At my workplace, there’s a gate in the courtyard. It has a lock. It has a key. For a brief, shining period in institutional history, it worked exactly as gates are supposed to work: it stayed locked, and people with keys used them. I understand this era has ended elsewhere as well.
Then someone left it unlocked. It was more convenient that way — one fewer step between the parking lot and the door, and in the landscape of modern office life, one fewer step is approximately the highest currency we trade in. Within a week, everyone was using the unlocked gate, and the key had entered a quantum superposition where it simultaneously existed and did not exist, depending on whom you asked and what mood they were in. Some colleagues said we had a key. Others said we didn’t. Some said the gate locks. Others said it doesn’t. The gate, lacking opinions of its own and presumably unbothered by epistemology, remained open.
Then someone stole the ashtrays from the courtyard.
This prompted a brief period of institutional concern, which is the phase between “nothing is wrong” and “nothing can be done.” A key was found. The gate was locked. It stayed locked for two days — a personal record for the organization, possibly a regional one. Then, because two days of mild inconvenience is approximately the half-life of human discipline, someone unlocked it again. The solution, arrived at after what I can only assume was significant deliberation by people paid significantly more than I am, was to install fake security cameras.
Not real cameras. Fake ones. Decoys. The organizational response to a security problem caused by people refusing to use a lock was to install a theatrical representation of surveillance rather than make copies of the key and distribute them.
I work in AI. I think about systems for a living. And I’m telling you, if you want to understand how the universe actually works — not how we wish it worked, but how it actually, measurably, observably works — you don’t need thermodynamics textbooks. You need to watch what happens to a gate key in a mid-sized company over the course of a month.
The word for this is entropy. And it explains more than it has any right to, which is saying something, because entropy explains quite a lot.
Through the lens of thermodynamics
The second law of thermodynamics states that in an isolated system, entropy — the measure of disorder, or more precisely, the number of possible microscopic configurations consistent with a system’s observed macroscopic state — never decreases. Ludwig Boltzmann formalized this in the 1870s with his statistical mechanics, which was such a disturbing insight to his contemporaries that he was professionally hounded for most of his career and eventually, tragically, hanged himself in 1906. Science does not always reward the messenger, especially when the message is “everything tends toward mess.”
Boltzmann’s point was this: entropy isn’t a mystical preference the universe has for chaos. It’s a consequence of probability. There are astronomically more ways for things to be disordered than ordered. A shuffled deck has roughly 8 × 10⁶⁷ possible arrangements and exactly one that’s perfectly sorted by suit and rank. Order is not forbidden. It is simply so statistically unlikely as a spontaneous outcome that expecting it is, in a rigorous mathematical sense, unreasonable. The universe is not out to get you. It is just committed to its statistics.
Arthur Eddington, in The Nature of the Physical World (1928), gave entropy what amounts to diplomatic immunity among physical laws: if your theory contradicts the second law, he argued, there is nothing for it but to collapse in deepest humiliation. Among physicists, this is considered gentle. Among the rest of us, it sounds like something you’d say to a colleague you were trying to get fired.
What makes the second law genuinely interesting — beyond predicting that your desk will get messy, your car will break down, and your enthusiasm for this article will fluctuate — is that it defines the arrow of time. Entropy is the reason the past is distinguishable from the future. Eggs break but don’t unbreak. Coffee cools but doesn’t spontaneously heat. Relationships fray faster than they form. The second law is the only fundamental law of physics that has a preferred temporal direction. Everything else — gravity, electromagnetism, quantum mechanics — works the same forwards and backwards. Only entropy knows which way time goes, which is a lot of responsibility for a concept that mostly manifests as your kitchen.
My gate went from locked to unlocked in one day. It went from unlocked to locked in a burst of organizational willpower that lasted forty-eight hours. The asymmetry is not a coincidence. It’s the second law, applied to a courtyard in Brașov.
Through the lens of organizational theory
In 1988, James March published an article in Management Science called “The Technology of Foolishness,” arguing that organizations are not rational systems that occasionally malfunction but fundamentally ambiguous systems that occasionally produce the appearance of rationality. This is an underappreciated insight, possibly because people who work in organizations find it too accurate to be comfortable reading.
March, together with Johan Olsen, developed the “garbage can model” of organizational decision-making in a 1972 paper in Administrative Science Quarterly. The model proposes that organizational choices arise not from deliberate analysis but from the semi-random collision of four independent streams: problems, solutions, participants, and choice opportunities. The “decision” is whatever falls out when these streams happen to intersect. They called it the garbage can model because that is, in fact, the mental image. This is one of the most cited papers in management science. It is also, to anyone who has ever worked anywhere, completely unsurprising.
The gate situation maps onto this with a precision that would be flattering if it weren’t so depressing. The problem (courtyard security) and the solution (fake cameras) arrived independently of each other. The participants who made the decision were not the same participants who had the key. The choice opportunity was the theft of the ashtrays. Nobody sat down and said, “What is the most logical response to our security vulnerability?” Someone said, “We should do something,” and the most available something was fake cameras, which had the dual advantage of being cheap and requiring no behavioral change from anyone. Behavioral change is, in organizational terms, more expensive than almost anything else, including actual cameras.
This is how entropy works in organizations. The system doesn’t degrade because people are incompetent — or not only because of that. It degrades because maintaining order requires continuous energy input — policies enforced, norms upheld, keys distributed, uncomfortable conversations had — and the moment that energy input drops below a threshold, the system relaxes toward its natural low-energy state. Which, in the case of a gate, is unlocked. In the case of a team, is siloed. In the case of a civilization, is harder to characterize on a Tuesday morning but presumably involves worse roads.
Through the lens of psychology
Michael Gazzaniga and Joseph LeDoux identified what they called “the left-brain interpreter” in 1978, based on experiments with split-brain patients at the California Institute of Technology — patients whose corpus callosum had been surgically severed to treat severe epilepsy, which is itself a reminder that neuroscience owes a lot of its foundational insights to people who signed up for surgery and got a research career as a side effect. In the most famous demonstration, they presented a snow scene to a patient’s right hemisphere and a chicken claw to the left hemisphere, then asked the patient to select related images with each hand. The left hand (controlled by the right hemisphere) pointed to a shovel. The right hand (left hemisphere) pointed to a chicken. When asked to explain his choices, the patient — whose left hemisphere had no access to the snow scene — didn’t say “I don’t know why my left hand picked the shovel.” He said, with complete confidence, that the shovel was for cleaning the chicken coop.
The left hemisphere fabricated a coherent explanation instantly, believed it completely, and moved on with its day. Gazzaniga called this module “the interpreter” — a system whose function is not to determine why you did something, but to construct a plausible narrative after the fact. It’s not a bug. It’s the headline feature.
We all do this. Not occasionally — constantly. The colleague who argued passionately for shipping fast on Monday and for code quality on Wednesday isn’t lying on either day. He has two narratives about himself — “I deliver results” and “I care about craft” — and his interpreter activates whichever one the current social context rewards. He is not one consistent character. He is multiple locally coherent stories that contradict each other globally, which is also, I should admit, a reasonably accurate description of me. And probably of the person reading this. The interpreter is democratic. It comes standard.
This is psychological entropy: the natural state of human identity is not a unified narrative but a shifting, context-dependent set of partial stories. Consistency is the exception — a temporary, energy-intensive state that requires constant maintenance and usually therapy. Stop maintaining it, and the self reverts to its default: a committee of voices that don’t always agree, rarely take minutes, and occasionally issue press releases contradicting what last Tuesday’s committee said.
Through the lens of ecology
In ecology, the concept of succession describes how an ecosystem evolves over time after a disturbance. A cleared field doesn’t stay cleared. First, fast-growing weeds colonize the open ground — nature’s equivalent of the person who joins the new project first and then complains it has no structure. Then grasses arrive, then shrubs, then trees. Each stage changes the conditions enough to enable the next. Left entirely alone — no mowing, no farming, no human intervention — the field will converge on what ecologists call a “climax community,” a relatively stable state determined by local climate and soil.
The critical insight is that the climax state is not the most ordered state. It’s the state that requires the least energy to maintain. It’s the ecosystem’s thermodynamic compromise — complex enough to be self-sustaining, but no more complex than it needs to be. Ecosystems, it turns out, are more disciplined than most startups.
Human organizations work the same way, just faster and with worse meetings. A freshly instituted policy — lock the gate, write tests, document your code, no Slack after 7pm — is a disturbance. A sudden injection of order into a system that was happily coasting at lower entropy. If no one continuously maintains the policy, the organizational ecosystem will undergo succession: first, exceptions (“just this once”); then, workarounds (“it’s easier if we skip that part”); then, the workarounds becoming the new norm (“we never really did that anyway”); and finally, a stable state where the policy exists on paper and is ignored in practice. This is not decay. It’s equilibrium. The policy was the anomaly.
This is not cynicism. It is ecology. The gate is not locked because nobody is supplying the energy required to keep it locked. The fake cameras represent the organizational equivalent of a scarecrow — a low-energy simulation of order that satisfies the need to have done something without actually changing the system’s equilibrium state. Scarecrows, to be fair, at least move in the wind.
Through the lens of philosophy
Albert Camus, in The Myth of Sisyphus (1942), identified what he considered the fundamental philosophical problem: the collision between our demand for meaning and the universe’s total indifference to that demand. He called this collision “the absurd” — not a property of the world, not a property of the mind, but the relationship between the two. We want order. The universe doesn’t want anything. The gap is permanent and cannot be resolved, which Camus considered important enough to write 150 pages about, because the alternative was apparently to do nothing, and Camus was French.
Camus explicitly rejected the existentialist label often applied to him — a rejection ignored by approximately every philosophy syllabus since. He considered the existentialist responses to absurdity — Kierkegaard’s leap of faith, Heidegger’s ontological framework — to be forms of what he called “philosophical suicide”: abandoning the tension instead of living within it. His alternative was revolt, freedom, and passion — not revolt against any particular injustice, but revolt against the condition itself. His concluding metaphor, Sisyphus condemned to roll a boulder up a hill for eternity, is offered not as despair but as defiance. The struggle itself is enough. One must imagine Sisyphus happy, Camus insists, though he does not provide instructions.
I think Camus is right about the universe and incomplete about humans. The absurd isn’t only out there — in the indifferent cosmos, in the heat death, in the entropic decay. It’s also in here, in the person trying to revolt. The person revolting on Monday is not the same person on Wednesday. Sisyphus rolling the boulder up is not the same Sisyphus watching it roll down. The myth assumes a stable protagonist. The reality is that the protagonist is also subject to entropy — inconsistent, forgetful, occasionally distracted by his phone, occasionally installing fake cameras instead of distributing keys.
Perhaps the real absurdity is not that the universe is indifferent. It’s that we’re not even consistent enough to be properly outraged about it. The gods should have cursed Sisyphus with a slightly different punishment: roll the boulder up, watch it roll down, then try to remember why you were doing that in the first place.
Through the lens of remote work
A 2024 meta-analysis by Gajendran and colleagues, published in Personnel Psychology, found that remote work operates through a dual pathway: it simultaneously increases autonomy (which improves satisfaction) and decreases social connection (which increases isolation). These aren’t opposing findings from different studies. They’re happening to the same people at the same time, which is exactly the kind of thing the human mind is terrible at processing, because we prefer stories where the protagonist is either winning or losing and not both in alternating slides.
A study using the 2024 U.S. Household Pulse Survey — 87,317 respondents, nationally representative — found that employees working remotely three or more days per week had significantly higher odds of experiencing loneliness compared to those who never worked remotely. The effect persisted after adjusting for demographic variables and other social isolation indicators. This is the kind of finding that would have been surprising in 2019 and is now, in 2026, considered a fact so obvious that stating it feels faintly apologetic.
I see this daily. Remote work was supposed to give people freedom — flexible hours, no commute, autonomy to organize their own time. And it did, to the extent that flexibility is freedom, which is further than I would have guessed. But freedom without structure is just another word for entropy. Each person, freed from the synchronizing effect of physical proximity, drifts into their own orbit. Communication degrades from real-time conversation to asynchronous messages to occasional silence. Shared context evaporates. People build private mental models of what the team is doing, and those models quietly diverge until someone says “I thought we agreed on X” and someone else says “We never agreed on X” and both are right, because in the absence of continuous information exchange, agreement is just a snapshot that expired weeks ago and was never formally filed.
The irony that kills me: the freedom was designed to help people. Flexible policies, no rigid rules, trust the individual. It comes from a genuinely good place — the recognition that treating adults like adults generally produces better results than treating them like office furniture. But a system with no constraints is a system at maximum entropy. The gate stays unlocked because unlocking it was framed as freedom. The fake cameras go up because the cost of actually enforcing the lock — distributing keys, creating accountability, having the conversations nobody wants to have — exceeds the organization’s appetite for friction, and the organization, like all systems, prefers friction-free failure to friction-ful success.
We confused the absence of structure with the presence of trust. They’re not the same thing. Trust is a low-entropy state maintained by continuous investment, vulnerability, and the willingness to enforce uncomfortable norms. The absence of structure is just… entropy wearing a PR team.
Connections
If Overfitting was about imposing too much order — optimizing so aggressively that you fit to noise — then Entropy is the opposite problem: what happens when you stop imposing order at all. They’re the two failure modes of the same system. One is too much signal extraction; the other is too much noise acceptance. Overfitting and entropy are, in this sense, the Scylla and Charybdis of anyone trying to live deliberately, except there are also many other monsters and the ship is on fire.
Entropy connects to Fragility because fragile systems are, among other things, systems that have optimized away the redundancy that would have buffered them against entropy’s arrival. Removing structural margin to improve efficiency is a way of betting that entropy will be polite and announce itself before breaking things. Entropy is not polite. Entropy also connects to Enough because the question “when have I imposed enough order?” and “when have I let things slide into enough disorder?” are the same question asked from opposite sides of the same problem. And it connects to future pieces — Noise vs. Signal, Attention, Calibration — because each of them is, at its core, about trying to distinguish meaningful pattern from the background disorder that everything eventually turns into.
What I Don’t Know
I haven’t explored how the Buddhist concept of anicca — the doctrine that all conditioned phenomena are impermanent — frames entropy not as a problem to be solved but as a foundational truth, which would make this entire article a 2,000-word elaboration of something a monk could say in four syllables. I haven’t looked at this through chemistry, where Le Chatelier’s principle describes how equilibrium systems resist disturbance — a kind of molecular stubbornness that might explain why institutional change feels the way it does, which is roughly the way uphill feels. I haven’t examined entropy in families across generations — the way values, stories, and traumas degrade and mutate as they transmit from parent to child, like a game of telephone played over decades with emotional stakes and no referee. I haven’t looked at the political dimension, though I have my suspicions — specifically, that the appeal of authoritarian leadership is often an entropy-panic response, a population so overwhelmed by disorder that it will trade freedom for the appearance of structure. Which is, if you think about it, the political equivalent of fake cameras, except the cameras vote. And I haven’t engaged with information theory here, where Shannon’s entropy measures uncertainty in communication and might formalize why every message we send gets a little more lost in the noise as it travels.
Where I Stand
I’ve stopped trying to resolve the tension between wanting coherence and living in a universe that doesn’t provide it. I’ve started trying to notice the tension instead, which turns out to be a more sustainable practice and a less exhausting one.
Concretely: I maintain the systems I can maintain, and I’ve made peace with the ones I can’t. I no longer consider this a character flaw — I consider it statistics. I show up to work knowing that the organizational entropy I described in this article will continue regardless of my feelings about it, and I contribute the energy I have toward the parts that matter most, without pretending I can maintain order everywhere simultaneously. That’s not resignation. That’s triage. And triage, done honestly, is more useful than the illusion of total control.
The desire for coherence — in my work, in the people around me, in myself — is probably both a noble human trait and a cognitive limitation, simultaneously and without contradiction, which is exactly the kind of answer the universe would give if you asked it directly. I consider myself, on balance, a reasonably well-maintained local decrease in entropy temporarily resisting my environment, which is approximately the best any of us can hope for and measurably better than I was doing a year ago, when I thought the right response to chaos was to analyze it into submission rather than to walk through it with my eyes open and my expectations calibrated.
The gate at work is still unlocked. The cameras are still fake. But I’ve bought my own key, and I use it, and I’ve stopped expecting the system to match my preferences. Some entropy you fight. Some you navigate. The skill is learning which is which — and doing both without losing your sense of humor about it, because the moment you stop finding it funny, you’ve lost the one advantage you actually had.
Written: April 2026 Version: 1.0 This is how I understand this concept today. It will change.
Leave a Reply