Crash-Only Thinking

A few weeks ago, I learned about something called crash-only software  (ht, Robert Greco). This is software that has no normal “start” or “stop” mechanisms. It can only be stopped by crashing it. Often this means unplugging the computer physically. It can only be restarted through some sort of failure-recovery routine, with a hard reboot being the most extreme kind. There’s a whole theory of crash-only software design apparently.

The idea of crash-only design  steelmans a strawman idea of mine that has cropped up in multiple recent posts. In  How To Fall Off the WagonI argued that falling-off-the-wagon is the right focal point for understanding self-improvement efforts.   On the Unraveling of Scripts was about why major life transitions are necessarily messy. In The Adjacency FallacyI argued that career transitions necessarily involve a period of anomie caused by status and value disorientation. Crash-only is the more powerful version of all those ideas. From a crash-only perspective, falling off the wagon (and getting back on) isn’t the main thing. It’s the only thing.

A self-improvement system, or a management model for a business, that doesn’t solve for crash-only constraints isn’t a solution, because it will cost more in crash-recovery effort than the value it creates. Transition management of any sort has to be entirely about crash-only mechanisms.

Software ideas of crash-only design don’t port well to human lives and businesses. So how do you port the thinking?

The software in your head — all that neural code that drives all your thinking and habits — is largely crash-only. You can start and stop relatively simple behaviors in non-crash-only ways (like closing your eyes, starting and stopping coarse bodily movements like walking), but the rest is crash-only.

Management models embodied by an organization — all those organizational habits, incentive structures and CEO-message-reinforcement — is largely crash-only.  You can change how you order paper clips and your lunch buffet vendor in a graceful non-crash ways, but the rest is crash-only.

Crash-only means there is no such thing as gracefully starting and stopping non-trivial parts of your life or business.  You have to crash what you’re doing and recover in a more promising direction. The fact that it is a crash means that, unlike normal decisions, there is a sharply increased probability of not coming out the other end.

Crashes are traumatic, high-entropy, messy ways to achieve transitions, which is why transitions in life and business are so hard. We want clean, smooth turns that smoothly and isentropically shed momentum in one direction and accumulate it in another direction.

It doesn’t happen.

We recognize this intuitively and try to minimize crashes. One way we do this is to burden a single crash-and-recovery event with a huge pile of accumulating necessary changes. At some point, the burden gets so high, we actively wish for a disaster that can give us a fresh start. This makes the rare crash events worse, since we are not used to dealing with crash/recovery periods and have no good transition-management meta behaviors.

In software, I think crash-only design thinking has actually gone mainstream in a disguised form. Though not much actual software is written in crash-only ways, the emerging discipline of DevOps incorporates crash-only thinking in deployment processes.  But that’s a bunny-trail for software engineers.

If life and business, crash-only is not a choice but an operating condition, because both are bundles of entrenched habits without on/off switches. Not only are there no kill commands (or the equivalent of Task Manager in Windows to force misbehaving programs to shut down) to forcibly stop behaviors, there is no equivalent to pulling the plug, short of perhaps some sort of controlled brain damage.

Crash-Only Life Phenomenology

Here are some illustrations from personal life at various levels of abstraction.

  1. You can’t change careers by stopping one career and starting another. You can only crash one career, more or less gracefully, and do some bricolage with the broken pieces to restart in another domain.
  2. You can’t live life as a series of projects, with arrival and departure ritual events marking actual clean beginnings and endings. You can only abandon one project, leaving some residual marginal value on the table so to speak, and dive into the deep end of another, where there is already a pre-existing party that you have to gatecrash.
  3. You can’t fall asleep by flipping a switch and turning off your thoughts. You can only let the thoughts exhaust themselves, and perhaps kill some runaway thought trains with milk and cookies.
  4. You can’t wake up with a clean mind. Only with fragmentary memories of dreams, dormant worries rushing back in with a vengeance, and so forth.
  5. You can’t “calm down” when you are upset simply by flipping a switch. You have to meditate or something. You can’t “get your head in the game” by an act of sheer will, you have to fake enthusiasm and energy until the real thing takes root.
  6. You can’t perfectly pack up your stuff and unpack it in a new home and resume your life. You have to disrupt and crash several of your routines, and hack together new, shaky ones and stabilize them.
  7. You can’t “win” life as though it were a staircase of finite-game “levels” (in both the James Carse sense and the video-game sense). You have to refactor your understanding of what life is, through some infinite-game immersion periods between levels; what I called liminal passages in Tempo.

And here are some from business.

  1. You can’t just cleanly exit a dying line of business with flatlined revenue and shrinking margins. You have to go through some traumatic layoffs and restructurings, create a pool of cash to restart with (refinancing if necessary), and then restart in the mess.
  2. You can’t just decide to switch out the technology on which your company is built for a new, better kind. Too many people have invested careers, learning time and so forth. You have to slowly starve the things you want to kill and feed the new stuff and grow it as a new competency.
  3. You can’t just abandon an unprofitable market and gracefully pivot to a new one even if the two take exactly the same competencies and technology. If you stop selling apples and start selling bananas, your loyal apple-only customers will punish you with bad PR, new banana-only customers will need to be stolen in messy ways from banana incumbents, and apple+banana customers will have to go through something of a renegotiation with you, especially if they already buy their bananas elsewhere but like you enough to give you that business.
  4. A board of directors cannot just swap out a CEO. They can only crash the entire loyal part of the executive team, put in a lame-duck temp, and have the new CEO rebuild via some delicate surgery that will involve tests of loyalty, new people and persuasion of indispensable but hostile people.

A common feature of all these examples is that there is no true “off” state for life, other than death. Not even sleep is an “off state.” While you’re alive and breathing, things are changing. The change is predictable and controllable to a degree, but during periods when it isn’t, it still goes on. 

A crash is fundamentally “life going on” even after you lose predictability and control. A recovery is not resumption of life. Life never stops. Recovery is about regaining predictability and control of an uncontrolled, unpredictable process.

The Zero-State Fallacy

We don’t like crashes. That’s stating the obvious. Our basic crash-management approach is to stay away from crash-inducing conditions. When crashes start, we switch to crash aversion behaviors. When that fails, we start looking around for a way to achieve a clean, painless restart in a safe state. A zero state.

Imagining that there is a zero-state you can return to is the zero-state fallacy. We are not computers.

Examples include: going on a business retreat to reaffirm the “vision”; going on a vacation to “fix” a marriage;  getting drunk;  moving to another continent for a “fresh start.”

These might all very well work, but not in the hoped-for painless ways.

Here’s why.

One consequence of life-being crash-only is that it has an irreversible element to it. Since you can never unplug and reboot into a clean, known state,  something is accumulating, somewhere. That’s practically what it means to have a life or a business: an accumulation. When nothing accumulates anywhere, you get a stateless (in the computing sense) system with very limited capabilities, because it remembers nothing, learns nothing and can get better at nothing.

Long-term memories are of course the most obvious part of what is accumulating, but there are also layers and layers of ingrained habits.

The best you can do is go through grounding rituals by temporarily focusing both attention and behaviors on a minimalist state of being.

But unlike the safe states you can reboot computers into, grounding rituals don’t actually suspend other stuff. They just give you a breather to regain some composure. You have to hope things don’t go even more irretrievably out of control while you’re doing that. A safe assumption if you take five minutes off to get a grip on yourself during a business or personal crisis, but not something that helps if you take weeks or months off in the middle of a crisis to go on a meditation retreat. That’s just procrastination, escapism and denial bundled together.

Effective grounding is mainly about observing without yielding to the impulse to control or predict or exit the situation. This builds capacity for emotional self-management and impulse control. Unfortunately, grounding doesn’t actually make a crash-only system capable of graceful starts and stops. Just because you have something of a temporary safe space from which to view and respond to things, doesn’t mean there are meaningful pain-averting responses ready and waiting. There’s no such responsive creativity on tap the way there is electricity on tap.

This means grounding is of limited value, and once you’re done with your meditation session, you might still be screwed.  It might very well be a case of breathe, count to ten, say the serenity prayer, then panic.

The zero-state fallacy can be restated as the belief that if you just think up the right grounding ritual, you can avoid all pain.

If there are no zero states, what do we actually have?  We have crash states.

Crash states are unsustainable (but not necessarily short-lived) conditions of lack of control, predictability, or both. States where you know you’re in the process of crashing, but don’t know how long it’s going to take, or whether you’re going to come out the other end alive.

What can you do in a crash state, once you’re done grounding and composing yourself and snap out of the zero-state fallacy?

Most people get back to trying to predict and control. Unfortunately, there may be no candidate ways to influence the environment, which means your prediction and control will mainly focus on your internal state and bracing for the crash. There’s a 2×2 here that I won’t sketch out graphically:

  1. Predictable and controllable crash-state behaviors are sort of like equanimity in the face of pain. Like Liam Neeson meditatively closing his eyes in Batman Begins as he realizes he is going to die with the crashing train.
  2. Predictable but uncontrollable crash-state behaviors are obsessive-compulsive behaviors. You cannot help yourself. In the computing analogy, they are processes that you cannot kill, forcing you to physically unplug the computer. Except you cannot. In business, they are like rearranging the deck chairs on the Titanic, or the string quartet continuing to play while all panic.
  3. Unpredictable, but controllable crash-state behaviors are self-destructive risk-taking behaviors like binge-drinking every night. You may get into the state deliberately (assuming you’re not an alcoholic), but you can’t predict what will happen as a result. With businesses in a crash state, you often get a bunch of wild-eyed attempts at last-ditch innovation.
  4. Unpredictable and uncontrollable crash-state behaviors are what we generally think of as acting crazy. There is no logic to the behavior. It simply burns off anxious energy.

The most common crash-state behavior is a degenerate predictable and controllable behavior: terminal inaction, a sort of dull, walking-dead state of foggy anticipatory shock.

None of this actually returns you to a true zero state or achieves any sort of “reboot.” To the extent they are even relevant to the crash, they may achieve some amount of damage control. But lowering the damage alone does not mean you’ll be able to recover.

The zero-state fallacy is really mortality denial. There is a more effective response that increases the probability of coming out alive on the other end. But to use it, you have to understand how identity plays out during crash-and-recovery periods.

Identity Crash-States

The most critical part of the accumulating reality of your life is through the notion of identity. Identity is a combination of things you believe about yourself (self-archetype) and things you believe about the rest of the universe (doctrine).

Identity is also the uncrashed locus from which you are able to view crashed, crashing or recovering parts of your life. The best approximation we have for a safe-reboot state.

But when a part of your identity itself has crashed, you cannot go through your grounding rituals. The four crash-state behaviors I outlined in the last section get far worse when identities crash, since there is no emotional self-management, no regained composure, no periodic grounding. Crashed identities radically lower the probability of surviving the crash. In fact, as a first approximation, you could say that identity crashes are likely irrecoverable.

The smaller your identity, the lower the risk of such irrecoverable crashes. But small identities do come with a cost. They involve forgoing the power of highly passionate emotional drives during non-crash-states.

The analog for businesses is brand equity, the most important component of goodwill. A normal business crash-and-recover routine can tarnish the brand in ways that can be fixed with future successes. Amazon routinely does that. But a crash that impairs the goodwill of the brand in a fundamental way makes recovery far harder.  This is not about market sentiment as much as it’s about employee sentiment. When employees grow attached to a particular sense of status and meaning associated with working for say, Google, a threat to the brand is a threat to the identity of the humans who’ve grown attached to it.

When that sort of brand impairment happens, important employees start to leave, mumbling standard lines about values diverging. Since these are typically the talented people who drive creative decision-making, the people left behind tend to be the ones with ingrained, hard-to-change, robotic behaviors: the Clueless.

Accelerating into a Crash

People and businesses in crash-states often have a strange kind of eagerness to “get it over with” just to get out of the stress of waiting for an outcome. Unfortunately, crash states can remain unsustainable longer than you can maintain the sanity to inhabit them.

A life or business crash is not necessarily a naturally accelerating, self-terminating behavior. Unlike a plane crash, it need not hit some sort of degenerate, no-energy state very quickly after control and predictability are lost. A plane that goes out of control quickly crashes, burns and then cools down into a heaping mass of twisted metal. Lives and businesses that go out of control can thrash around in misery for a very long time.

This is because life or business crash states can drain energy steadily through persistent behaviors that are unpredictable, uncontrollable, or both, and painful to endure.

This happens in environments that are themselves more complex crash states. The crash state can be locally adaptive, as Jordan argued recently.

When humans are in crash-states that are enabled by a social environment of others in crash-states, you get patterns of co-dependency. When businesses are in crash-states enabled by an environment of regulatory capture and protectionism, you get patterns of crony capitalism. Sustainable, but unpleasant in the short-term, unsustainable over longer time horizons nobody cares about.

So how do you actually crash and recover well when you are in such slow-motion crashes that seem to go on and on forever?

You accelerate into the crash. It is one of those counter-intuitive things, like turning into a skid while driving, getting an airplane out of a tailspin by pushing the stick forward instead of back (I think I got that right?), or emptying your lungs when exiting a sinking submarine instead of exiting with a big lungful (the pressure difference can be fatal as you surface otherwise).

Accelerating into a crash helps you regain actual control authority and predictability. If you force a crash into unfolding faster than it naturally wants to, you gain control over it.

This sounds similar like the impatience of “getting it over with” but is not. Get-it-over-with impatience tends to focus on actions that visibly and demonstrably shorten time-to-live. A significant crash milestone is moved ahead, like an announcement of a product being discontinued, or giving notice at a job. This does not mean you gain any control or predictability. The actual crash-time may still be too far out and uncertain. You may get no indication that the crash has come nearer. When you quit your job today instead of next week to work on your startup, you may get no indication of when your business will go through its existential crash moment.

So what does it mean to accelerate into a crash if it isn’t moving milestone events forward?

I am still thinking about this. Accelerating into a crash seems to be a very situation-specific behavior. It’s easy to convince yourself you’re doing it when in fact you’re bracing for the crash, trying to get it over with, or wishfully seeking a zero state.

I can recognize it when I see it: the person doing it seems to have a death-wish of sorts, but seems strangely exhilarated about it. It’s not an isolated event. It’s a changed pattern of behavior that might continue indefinitely. You could accelerate into a crash for weeks or months on end. Sometimes even years in the case of businesses.

At the moment, the best heuristic I have is this: if a behavior change ( not an event or a single decision) brings you immediate relief from anxiety and makes you busier, but without any immediate material change in the environment, it’s likely a case of accelerating into a crash. Often this means giving up doing certain things you’re used to doing, and giving up your resistance to certain things others want to do (which might cause them considerable surprise).

Lame duck leaders are one example of people in this state. A marriage that slips from active hostilities or cold mutual contempt, and into a relatively peaceful detente, is another.

About Venkatesh Rao

Venkat is the founder and editor-in-chief of ribbonfarm. Follow him on Twitter


  1. >> Effective grounding is mainly about observing without yielding to the impulse to control or predict or exit the situation.

    This is a great explanation of this new age notion I have always struggled with.

    This article is giving me a wonderful sensation of synchronicity, as if you were addressing my personal problems directly with astonishing timing.

    A tangent analogy comes to mind: when one feels suicidal for instance, perhaps really they are just truly ready to make a radical necessary change that they couldn’t allow themselves to contemplate before because of pride and long term energy investments that didn’t pay out as expected.

  2. Ditto. It is fascinating how timing of events lines up sometimes. Great article!

    I am a bit unclear on how devops incorporates crash-only thinking though.

    • Maximizing frequency of releases, pushing lots of critical updates at peak load (Intuit example), chaosmonkey, are all examples of forcing the system to evolve faster by putting more frequent crash-causing stressors on it. This forces the dev and deployment process to be more robust, easier to recover/rollback etc.

      Traditional waterfall = minimize release frequency, overload massive launches with huge feature set updates.

    • is a great introduction to the principle from a systems perspective.

  3. The only way to accelerate anything is by pouring more energy into it. In the acceleration examples you listed, gaining control means aligning energy output with the direction of the crash event. It would be an application of the “non-resistance” tactic… think Muad’Dib bending like a reed. Whatever you resist, persists & etc. I might be using this analogy wrong, but in OODA terms, you’d accelerate a crash by getting inside its loop & taking whatever action is necessary to align existing energy output with the crash direction.

    • Yes, that’s the general principle. It’s just very hard to recognize in practice what actions they map to in a given situation, and many temptations to believe you’re accelerating into a crash when you’re really doing something else, like displacing anxiety, giving up, etc.

  4. There’s one simple example of crash-only behaviors in life that I think describes them perfectly: with very rare exceptions, ending a romantic relationship is a crash behavior. The amount of entropy in doing so is so high that most people at any given time know someone who has been putting off breaking it off with someone for months if not years.

  5. Also, here is the evil twin version of your line of thinking:

  6. I am the 3rd case so don’t be surprised if you hear more people mentioning that this article seems to directly address some of their thoughts. Maybe it’s just the times… and why and especially their book, The Escape Manifesto, struck a chord with many.

    I visualized the difference between impatiently bracing for the crash and accelerating into it as follows: someone rigidly leaning back with clenched teeth in a derailed train versus someone deciding to lean in and sway with the hurtling train, enjoying the jumping-shaking scenery with an equanimous curiosity. Who is to tell whether this latter state was achieved through something snapping in the brain (gone “crazy”) or through a mature, self-aware and logical assessment of the situation? Maybe the post-crash behavior is the only proof.

    The last few posts here at ribbonfarm, constitute a good pack of Advanced Career Thinking–along with one or two of the older ones, they could be your next ebook.

    • In terms of sequence, these last few posts kinda pick up where I left off last year with the sequence on freedomspotting, quality of life etc.

      As for whether this is timely, I think so. We’re going through a high-crash period at a macroeconomic level. So the wishful thinking behind hopes for smooth transitions of any sort is particularly ineffective at the moment. In more stable times, you can sort of convince yourself that transitions are painless and isentropic.

    • Uber Manager, Taste Manager, Turtle Conservation Manager, Operations Director, Cave Angel, All round superstar … LOL

      Perfumed bullshit jobs in exotic locations – that’s what people want to escape for? I missed “Islamic State City Major” but I guess humor is not on their site.

  7. Stoicism and Buddhism both propose methods for reducing the surface area of one’s identity as a means of damage-minimization.

    The accelerate-into-crash seems to more closely parallel models of death/rebirth or conversion. Politically, this seems to dovetail with the recent resurgence of talk of Marxist #accelerationism.

    Lots of food for thought. The Yegge piece linked in the first pp is fantastic.

  8. When you’ve gone so far to deny certain rationalist ideas such as off or zero states for human life or organizations why not also reject the idea of pure states at all? Pure states are then projections onto certain views or stories which might contradict each other and coexist. There are inner changes which are hidden from view, and which happen within the realm of the subconsciousness. A disillusionment for example might be experienced as a critical passage because the projection is still active but will be consciously rejected. You know that the story is wrong and doesn’t make sense anymore but you cannot voluntary turn it off. It begins to blend in and haunt you. You may even have experienced that multiple times and can’t do much about it. Only you know that by the time it gets overridden and you might stay calm and civilized for that reason. This is maturity.

    It looks to me that ideas such as “accelerating into a crash” still gives too much credit to the active subject which is a more or less useful philosophical fiction. We always want to look like masters of our self both towards others and ourselves. This is what makes us modern. We want to tinker around, try this and the other nutrition supplement and vaccine to keep the meta-narrative alive that we have an infinite potential of action and will ultimately succeed for good reasons, while something entirely different happens to us and also within us.

  9. Example of acceleration ( from Seattle basketball):

    Can’t find the specific quote, but Nate McMillan was a lame duck coach, and at some point he decided to intentionally do things his way because he expected to lose his job no matter what. Resulted in successful record and a large raise.

    If you knew your phone was going to crash in 30 minutes and never recover, would you do anything differently?