, and it locks in.

Where the Argument Is Weakest

Three honest objections.

"The form sometimes is the mechanism." Austin's How to Do Things With Words (1962) showed that some speech acts constitute the act they describe — saying "I promise" makes the promise. Some organizational rituals are like that. Wearing the uniform creates the discipline; running the standup, even badly, sometimes creates the conversation. This is real. The argument-from-mimicry holds only when the causal pathway from form to function genuinely requires hidden infrastructure — culture, capability, trust — that the form alone cannot summon.

"Mimicry is a necessary first step." It is. Scaffolded mimicry (Level 1-2) is how apprentices become experts. The cognitive apprenticeship framework (Collins, Brown, & Newman, 1989) puts modeling first among its four core moves. The danger is when mimicry optimized for legitimacy displaces mimicry oriented toward learning. GM photographed NUMMI for 15 years. The photographs were Level 1. The behavior was Level 4.

"The metaphor is colonially condescending." This is the most serious of the three, and Gelman and Higgs (Theory and Society, 2025) make it persuasively. The original Melanesian movements were rational responses to genuinely unprecedented events within the frameworks available; calling them "cults" strips both context and agency. The right response, I think, is to keep the diagnostic value of the term while changing whom it is aimed at. The metaphor's bite comes from turning it on the people who used to deploy it dismissively. That is the move that earns it the right to keep existing.

What to Actually Do

The reliable antidote is not the avoidance of mimicry. We cannot avoid it; it is how we learn. The antidote is preserving the feedback loop that connects form to function long enough for the form to start meaning something.

A few practical translations for developers and tech leaders:

1. When you copy a pattern, write down what you think it is doing. A code snippet, a Scrum ritual, a design pattern. Write down the mechanism you believe is operating. Then have someone who knows the original tell you whether you are right. This is the apprenticeship move, applied unilaterally.

2. Treat unbroken mimicry as a smell. If a practice has worked smoothly for years and no one can explain why — what it is defending against, what it is optimizing for — the mechanism may have decayed faster than the form. The form is still standing. The function may have left the building.

3. Refuse to measure form when you can measure function. "Number of standups attended" is a form metric. "Did anyone change what they were going to do based on what they heard?" is a function metric. Function metrics are harder to gather and easier to game, which is exactly why they are worth gathering carefully.

4. Watch for immunization. The most damaging cargo cult is the one whose failure teaches the organization that the original idea does not work. If you are inheriting a system that someone tells you "we tried that and it failed," the next question is: did you try it, or did you try its photograph?

5. Do not cargo-cult the anti-cargo-cult. Paul Graham's "Founder Mode" essay (2024) was an explicit rebellion against management mimicry. Companies are now mimicking Founder Mode. Design thinking workshops, first-principles offsites, blameless post-mortems — every one of these is a real practice that can be photographed, photocopied, and dropped into an organization that has never asked why it works. The form will look perfect. It will not work.

The thing Feynman put at the bottom of all this in 1974 is still sitting at the bottom: the willingness to report the results that make your own approach look bad, to look hard for the errors that would invalidate your conclusions, to lean over backwards toward honesty. That is not a form. There is no photograph of it. It is the mechanism. It always was.