The Oblique
Solution

On non-linear cognition, the incubating mind, the neuroscience of the shower, and where the answer actually lives

The model of productive cognition that the modern organisation implicitly holds is simple, linear, and wrong. It runs as follows: a problem exists; a person is assigned to solve it; the person sits at the problem for a sufficient number of hours; the solution emerges. From this model follow the desk, the schedule, the status meeting, the time-tracking software, the open-plan floor that allows visibility of occupancy, the Slack message sent at half past eight in the evening to a person who is supposed to be at the desk at nine the following morning. The model treats the brain as a deterministic processor — slower or faster depending on experience, more or less accurate depending on skill, but fundamentally a machine that produces output in proportion to directed input, whose utilisation can be measured by the number of hours in which it is pointed at the problem and whose underutilisation is the gap between those hours and the contracted total.

This model is not a fringe assumption. It is the foundational premise of almost every management practice in the contemporary technology organisation. It is also contradicted, comprehensively and specifically, by a century of psychological research and several decades of neuroscience, and by the private testimony of essentially every person who has ever solved a hard problem: that the solution did not arrive at the desk. It arrived in the shower, or on the morning run, or at the moment of waking from a night's sleep, or on the motorway at a speed that made it inadvisable to write anything down. It arrived, in other words, when the directed conscious attention that the organisation is paying for had been withdrawn, and something else — less visible, less schedulable, less amenable to management — had been given room to operate.

In 1926, the English social psychologist Graham Wallas published The Art of Thought, a book that has been cited in almost every subsequent serious treatment of creativity and insight and that remains, nearly a century later, the most useful single framework for understanding why hard problems resist direct assault. Wallas proposed a four-stage model of creative thought drawn from the introspective accounts of scientists, mathematicians, and artists, and from his own observations of the thinking process. The four stages were: Preparation, in which the problem is consciously examined, existing knowledge is applied, and the dimensions of the difficulty are established; Incubation, in which conscious attention is withdrawn from the problem; Illumination, in which the solution or a significant insight arrives, typically suddenly and involuntarily; and Verification, in which the insight is tested and elaborated. The model is not a procedure. It is a description of what actually happens — an attempt to account for the observed fact that the most significant cognitive events in the lives of productive thinkers occurred not during the preparation phase, when directed attention was maximum, but during or immediately after the incubation phase, when directed attention had been, deliberately or incidentally, suspended.¹

The primary evidence Wallas drew on was the introspective literature of accomplished scientists, and the most detailed and analytically careful of those accounts was the one produced by the French mathematician Henri Poincaré, who had described his own experience of mathematical discovery in a lecture to the Psychological Society of Paris in 1908, subsequently published in Science and Method. Poincaré's account is worth attending to closely, because it is not the anecdote of a mystic but the precise observation of a working mathematician examining his own cognitive processes with the same rigour he applied to his mathematics. He described working intensively on the theory of Fuchsian functions, failing to make progress, and then setting the problem aside to join a geological excursion. At the moment he stepped onto a bus at Coutances — mid-conversation, foot on the step — the key insight arrived, fully formed and certain, with no preparatory conscious reasoning he could identify. He did not write it down immediately. He was confident it was correct. He completed the excursion, returned to the mathematics, and verified it. The insight had not occurred at his desk. It had not occurred while he was thinking about mathematics at all. It had occurred in a transition — a moment of suspension between one engagement and another, when the mind was occupied with something else and the problem had been handed to a process that did not require his conscious participation.²

Poincaré was not describing a lucky accident. He was describing a mechanism, and he was sufficiently precise about it to formulate a hypothesis: that unconscious mental work continued after conscious attention had been withdrawn, and that the incubation period was not empty but active — differently active, invisibly active — and that the products of this hidden activity surfaced into consciousness not gradually but suddenly, at moments of threshold crossing whose timing was not under voluntary control. The hypothesis was, in 1908, speculation informed by introspection. It is, a century later, one of the most thoroughly confirmed propositions in cognitive neuroscience.³

In 2001, the neurologist Marcus Raichle and his colleagues at Washington University in St. Louis published a paper in the Proceedings of the National Academy of Sciences that identified and described what they named the default mode network: a set of brain regions — the medial prefrontal cortex, the posterior cingulate cortex, the angular gyrus, the hippocampal formation, and associated areas — that displayed a systematic and reproducible pattern of coordinated activity during rest, and that were consistently deactivated during tasks requiring directed external attention. The finding was initially treated as a methodological nuisance — the baseline activity against which task-related activations were measured. Raichle recognised it as something more fundamental: a network with its own coherent organisation, consuming roughly twenty percent of the brain's total energy budget at rest, performing functions that were not incidental to cognition but central to it. The network's activity was not the absence of thought. It was a different mode of thought — and the question of what that mode was doing became one of the most productive research directions in cognitive neuroscience of the following two decades.⁴

What the default mode network does during its periods of unconstrained activity has been studied through a combination of neuroimaging, experience-sampling research, and creative performance tasks, and the picture that has emerged is of a network specifically organised for the cognitive operations that directed attention suppresses: the integration of autobiographical memory with present context, the simulation of alternative futures and hypothetical scenarios, the detection of remote conceptual associations, the construction of narrative coherence across loosely related elements. These are not peripheral functions. They are precisely the functions required to see a problem from a new angle, to recognise that two apparently unrelated structures are instances of the same underlying pattern, to find the solution path that a fixated mind has been systematically excluding from its search space. The default mode network is, in the language of the research literature, the brain's creative infrastructure — and it is maximally active when the directed, task-focused, externally oriented attention that organisations measure and reward is minimally engaged.⁵

Roger Beaty and his colleagues at Pennsylvania State University, in a series of studies published between 2014 and 2018, used network connectivity analysis to examine the neural basis of creative thinking and found that high creative achievement was associated not with elevated activity in any single brain region but with the strength and flexibility of the functional connections between the default mode network and two other large-scale networks: the executive control network, which governs directed attention, and the salience network, which detects and prioritises incoming information. What distinguished highly creative individuals was their ability to engage the default mode and executive networks simultaneously — to allow the associative, wandering, memory-integrating activity of the default mode while retaining sufficient executive control to evaluate and develop the associations it produced. The creative mind, in Beaty's data, is not the focused mind or the wandering mind. It is the mind that can do both at once, and shift fluidly between them.⁶

The neural correlate of the insight moment itself — the instant at which a solution arrives — was captured with unusual precision in a 2004 paper by Mark Jung-Beeman and John Kounios, published in PLOS Biology, that used simultaneous EEG and fMRI to image the brain during the moment of insight on a verbal problem. The finding was specific: approximately 300 milliseconds before the participant reported the solution, there was a sudden burst of high-frequency gamma-wave activity in the right anterior temporal lobe — a region associated with the integration of distantly related semantic information. The burst was distinct from the neural activity associated with solving the same problems analytically: the analytic solution showed a progressive, bilateral activation pattern consistent with systematic search, while the insight solution showed a sudden, lateralised, high-amplitude event that had no equivalent in the analytic condition. The gamma burst was preceded, in the seconds before the insight, by a brief increase in alpha-wave activity over the visual cortex — a neural signature of reduced visual processing, as though the brain were briefly shutting out external input to attend to something internal. The mind, at the moment of insight, was not more alert to the environment. It was less.⁷

This is not a marginal finding in an obscure subfield. It has been replicated and extended across multiple laboratories, across different problem types, and using different neuroimaging modalities. The picture it paints is consistent: insight — the class of cognitive event that Poincaré described stepping onto the bus and that Archimedes described in his bath — is associated with a specific neural signature that is distinct from analytic problem-solving, that is preceded by reduced external attention, that is lateralised to brain regions associated with semantic integration rather than logical analysis, and that is not under voluntary control. You cannot decide to have an insight. You can only create the conditions in which one becomes possible, and those conditions are characterised by the withdrawal of the very focused, directed, externally engaged attention that the organisation's model of productive work assumes is the mechanism of solution.⁸

The mechanism by which incubation produces insight has been the subject of theoretical debate for as long as the phenomenon has been studied, and two accounts have accumulated the most empirical support. The first is spreading activation. Proposed by Sarnoff Mednick in his 1962 associative theory of creative thinking and developed through subsequent decades of cognitive psychology, spreading activation describes the way that concepts in semantic memory are organised not as discrete items but as nodes in a network of associations, and the way that activation of one node propagates, with decaying strength, through its associative connections to related nodes. In directed, analytic problem-solving, this propagation is constrained by attention: the executive control network suppresses activations that are not relevant to the current search direction, preferentially spreading activation along well-established, high-frequency associations while inhibiting remote, low-frequency ones. This is efficient for problems whose solutions lie close to the search's starting point. It is actively counterproductive for problems whose solutions require a remote association — a connection between concepts that are infrequently linked, that the analytic mind has not considered adjacent, that only become visible when the inhibitory constraint of directed attention is relaxed.⁹

During incubation, with directed attention withdrawn and the default mode network operating without executive suppression, spreading activation is less constrained. It propagates further along the associative network, reaching nodes that focused attention would have suppressed as irrelevant, exploring the semantic neighbourhood of the problem in directions that the analytic mind had not sanctioned. The shower, the walk, the motorway — these are not cognitively empty. They are cognitively diffuse, maintaining the problem in a background state of activation while allowing that activation to spread in directions that directed attention would have prevented. The solution that arrives in the shower is not a product of the shower. It is a product of the previous hours of preparation — the conscious engagement with the problem that loaded the relevant nodes — combined with the incubation period's release of the associative constraints that had been keeping the solution out of reach. The two phases are not alternatives. They are sequential dependencies. The incubation does not work without the preparation. The preparation does not produce the solution without the incubation. The organisation that understands only the preparation phase has understood half the process and scheduled for none of the rest.¹⁰

The second account is constraint relaxation. Research on fixation — the tendency of a mind that has been working on a problem to become progressively more committed to a particular solution approach, even when that approach is not working — established that one of the primary functions of incubation is the dissolution of this fixation. The mind that is prevented from thinking about anything else searches the same territory repeatedly, reinforcing the same incorrect approaches with each pass, becoming progressively more, not less, stuck. The mind that is given permission to think about something else gradually loses its attachment to the failed approach: the activated representations of the fixated strategy decay as attention is redirected, the search space contracts back toward a more neutral starting configuration, and the next engagement with the problem begins from a less constrained position. Incubation, in this account, is not just spreading activation into new territory. It is also the clearing of old territory that had been mistakenly occupied.¹¹

The shower warrants its own treatment, partly because it has become the canonical location of the programmer's insight — sufficiently universal to be a cliché, sufficiently consistent across populations and decades to be something more than coincidence — and partly because the research on why it works is more specific than its reputation suggests. Scott Barry Kaufman, a psychologist at Columbia University, surveyed a population of more than three thousand people on the contexts in which they experienced their most creative moments, and found that seventy-two percent reported having creative insights in the shower — a figure higher than the proportion reporting insights at their desk, in a meeting, or during any other scheduled work activity. The finding is striking not because the shower is magical but because of what it reliably provides: a combination of warmth, enclosure, white noise, the absence of social obligation, the absence of screens and notifications, the performance of a task so thoroughly automated that it requires zero cognitive allocation, and — critically — the absence of any expectation of productive output. No one enters the shower expecting to solve a problem. The removal of that expectation removes the pressure of directed search, which is precisely the thing that constrained the spreading activation in the first place.¹²

The walk produces the same conditions through different means, and the research on its cognitive effects is, if anything, more precise. In 2014, Marily Oppezzo and Daniel Schwartz of Stanford University published a study in the Journal of Experimental Psychology that measured creative output — specifically, performance on divergent thinking tasks requiring the generation of novel uses for common objects — under conditions of walking and sitting, indoors and outdoors. The results were not modest. Walking increased divergent thinking output by an average of eighty-one percent relative to sitting, and the effect held for walking on a treadmill facing a blank wall — removing any contribution from environmental novelty or fresh air — suggesting that the mechanism was the walking itself rather than the landscape it traversed. The effect on convergent thinking, which requires the identification of a single correct answer rather than the generation of multiple creative ones, was not significant. Walking did not improve analytic problem-solving. It improved the associative, generative, multi-directional thinking that is the cognitive substrate of insight.¹³

Poincaré had been boarding a bus. He had not been thinking about mathematics. He had been going somewhere. The eighty-one percent figure quantifies what he described from the inside.

Sleep is the most extreme and most studied form of the withdrawal of directed attention, and the research on its relationship to insight is among the most dramatically counterintuitive in cognitive science. Ullrich Wagner and his colleagues at the University of Lübeck, in a 2004 paper in Nature, designed an experiment specifically to test whether sleep could enable insight — whether the resting mind could produce, during the consolidation processes of sleep, an understanding of a problem's structure that the waking analytical mind had failed to find. Participants were trained on a numerical sequence task that had a hidden shortcut — a structural regularity in the sequence that, once recognised, allowed the task to be completed in a fraction of the time. Participants were not told about the shortcut. They were tested immediately after training and then again after an interval of eight hours. Half had spent those eight hours awake; half had slept. The participants who slept were nearly three times more likely to have discovered the hidden shortcut than those who had remained awake. Sleep had produced an insight that wakefulness had not, in participants who had spent the eight-hour interval doing nothing they would have described as working on the problem.¹⁴

Otto Loewi, the German pharmacologist who won the Nobel Prize in Physiology or Medicine in 1936 for his discovery of chemical neurotransmission, reported that the experimental design that confirmed the existence of the nerve impulse — a finding that would become one of the foundational results of modern neuroscience — came to him in a dream. He woke, wrote notes, returned to sleep. In the morning, the notes were illegible. The following night, the dream returned. This time he went directly to his laboratory, at three in the morning, and performed the experiment that would win him the prize. The experiment had not occurred to him in years of focused research on the relevant problem. It occurred during sleep, in a hypnagogic state, in a mind that was not trying. August Kekulé described the ring structure of benzene as arriving in a reverie — a daydream in which he saw atoms forming themselves into chains that curled back on themselves like snakes biting their tails. The structural insight that resolved one of the central open problems in organic chemistry of the nineteenth century did not come from additional analysis of the problem. It came from a metaphor generated by a mind that had been allowed to wander.¹⁵

The scientist who has spent the day in meetings has not merely lost the day's analytical hours. They have lost the day's incubation hours — the unconstrained mental time during which the day's preparation might have produced something. The meeting filled the incubation period with coordination overhead. The problem sat, unattended, in the wrong kind of waiting.¹⁶

The application to software development is not merely analogical. The hard problems of software — the concurrency bug that reproduces only under load, the performance regression whose cause crosses three abstraction layers, the architectural decision whose implications ramify into subsystems that were not imagined when the design was chosen, the security vulnerability that requires simultaneously holding in mind the attacker's capabilities, the system's state machine, the runtime's memory model, and the network's timing behaviour — are precisely the class of problem for which direct analytical search is most likely to fail and non-linear, incubation-mediated insight is most likely to succeed. They are hard not because they require more time at the desk but because they require a connection that the analytic mind has not made, between concepts that the directed search has not considered adjacent. They require, in Mednick's framework, remote association — and remote association is the product of the diffuse, unconstrained, default-mode-mediated spreading activation that the desk, the meeting, and the notification pipeline are designed to prevent.

The programmer who reports that the solution to a three-day debugging problem arrived in the shower on the fourth morning is not reporting an anomaly. They are reporting the normal operation of a mind that had been given, for the first time in three days, the conditions under which the relevant cognitive process could complete. The three days at the desk were not wasted: they were the preparation phase, without which the incubation would have had nothing to work with. But the solution itself — the moment of recognition, the connection between the thread scheduler's preemption window and the cache invalidation timing that produced the race — was not produced at the desk. It was produced in the shower, which is to say, in the default mode network, which is to say, in a brain that had finally been released from the inhibitory pressure of directed attention and allowed to search its own associative space in directions that directed attention had been excluding for three days.¹⁷

Ap Dijksterhuis and his colleagues at the University of Amsterdam published a series of papers between 2004 and 2006 documenting what they termed unconscious thought theory: the finding that on complex, multi-attribute decisions — decisions involving many variables that must be weighted and integrated simultaneously, decisions that exceed the capacity of working memory to hold in its entirety — deliberate conscious deliberation produced worse outcomes than a period of distraction followed by an intuitive judgment. The effect was reliable and, to anyone trained in the assumption that more conscious deliberation produces better decisions, counterintuitive. For simple problems, conscious deliberation was superior. For problems of sufficient complexity — and the problems that software development at its limits produces are problems of sufficient complexity — the unconscious process that operated during the distraction period outperformed the conscious process that had direct access to the problem's materials. The unconscious process had more working memory available, could integrate more variables simultaneously, was not subject to the attentional bottlenecks that constrained conscious reasoning, and was not distracted by the irrelevant cognitive accessory of producing a verbal account of its own reasoning. The shower does not understand what it is thinking about. It does not need to. It just thinks.¹⁸

The research programme on incubation effects, surveyed in a 2009 meta-analysis by Ut Na Sio and Thomas Ormerod across one hundred and seventeen experimental studies, produced a finding that is both clear and practically important: incubation effects are real, consistent across problem types and populations, and specifically associated with periods of low-cognitive-demand activity rather than with periods of high-cognitive-demand activity or rest. That is to say: the optimal incubation condition is not sleep, or emptiness, or the complete absence of thought. It is the presence of a task that occupies the surface layers of cognition without consuming the deeper resources that the background processing requires — the shower, the walk, the routine drive on a familiar road, the mechanical domestic task whose completion requires just enough attention to prevent the self-monitoring that would otherwise interfere with the incubation process. The mind needs something to do at the front of the house so that the work at the back can proceed without interruption. Crucially, the meta-analysis also found that incubation effects were larger for tasks that required creative or insight-based solutions, and smaller for tasks amenable to systematic search. The harder the problem — the more it requires a connection the analytic mind cannot force — the more the incubation period matters, and the more the conditions of that period determine whether anything emerges from it.¹⁹

The implication is not that software developers should spend their days showering. It is that the cognitive process that produces solutions to hard problems is not a single-phase process, and that the phase that the organisation cannot see, measure, or schedule is not the dispensable one. The preparation without the incubation is a loop without a termination condition — a search that continues to search the same space, with increasing frustration and decreasing return, until either the deadline forces a compromise solution or the incubation is provided accidentally by the weekend. The organisation that structures its developers' time to maximise directed cognitive engagement — meetings, standups, check-ins, the always-on Slack channel, the open-plan office with its continuous low-level social demand — is an organisation that has optimised the preparation phase and systematically destroyed the incubation phase. It has built a machine for loading the problem into the mind and then preventing the mind from doing anything useful with the load.

There is an objection to this argument that is made in good faith and that deserves a direct response, because it is the objection that will be made by anyone whose working model of productive cognition is the linear one that this essay has been attempting to dislodge. The objection is that this account, if accepted, provides a principled justification for any developer who wishes to avoid desk work — that if solutions arrive in showers and on walks, then perhaps one should spend more time in the shower and on walks and less time at the desk, and that this recommendation will be indistinguishable, in its practical consequences, from laziness dressed in the vocabulary of cognitive science.

The objection misreads the mechanism. Poincaré was explicit, and the research has confirmed, that incubation without prior preparation produces nothing. The default mode network performing spreading activation across a problem that has not been loaded into the system has no relevant material to activate across. The walk after three days of intensive engagement with a hard problem is cognitively different from the walk taken instead of the engagement — as different as the consolidation of learning during sleep is from the sleep of someone who did not study. The insight requires the preparation as its necessary antecedent. What it does not require is the preparation's continuation into the space that the incubation needs. The developer who spends three days at the desk, achieves saturation on the problem, and then takes a walk is doing something productive. The developer who is prevented from taking the walk — by the three o'clock status call, by the afternoon retrospective, by the Slack thread that requires attention and response — is not being kept productive. They are being kept from the conclusion of a cognitive process that the previous three days have been building toward. The organisation has paid for the preparation and is systematically preventing the return on its investment.²⁰

The deeper difficulty is that the timeline of non-linear cognition does not map onto the project schedule, the sprint calendar, or the quarterly review. Insights do not arrive on deadline. The gamma burst in the right anterior temporal lobe does not respond to the product owner's request for an update. The default mode network cannot be asked to complete its operations before the Thursday stand-up. This is genuinely inconvenient for organisations whose planning horizons are defined by the sprint, and it does not become less true because of the inconvenience. The hardest problems in software — the ones whose solutions determine whether a product works or fails, whether an architecture is maintainable or is rewritten in eighteen months, whether a system is secure or is the subject of a disclosure — are precisely the problems least amenable to the linear, scheduled, deliverable-by-Thursday model of solution production. They are the problems that require the bath, or the bus step at Coutances, or the three in the morning laboratory visit when the dream comes back. The organisation that cannot make room for those moments will solve the easy problems on schedule and the hard problems never, or badly, or at a cost in time and revision that exceeds, by a considerable multiple, what the room would have cost to make.