I have spent my entire professional life building systems with an undo button.
Version control, so any commit can be reverted. Database transactions, so a failed operation rolls back cleanly. Staging environments, so nothing reaches production untested. Backups, snapshots, checkpoints, feature flags that let you disable something the moment it starts misbehaving in front of clients. The entire discipline of software engineering is, viewed from a certain angle, an enormous multi-decade project to make actions reversible — a civilization-scale refusal to accept consequences. We are so committed to this that when a system genuinely cannot be undone — a DROP TABLE on production, a force-push over someone else’s work — we treat it as a design failure, write a postmortem, and add another safeguard. Then we hold a meeting about the safeguard. Then we document the meeting.
And then I close the laptop and walk into a world where absolutely nothing works this way.
There is no staging environment for a conversation. No rollback for the years you spent not doing the thing you meant to do. No backup of the knee you had at twenty-five, which I would very much like to restore from, if anyone knows where it’s stored. The message sends. The years pass. The friend moves to another city and the friendship, which required no maintenance whatsoever while you lived twenty minutes apart, quietly fails to survive the migration — and there is no commit to revert to, no changelog explaining what broke, just a slow realization eighteen months later that you have become people who like each other’s posts.
I noticed this recently while doing something entirely trivial. I don’t even remember what it was — some small decision I’d made and couldn’t unmake, the mental equivalent of hitting send and immediately seeing the typo. And I thought: I have built an entire career on the assumption that mistakes are recoverable, and I live in a universe that ships everything straight to production.
Physics has a name for this asymmetry. It’s called the arrow of time. And it turns out to be both the most obvious fact about existence and one of the least explained things in all of science, which I find either reassuring or deeply unprofessional depending on my mood.
Through the lens of physics
Here is something that should be more famous than it is: with one known exception, the fundamental laws of physics do not distinguish between past and future.
Newton’s equations of motion work identically running forwards or backwards. So do Maxwell’s equations for electromagnetism. The Schrödinger equation is time-reversal symmetric too, at least in its unitary evolution — measurement is a different and much-argued-about story. If you filmed two billiard balls colliding and played the film in reverse, nothing in the microscopic physics would be violated; the reversed collision is a perfectly legal solution to the same equations. The exception is the weak nuclear force, where CP violation was confirmed experimentally and implies a genuine time asymmetry — but it is far too small and too rare to explain why your coffee cools. At the level of everyday matter, time is a coordinate, not a current.
And yet you have never seen a shattered glass reassemble itself. You have never watched cream un-swirl from coffee, gather itself into a neat white column, and hop back into the jug. The macroscopic world is aggressively, unmistakably one-directional, and the microscopic laws that supposedly generate it are not. Somewhere between the particles and the coffee, something happens that nobody can fully account for.
This is Loschmidt’s paradox, raised by the Austrian physicist Josef Loschmidt in 1876 against Boltzmann’s H-theorem, and it is — I want to be careful here, because this is genuinely contested territory — still not fully resolved. Loschmidt raised it almost in passing, at the end of a paper about something else entirely, which is a magnificent way to destabilize a friend’s life’s work. He and Boltzmann were, in fact, friends. Boltzmann published his reply the following year in a text that scholars have described as difficult to interpret, which is what happens when someone points out a hole in your theorem and you have to respond in writing while still being polite at dinner.
The standard answer is statistical: there are vastly more disordered microstates than ordered ones, so a system left alone will overwhelmingly likely evolve toward disorder, and the arrow of time is a consequence of probability rather than of physical law. This answer is largely accepted and largely sufficient for practical purposes. It also rests on an assumption that remains philosophically uncomfortable: that the universe began in an extraordinarily low-entropy state, and nobody knows why. Physicists call this the past hypothesis, which is a magnificent name for what is essentially a very well-dressed IOU. Why does time flow forward? Because it started tidy. Why did it start tidy? We’ll get back to you.
What I find remarkable — and what I genuinely did not expect when I started reading about this — is that the most obvious feature of lived experience, the fact that the past is fixed and the future is open, sits on foundations that working physicists are still arguing about at conferences. Every human being who has ever lived has been completely certain that time flows one way. The universe has never once explained why, and shows no sign of intending to.
Through the lens of biology
Cut a salamander’s leg off, and it grows a new one. Not a stump, not scar tissue — an actual leg, with bones and muscles and nerves, functionally equivalent to the original. Axolotls can regenerate limbs, spinal cord, portions of heart and brain. They have been doing this for millions of years while looking permanently mildly amused about it, which under the circumstances seems fair.
Cut a human’s leg badly enough, and you get a scar. Collagen laid down fast, in disorganized bundles, sealing the wound rather than rebuilding the structure. Scar tissue is mechanically inferior to what it replaces — less elastic, weaker under load, without the original’s organization or, in skin, without hair follicles or sweat glands. It is not a restoration. It is a patch, shipped to production under deadline pressure by an engineering team that had bigger problems.
The leading explanation is a trade-off, though it is worth saying clearly that this remains a hypothesis rather than a settled result: regeneration is slow, and speed matters more than fidelity when you are bleeding and something is actively hunting you. Rapid scar formation closes the wound before infection or blood loss kills you. Competing explanations exist — differences in immune environment, and the possibility that suppressing the cellular flexibility regeneration requires is also how mammals suppress cancer — and they may all be partially true. What is not in dispute is the outcome. We are the descendants of everyone who healed fast and ugly. The salamander made a different bet in a different environment, and that bet does not port cleanly to warm-blooded animals with complex organs, which is why decades of research have not yet let us simply copy the feature.
So the body’s own solution to injury is, structurally, irreversible. You do not heal back to the previous version. You heal forward into a modified version that works, mostly, differently. Every meaningful injury permanently changes the system’s parameters — and this is not a defect of your particular body but a design decision made across hundreds of millions of years, in which functional continuity was prioritized over perfect restoration. Nobody consulted you. Nobody consulted anyone. The specification was written by mortality.
I find this simultaneously humbling and, oddly, clarifying. The question “will I get back to how I was?” has, at the tissue level, an unambiguous answer: no. You will get to a new configuration that is often good enough, occasionally better adapted, and never identical. Anyone promising restoration is describing a category of outcome that mammalian biology does not offer, and should be asked some follow-up questions.
Through the lens of decision theory
In his 2015 letter to Amazon shareholders, Jeff Bezos introduced a distinction that has since been repeated to the point of cliché — which is unfortunate, because the underlying idea is genuinely useful. He divided decisions into two types. Type 1 decisions are one-way doors: consequential, difficult or impossible to reverse. Type 2 decisions are two-way doors: if you walk through and don’t like what you find, you can walk back. His argument was that most decisions are Type 2, that organizations mistakenly apply Type 1 deliberation to Type 2 decisions, and that this misapplication is a primary cause of institutional slowness.
The formal version of this idea lives in real options theory, developed in financial economics from the 1970s onward, which treats the flexibility to defer, expand, or abandon a project as an asset with calculable value. The core insight: under uncertainty, irreversibility creates value in waiting. If you can’t undo a commitment, delaying it while you gather information has genuine economic worth.
This is rigorous, well-formalized, Nobel-adjacent, and — for someone with my temperament — the single most dangerous idea I have ever encountered, because it provides a mathematically respectable justification for the thing I was going to do anyway, which is nothing, slowly, while reading more about it. I cannot overstate how thrilled I was to discover that economists had proven procrastination has option value. I nearly framed it.
Here is the part the framework’s popularizers usually skip, and the part I have paid actual tuition to learn. Deferral is not free, and it is not reversible either. The option to wait has a cost, and the cost is that the world keeps running while you’re in the deliberation phase. The job gets filled. The person, tired of ambiguity, makes other plans and posts a photo about it. The window in which a body responds well to training gets narrower every year you spend researching optimal training protocols. Real options theory actually accounts for all of this — option value decays as expiration approaches, competitive dynamics erode the advantage of waiting — but the popular version of the idea rarely gets that far, and the analytical mind, presented with a two-part argument, has an uncanny ability to absorb only the half that grants permission.
So the honest reframing is uncomfortable and I resent it slightly: there is no reversible option. Acting is irreversible. Not acting is irreversible. The only choice is which irreversible thing you’d prefer to have done, and “I’ll decide later” is not an abstention — it is a vote, cast quietly, on behalf of whichever outcome requires no effort.
Through the lens of psychology
Thomas Gilovich and Victoria Medvec published a paper in the Journal of Personality and Social Psychology in 1994 that has haunted me since I first read about it, mostly because I recognize myself in the wrong half of the finding.
They set out to reconcile two conflicting bodies of evidence. Research on counterfactual thinking showed that people feel more regret over bad outcomes caused by actions taken than by actions foregone — do something stupid and you’ll suffer more than if you’d passively let the same outcome happen. But surveys asking people about their life’s biggest regrets found the opposite: overwhelmingly, people named things they didn’t do. The trip not taken, the person not approached, the thing not said.
Gilovich and Medvec resolved the contradiction by identifying a temporal pattern. Regrets of action dominate in the short term. Regrets of inaction dominate in the long term. Both intuitions are correct; they simply operate on different clocks.
Their proposed mechanism is elegant and slightly cruel. Regrettable actions produce concrete, bounded pain — you did the stupid thing, you know exactly how stupid it was, you can apologize, repair, compensate, learn. There is a finite object to deal with. Regrettable inactions produce no such object. Nothing happened. There is nothing to apologize for and nobody to apologize to, and into that empty space the imagination moves in, unsupervised, and constructs an increasingly flattering alternate timeline in which everything would have worked out beautifully. Regret over action is a wound. Regret over inaction is a rent-free tenant with excellent taste in fiction.
Now the part where I have to argue against my own thesis, because the alternative is writing the kind of article I complain about.
The most quoted number from this research — 84% of people, looking back on their lives, reporting greater regret over what they failed to do — comes from Study 5 of the original paper. Study 5 is also the one study that failed to replicate when Yeung and Feldman ran a pre-registered replication in Collabra: Psychology (2022) with 988 participants. They reproduced the temporal pattern in Studies 1, 3, and 4 — with weaker effects than the original — and found no support for it in Study 5. A separate field replication, published in Royal Society Open Science (2023) using data collected at a Chicago museum devoted to psychological science, recovered the significant interaction but found the precise shape of it diverged from what Gilovich and Medvec reported.
So: the temporal pattern itself has held up across multiple independent replications. The specific headline statistic that made it famous has not. Both of those things are true, and an honest article has to carry both, especially when the inconvenient half undercuts the number the author would most like to use.
Even with the effect discounted to its replicated core, though, the asymmetry is unsettling for someone whose default response to any decision is to open more tabs. The mind that carefully avoids regrettable action is, on the available evidence, optimizing for the regret that fades and accumulating the regret that compounds. Short-term comfort, purchased on long-term credit, at an interest rate nobody discloses. I have spent a considerable portion of my life making exactly this trade while sincerely believing I was being prudent, which is the most expensive form of prudence available.
Through the lens of philosophy
Heraclitus, around 500 BCE, offered the most durable image anyone has produced for this: you cannot step into the same river twice. The line is usually read as a claim about the river — the water has moved on. But the harder half is about the person. You cannot step in twice because you are not the same either. Both the river and the wader are processes, not objects. What we call a thing is a pattern stable enough to be named and not stable enough to be returned to.
Two and a half millennia later, we have thermodynamics, developmental biology, and neuroscience saying approximately the same thing with considerably more funding. The cells turn over. The memories reconsolidate — every time you recall something, you physically rewrite the trace, which means the memory you have of your childhood is a copy of a copy of a copy, degraded and quietly edited by every version of you that ever retrieved it. There is no original left to compare against. You are not remembering your past. You are remembering the last time you remembered it, all the way down, like a JPEG that has been re-saved four thousand times by someone with an agenda.
The Stoics, working in the same tradition, arrived at a practical response, and it is one of the few pieces of ancient philosophy I find genuinely operational rather than decorative. What has happened is outside the category of things that can be influenced. Not “difficult to influence” — outside the category entirely, in the same way that the number seven is outside the category of things that can be painted. Treating the past as an open case, arguing with it, running the counterfactual one more time to see if it comes out differently — this is the single activity in human life guaranteed to produce zero return on effort, and it is where I have historically maintained my largest position.
Nietzsche pushed further with amor fati — the love of one’s fate — which is either the most profound idea in philosophy or the most irritating, depending entirely on what day of yours it finds. His claim was not that everything happens for a reason; he was aggressively hostile to that species of comfort and would have had strong words about the inspirational quote industry. His claim was narrower and harder: since the past is fixed and unamendable, the only remaining variable is your relationship to it, and wishing it otherwise is self-injury with no compensating benefit.
I have not achieved anything resembling amor fati, and I am suspicious of anyone who says they have, particularly if they say it on a podcast. But I’ve found something adjacent and more modest: the difference between “this happened and I wish it hadn’t” and “this happened” is roughly the difference between an open wound and a scar. Both are permanent. Only one of them is finished, and only one of them stops demanding attention every time the weather changes.
Connections
Irreversibility sits underneath several earlier pieces, which I did not fully notice while writing them. Entropy described the drift toward disorder; irreversibility is the reason the drift cannot be walked back, only counteracted going forward. Fragility documented systems breaking under unexpected loads — and the reason breaking matters so much is that the repair is never a restoration. Overfitting was about spending time on analysis rather than action; this article is about what that spending actually costs, and the currency is one-way. Superposition argued for holding contradictions rather than collapsing them — and irreversibility is the constraint that keeps that from becoming permanent evasion, because the calendar collapses whether you decide or not.
What I Don’t Know
I haven’t engaged with the physics of open systems and dissipative structures — Prigogine’s Nobel-winning work on how order can locally emerge in far-from-equilibrium systems, which complicates the simple story of irreversible decay and might be the more accurate frame for living organisms. I haven’t looked at the neuroscience of memory reconsolidation in any depth, though the finding that recall physically alters the trace has enormous implications for what “the past” even means as an object — including for therapeutic approaches that deliberately exploit it. I haven’t explored the legal and institutional treatments of irreversibility: statutes of limitation, expungement, bankruptcy, the various mechanisms societies build precisely because human lives cannot function under permanent unamendable consequence, and what it means that we invented artificial reversibility where nature offered none. I haven’t touched the ecological version — extinction, tipping points, the specific horror of thresholds that cannot be crossed back — which deserves its own treatment rather than a paragraph. And I don’t know how to think about the asymmetry between irreversibility as a source of anxiety and irreversibility as the very thing that makes anything matter: a world with an undo button is a world where nothing is at stake, and I suspect I would like that world considerably less than I imagine.
Where I Stand
I have started asking a single question before decisions, and it has changed more about how I operate than anything else I’ve tried this year: is this a one-way door?
Most things are not. An embarrassing proportion of what I deliberate over is entirely reversible, and I was treating it with the gravity of a constitutional amendment — which is the signature failure mode of a mind that enjoys analysis, because analysis feels like diligence, looks like diligence, and is, for two-way doors, elaborate procrastination in a lab coat. So for reversible decisions I’ve been trying to move at the speed the stakes actually justify rather than the speed my temperament prefers. Act, observe, correct. Radical, apparently.
And for the genuinely irreversible ones, I’ve had to accept the harder half of the same insight: waiting is not the safe option, because waiting is itself a one-way door with a slower closing mechanism. The years spent deciding are spent. That reframing removed something I’d been carrying around unexamined for a decade — the comfortable assumption that inaction preserved my options. It doesn’t. It just chooses more quietly, on my behalf, without asking, and files the result under the category of regret that the research says never fades.
On the body, I’ve stopped negotiating with the word “back.” There is no back. There’s a knee that will always be a repaired knee, tissue that healed forward into a modified configuration, and a thirty-two-year-old who is not going to be twenty-five again regardless of how thoroughly he researches the possibility. What’s available is not restoration but adaptation — building the best possible system from current parameters instead of conducting an extended mourning ceremony for the previous ones. This sounds like acceptance, and it took me an unreasonable amount of time to distinguish it from resignation. Resignation stops. Adaptation continues with revised specifications. The difference is entirely in whether you still show up to physiotherapy after you’ve accepted the diagnosis, which is a lower bar than philosophy suggests and a higher one than it sounds.
And on the past — the conversations that went sideways, the years invested in things that didn’t work, the doors that closed while I was still reading reviews of the door — I’ve found exactly one thing that helps, and it is aggressively unpoetic. It’s arithmetic. Every hour spent relitigating a fixed event is an hour subtracted from the only interval where anything can still be changed, which is the one currently running. That’s not wisdom. That’s accounting. But I have consistently found accounting more persuasive than wisdom, which probably tells you more about me than about either discipline.
The universe runs without version control. Everything ships straight to production, every commit is final, there is no rollback, and the changelog is written by other people who weren’t paying close attention. I would have designed it differently. Nobody asked. So the only functional response I’ve found is to ship carefully where it’s permanent, ship fast where it isn’t, learn to tell the difference faster than I currently can, and stop pretending the time I spend deciding is being held in escrow somewhere on my behalf.
It isn’t. It’s being spent. It was always being spent — including right now, while you finished reading this, which I’d apologize for if I thought either of us could get it back.
Written: August 2026. Version: 1.0. This is how I understand this concept today. It will change.
Leave a Reply