← Home

Map of ideas

416 ideas drawn from the essays, and the 723 links written between them. Each dot is one idea; its size shows how many links it has.

An idea left waiting loses its temperature rather than its subjectAsking a thought to be clean too early stops it from coming outA capture that lands in a workspace is already a positioningA passive writing tool cannot flag what is missingA tool that requires you to be settled misses the ideas that arrive in motionA short publication deadline selects subjects as much as it speeds up writingPrimary material that becomes clean too early stops being primary materialThe most useful moment in writing is the one where the gaps appearCritical feedback is only worth something if it compares the material with what already existsAn outline written before the material organizes emptinessAuthorship of a text rests on what isn't delegated, not on the keyboardShaping is real work, distinct from the arbitration it servesA writing setup is worth the chaining of its pieces, not each of themA deliverable carries the result without carrying the reasoningThe cost of uncapitalized knowledge only appears at the second deliverableTaking a deliverable as a source makes you inherit choices that are not knowledgeBetween sources and deliverables, a knowledge layer speaks to no audienceA note that shapes itself for a deliverable stops being reusableAI makes an intermediate layer pay off that didn't beforeAfter a deliverable, the question that counts is what remainsCapitalizing on a past form is not capitalizing on a future capabilityA deliverable is worth the constraint it imposes, not what it preservesEvery piece of production leaves a second output nobody collectsJudging consists first in recognizing what matters, not in choosingA decision matures through successive corrections, and that delay is paid forTreating every objection as a correction swaps judgment for complianceA vague unease becomes workable once it is broken into named handlesMultiplying possibilities increases the load on judgment instead of reducing itThe risk isn't lacking information, it is not testing itSpeeding up the confrontation frees time for the only slowdown that countsA well-phrased answer proves nothing about its correctnessAn imperfect idea only matures if something resists itThe displayed transcript is not what the model has to answer withThe value of persistence lies in the decision to keep, not in the keepingAsking where the tool already knows how to act unblocks what "how to make it act" was blockingSlowness blamed on the tool often comes from making it rediscover a known contextA confirmation that doesn't follow a read-back attests to nothingSeparating the exit point from the place of preservation unblocks the choice of toolA safeguard that bears on the agent isn't bypassed by disguising the requestA large account's technical request often expresses a culture of governanceA building block that creates no value can on its own kill a dealA customer demanding their data on their own side isn't questioning your competence, they're demanding controlThe customer controls where their data lives, the vendor controls how the product evolvesA change that requires an intervention at every customer makes versions divergeExternalizing a resource doesn't move the responsibility the user perceivesA split of responsibilities exists only if incidents can be attributedExternalizing a resource lowers infrastructure cost and raises cost of serviceReplacing a provider without changing the value of the service makes it substitutableA non-differentiating building block stays expensive to substitute if it holds the durable stateBreadth of technology support must follow commercial positioning, not feasibilityUnderstanding why a customer asks for control lets you calibrate the control you grantChoosing to sell to large organizations means accepting their governance as a market constraintA memory kept in versioned flat files follows the project, where a memory lodged in the tool stays on the machineAn agent's memory is layered by lifespan, not by subjectPermanent rules don't replace a journal of what happenedDelegating to a model the sorting of what deserves to be retained replaces a human decision with a written criterionPartitioning memory by project is what makes it reloadableSections fixed in advance make what a machine writes continuously rereadableA long memory stays a memory only if a ceiling forces its consolidationAn automation triggered by the end of a piece of work must recognize its own triggersA utility agent inherits the instructions of the directory it is called fromThe name of an automation event doesn't tell you how often it firesAn automation that runs at every exchange must cost little enough that you stop arbitrating itCapture can be automatic, recall must stay requestedA single classification axis forces you to choose between business perspective and precision of actionCustomer feedback is classified by the level at which it is expressed, not by its apparent topicThe number of categories in a taxonomy is decided on its adoption, not on its eleganceDescribing a need at the use case level avoids locking the analysis into a solutionA level of classification is only useful if it leads to a level of decisionA subject everyone agrees is strategic keeps being put off as long as its collection has to be paid for before any thinkingPart of the Product Manager job consists of compensating for the organization's frictionA layer of work passes for the core of a job as long as it costs enough to fill itHanding an ancillary task to a tooled system shifts the work towards framing and supervisionA Product Manager is judged on the problems he spots before everyone else, not on the volume of artifacts he producesTranslating a product only crosses the visible layer of a marketThe moment of paying exposes a country's economic infrastructure in a few secondsInstallment payment can be an ordinary way of consuming rather than a marker of a large purchaseA tax ID number built into ordinary payment shifts a country's privacy normA massively adopted means of payment stops being a tool and becomes a common languageThe choice of a means of payment can answer a physical risk rather than a preference for convenienceThe informal economy isn't defined by the absence of technology: it equips itselfA tolerated informal economy stands in for a social safety net where there isn't oneA few percent of commission are enough to steer a means of payment when the margin is vitalA dominant position doesn't transpose into a country already structured by other habitsA disorienting market is neither behind nor ahead: it answers different constraintsWhere institutional verification is weak, personal recognition stands in for itA cultural trait is only visible to the person for whom the gesture isn't normalA zero bug policy doesn't aim for zero defects, but for zero known defects left undecidedA defect excluded from planning consumes the team's capacity anywayA defect becomes a priority when it becomes visible, not when it becomes expensiveA stockpile of defects creates its own maintenance work, separate from fixingA bug ticket stays open because closing it would force someone to sign a refusalReducing the qualification of a signal to two outcomes prevents hiding the trade-off behind a taxonomyA user doesn't distinguish a defect from a missing capability: he feels a painA missing capability can hurt more than a proven defectSeparating bug cards from feature cards creates two priority systems that are never compared"To be prioritized later" is a refusal nobody signedIgnoring defects to hold velocity protects the metric and not the productA zero bug policy doesn't create the slowdown, it makes the system's defect rate visibleGetting out of a stockpile of defects starts with what comes back to support, not with a reprioritizationA stockpile-reduction phase has to be bounded, otherwise it becomes the normal backlogWhen qualifying defects costs more than fixing the system, the stockpile has only changed its nameEmptying a stockpile of defects and preventing it from re-forming are two distinct problemsBetween the specification and the release, unwritten arbitrations redefine what the customer will experienceDesigning a feature without carrying it through to production externalizes the cost of your own approximationsThe existence of a role says what an organization judged separable, not what that role does wellThe split between those who design and those who execute has already been abandoned on the development sideWhat the market hires for defines a role more reliably than its methodological definitionA consumable artifact can ground a function, not a professionA role's title manufactures the mandate granted to itSplitting product responsibility puts the consequence on whoever had no choiceA division of labor doesn't survive long once its economic justification disappearsA task that isn't the decision bottleneck can still be a ceiling on volumeAn AI can analyze an option, it cannot own the betAn interface role can be redefined by redistribution rather than removedAn interface role becomes harmful when it cuts developers off from the real needDomain expertise is no role's propertyA division of labor that suits whoever defends it keeps the part that tells well in a meetingAs long as the loss isn't named, the committee hasn't arbitratedA committee that gives each function a place also protects everyone from the moment of choosingA useful committee doesn't produce agreement, it shows where agreement stops"We'll adjust later" is legitimate under real uncertainty, not in front of an incompatibility already knownAn explicit disagreement locates the problem, surface alignment disperses itA dodged arbitration comes back later under the name of a coordination problemAn unsettled line is paid for in alignment meetings and cross-validationsA blurry decision dilutes responsibility to the point of making failure unattributableThe soft decision is a product of the organization, not a weakness in whoever had to decideAn arbitration reopened as soon as it displeases an influential function teaches the whole organization that no choice is finalA committee that doesn't know what it can decide and who decides after it becomes a theater of alignmentAlignment happens when disagreements become clear enough to be dealt with, not when the right people are in the roomA shared language and ranked success criteria do what no improvement to meetings can doA retrievable idea is not yet an available ideaA flash card differs from an atomic note by what it commits to, not by its sizeWhat you want to hold within you is recognized by resonance, not by a checklist of criteriaAn anchoring card stops holding as soon as it carries the reasoning that surrounds itA deck of convictions stops being one as soon as it grows without limitRereading a conviction without testing it against what has happened since tests nothingA conviction that has stopped carrying is thrown away, where a note is archivedPaper anchors a pillar of thought, digital only mirrors itThe code is the only artifact that describes the real state of the product, the others merely follow itNatural-language documentation exists for readers who cannot access the code, not for the quality of its informationThe divergence between documentation and code is a cost trade-off, not negligenceClean code stops being a technical-team matter when the code becomes the source of the product artifactsRebuilding a specification from the current code replaces updating it from an old documentWithout technical literacy, a pile of locally effective tools is paid for in the cost of connecting themA technical background gives the Product Manager a lever of access to the product's material, not a superior aptitudeA specialized SaaS sees only one face of the product and cannot carry the complete view of itA mockup rebuilt from the real code stops being a representation and becomes a working basisA tool's value shifts toward its ability to pull context from outside its own perimeterAn information system keeps the event without keeping what it teachesThe forms of the information system are technical answers, not forms of thoughtAdding an AI inside an existing application improves usage without moving the informationA business context runs through the applications instead of replacing themThe more AI is used, the more decisive information governance becomesWorking with an AI consists in tending the context around it rather than writing good promptsAs long as the user has to know the system's structure, it is the user who carries the context of the questionWhat holds back an organizational memory is cultural, not technicalSeparate business memories are better than a central brain, provided they are connectableA backlog's transparency rests on the legibility of its decisions, not on the number of lines it exposesA topic enters the backlog at the moment it becomes collective workDropping three lines into a collective tool delegates a fuzzy thought instead of a piece of workA ticket's level of detail depends on the team's shared context, not on a templateA user story triggers the conversation instead of replacing itAn unbounded backlog cancels the constraint the roadmap had just setAdding a topic without removing another amounts to denying the team's real capacityAn old ticket describes a context that has vanished rather than reliable memoryA ticket isn't an asset just because it existsA customer request is learning material, not an execution orderThe trace of a rejection only serves memory if it carries no work to be doneThe ticket is where you push the result of thinking, not where you do itA list of cards cannot carry material whose value lies in its relationsA coordination artifact is written for its useful life, not for the archiveA captured source is worth almost nothing until it has been picked up againThe criterion for capturing in the heat of the moment isn't interest but promise for a project under waySorting a capture happens in the cold light of hindsight, once the momentum of discovery has subsidedA stock of captures seen as a block makes you choose better than filtering piece by pieceAn external source and a thought in progress don't belong to the same gesture of captureA capture system is judged on the knowledge it produces, not on its completenessIn an agent's cost, what the model writes weighs several times what it readsA skill that loads an entire database to use three records pays for the volume, not the useA deterministic transformation handed to the model is paid for at the rate of reasoningRegenerating a deliverable by rereading the whole history makes its cost grow with the base, not with the new materialA skill's prompt is a fixed cost paid on every runThe model is chosen on the difficulty of the step, not on that of the skillIn a chain of steps, each step's output becomes the input of all the following onesHitting a consumption ceiling doesn't say where the consumption wentA shared quota turns a greedy skill into a cost for those who never launched itAn optimization can only be ranked as a share of the total costA token gain is weighed against the cost of implementing it in the existing codebaseAn estimate accurate enough to rank optimizations isn't accurate enough to validate oneA skill's cost drifts once it is in serviceA network of notes shows that everything is connected without saying howIn a network of linked notes, one false note contaminates the analysis without leaving a traceThe verifiability of what an AI produces comes down to the number of files a human can rereadAn AI turned loose in a borderless network of notes navigates instead of answeringA network of notes tracks neither its contradictions, nor its gaps, nor its decisionsA gap you cannot spot is more dangerous than an identified oneStudying a product opportunity is a deliberate narrowing of scope that the tooling must reflectInformation coming from another domain is copied and adapted rather than linkedModeling product work on development work removes a layer of translationUseful competitive intelligence starts from the target customer, and competitors are only revealers in itA competitor's feature is a visible answer to an invisible line of reasoningWithout defined positioning, competitive intelligence produces only a collection of misread signalsA real competitor is anything that is enough to prevent adoption, not what belongs to the same categoryAnticipating means learning earlier, not guessing the futureThe choice is never between risk and the absence of risk, but between exploring too early and arriving too lateThe level of expectation toward business software is set by experiences lived elsewhereConfusing benchmark, competitive intelligence, positioning and strategy passes off a pile of observations as a decisionRouting a defect's fix to whichever developer is free dilutes ownership of what was shippedSending a defect back to its author acts before the defect, on the way they shipFixing your own defect is the only moment when you see what was missing upstreamLooking for someone to blame after a defect makes defects disappear from conversations, not from the productProduct choice and technical quality are two zones of responsibility that dialogue must not dissolveA developer who does not understand the expected behavior must be able to refuse to buildA test written before the code turns part of the specification into a verifiable constraintQA placed at the end of the chain receives quality responsibility instead of framing itTomorrow's legacy is manufactured with every quality responsibility displaced todayKnowledge fixed by standards can carry a network of notes that opportunity knowledge cannotA setup that reassembles its answer with every question never consolidates the ideas it extractsA wiki generated by an AI stays derived: the source of truth remains the original documentIngesting a source into a consolidated base is measured by the existing notes it changes, not by the notes it addsA base of facts says what the standard prescribes, a personal system of thought says what it is worth hereThe decisive work in automated extraction is tuning the threshold between exhaustiveness and relevanceA tightened version of an extraction system is tuned by having it analyse what the previous one keptA search by semantic proximity finds notes that share no word with the questionA coherence tool is worth only what it forces you to make explicitA disagreement hidden under a shared word costs less at the glossary than at launchFailing to write the target or the promise on one page is a symptom of strategy, not of writingAs long as the elements of a decision stay scattered across several documents, each function keeps its own readingWriting what the customer is supposed to understand before building exposes the inconsistencies the internal point of view masksEvery function has a legitimate indicator, and it is their simultaneous sovereignty that prevents a shared definition of successPostponing an arbitration is already a choice: the choice to accept an unstated cost of delayAn arbitration with no written memory becomes negotiable again as soon as the reasons get simplifiedA decision stays clear while remaining revisable if the conditions for reopening it are written downWriting down why a point of view was set aside distinguishes it from a point of view that was ignoredA disagreement needs a maximum duration of openness, or prolonging the discussion replaces the decisionThe same question settled in several forums signals a missing rule, not a special caseRe-deciding for want of a record costs coherence before it costs timeA decision rule is documented at a level where it does not say what to buildA decision register loses its value as it fills upA decision written with its conditions for reassessment stays revisable without becoming negotiable againA superseded decision is written alongside the old one, never in its placeJudging a past decision without the context of its time produces an imaginary trialA structuring decision lodged in a ticket becomes invisible to the rest of the companyA document is judged on the discussions it removes, not on those it addsAccess to a technology stops being an advantage as soon as it becomes rentable on demandWhen producing becomes abundant, the advantage moves from the capacity to produce toward understanding the contextTwo companies expressing the same need can expect two different things from it, depending on their decision cultureGenerating an interface faster does not fix a poor understanding of the userA product adopted by its users can be blocked by the buyer's decision criteriaProducing more code produces neither architectural coherence, nor reliability, nor securityA traceability requirement reaches the field as an ordinary request for justificationA context document earns its value from the distinction it forces between what you know, what you believe and what makes you doubtA context document is organized by product risks rather than by disciplinesA roadmap replaces a logic of dates with a logic of certainty horizonsA roadmap does not produce the arbitration, it makes it visible and debatableThe column carries the decision, the card explains the valueA serious candidate differs from an idea by the written condition for moving to commitmentKeeping a topic visible without committing to it only makes sense if its reassessment trigger is written downA roadmap aligns through what it excludes, not through what it listsDisplaying more committed topics than the team can deliver manufactures an illusion of commitmentA waiting column with no limit becomes a political pre-backlogA limited number of places turns prioritization into comparative justificationA roadmap not tied to a strategy is only a more legible presentation of the existing messA roadmap must not be the sum of the incoming pressuresThe roadmap selects the topics, the backlog organizes the work on the selected topicsA customer roadmap is a manual translation, not an extract of the internal roadmapShowing an uncommitted topic creates an expectation that counts as a promiseA roadmap with no designated decision-maker becomes a compromise between power gamesA roadmap's review cadence is arbitrated between permanent discussion and driftA topic neither committed nor abandoned keeps consuming the organization's attentionWorking on a topic you don't know will be prioritized lowers the intensity of the workA roadmap protects against scatter and, in the same move, blocks the small obvious improvementsAn exception to the prioritization framework only holds if it is protected and boundedTime outside the roadmap only handles topics whose value is already knownTechnical debt is only a debt from the moment you decide to pay it downA note's atomicity rests on its cohesion, not on its lengthAs long as a note cites its author instead of stating the idea, it remains a captureA note's title is a stance, not a filing labelFailing to find a crisp title for a note signals an idea still unclear, not a lack of styleA note that is too broad blurs its links, a note that is too fragmented turns them into noiseFiling your thinking by concept rather than by source lets several readings enrich the same noteNotes that have already been clarified make writing incremental rather than dependent on one grand act of synthesisA single file cannot be organised for the human and for the AI at onceExcellence in a field comes from the depth of the context you have internalised, not from mastery of techniqueA model's attention degrades on whatever you bury in the middle of its contextA context is loaded as a constant core topped up on demand, not in one blockSeparating the raw material, what you retain from it and what you deliver is what makes a claim attributableA decision made with an AI can only be defended if the path leading to it remains traceableA deliverable regenerated from the context no longer waits for the end of a phaseA context with no outcome keeps its value through what it brings to neighbouring contextsIntelligence comes from bringing distant ideas together, not from accumulating knowledgeA long synthesis rules out the connections a set of short, linked notes makes possibleExplaining a context to an AI forces you to check that you understand it yourselfA perfectly documented context stays incomplete by everything the product manager never wrote downUnwritten knowledge is the only material of a context whose collection cannot be industrialisedA session source captures useful thinking before it reaches the threshold of a documentA micro-signal is only worth something through its buildup in a context that knows why it keeps itRepeating an idea with different nuances makes it mature instead of duplicating itKnowledge stays in people's heads because it is born too small to become a documentA synthesis is worth the angle it chooses, not the subject it coversAssembling several notes surfaces a level of reading none of them carries aloneA synthesis holds between two drifts: going deep on one idea, or being nothing but a list of linksExplaining the relationship between notes is worth more than copying out what they sayA synthesis can carry several ideas as long as they serve the same angleA thesis drawn from existing notes gives an article its architecture before a line is writtenA glossary entry delimits a word, an atomic note asserts a propositionA glossary entry is judged on the usability of the term, not on the exhaustiveness of the conceptA term is only delimited against the words it gets confused withA glossary becomes necessary when a note system serves several people or several purposesThe links in a network of notes are only as solid as the words that carry themA glossary replaces neither the notes nor the sources: it gives them a common vocabularyA question about how the product behaves calls for a reading of the source, not a product judgmentThe time spent checking what the product already does appears in no tracking instrumentAnswering a question outside your current subject costs the continuity of the work in progress more than the time it takesThe border between what you maintain and what you regenerate runs between what explains and what describesForcing a generated answer to cite its evidence constrains what it allows itself to assertAn answer that doesn't say where it gets stuck moves the risk instead of reducing itAn answer structured along angles serves several functions without duplicating itself per audienceRetrieving the right fragment and travelling to its neighbors are two distinct capabilitiesA repository's history carries the intent its current state has erasedA product's real behavior isn't in the repository alone: it depends on the deployed configurationQueryable code is code whose exceptions carry business rules, not code that is cleanA specification changes status when it ships: a tool for dialogue before, a regenerable artifact afterA coding assistant doesn't invent technical debt, it changes its scaleSlow production imposed an understanding that nothing takes over when it falls awayAn error that arrives at a scale where it is no longer analyzable stops teachingThe tasks that used to train junior developers are the ones an assistant automates bestA general acceleration makes indefensible the learning time it does not accelerateBest practices are born from the resistance of the field, not from rereading the previous onesAn understood system is a system you would know how to rebuildAccepting code nobody can explain takes on an understanding debtWithout a catalogue of patterns decided by the team, every generated feature invents its own way of doing thingsA failed output from an assistant signals a missing guideline before it is work to fixA feature's complexity is paid for outside the code, in adoption, support and documentationDeliberately reintroducing difficulty maintains a skill the tool makes unnecessary day to dayRebuilding a lost context doesn't only cost time: the next decision is made on an impoverished versionStructuring a piece of customer information means making it thinkable, not tidying it awayProduct evidence is built by the convergence of signals, not by demonstrationA context that brings together the lenses of several functions makes each one's blind spot visibleCirculating a summary without leaving access to the raw material rules out any later reinterpretationKnowledge capital is worth only what brings it back into decisionsOne and the same building block of context yields different deliverables depending on the function that picks it upA product is judged on the capability it adds to the customer, not on the task it carries outAn aggregate kept up to date erases the history that led to itA processing directive encodes the ground knowledge only the holder of the context possessesProcessing rules and the material they process don't fit in the same fileDelegating a split without saying how to split returns the model's structure, not yoursAn AI's acceptable error rate is set per domain, not in the absoluteA context system moves up one step each time the previous one stops being enoughThe first context system exists to teach you how the AI organizes, not to lastIn AI-assisted work, the disk carries the memory and the conversation is only a channelA mission's goal and the attention of the moment must be held in two separate filesA mission frame fills up through what gets said in passing, not at the moment you open itA directive encodes a way of working, never information about the subjectParking a future action outside the conversation takes it out of the reasoning under wayA working context must be able to receive a message when no session is openA resumption state is written from the question of someone arriving having read nothingA model that compacts its conversation chooses for itself what it keeps, and doesn't say soForgetting, in a workspace, is a loading decision and not a deletionHaving an assistant write a script rather than carry out the task makes the operation checkable and repeatableA running product, imperfect as it is, surfaces friction a mockup cannot showA tool built as throwaway becomes critical without anyone having decided itWith no stated volume constraint, an assistant designs for the test dataset and the flaw shows up as slowness, not as an errorSupervising a code assistant demands the work of mentoring a junior without the progress that pays for itRaising the capacity to produce moves the bottleneck onto whoever decidesAccelerating a single stage of a workflow moves the bottleneck to the next oneDelegating implementation to an agent forces the product decisions the developer absorbed in silence to be made explicitThe right to produce code is set by the relationship between competence and the product's risk, not by competence aloneShipping faster without access to customers or trade-offs makes an agent operator, not a product engineerSecurity is the one software requirement that does not scale with what the product has at stakeWhat a contributor can carry is measured by their level of capability to act, not by the job they came fromGoverning code produced outside the engineering team means connecting it to the tooling that already existsRequiring review of a throwaway personal tool recreates the very dependency on developers you were trying to reduceBuilding your own tool commits you to owning its failures, and that is the load the assistant does not take backTests are the foothold that remains on code you did not write yourselfPull request review becomes a governance mechanism as soon as non-developer profiles contributeOpening up a contribution perimeter is decided on the risk of the area, not on the contributor's jobTests written after the fact by the author of the code validate its implementation, not the business intentAn explicit technical framework speeds an assistant up instead of holding it backA product decision, a technical decision and an experience decision each keep their owner even when everyone contributesWorking with agents makes central the product responsibilities that used to be implicitBuilding alongside the product and building in its extension do not carry the same riskWhen other functions produce software, the developers' work shifts toward building the terrainThe quality of a probabilistic output is judged on its acceptability, not only on its conformityA tool that gives access to technical power does not hand over the culture that makes it safeLinks named with precise verbs describe a product where isolated definitions describe only wordsThe gap between the name in the code, the word on the screen and the word in the field is information to keep, not disorder to fixA product model passed on by word of mouth stops being passed on the moment the recipient is an agentA business rule stated in one sentence is already a test, without going through the codeUntil a reference model is actually in use, the rigour of its format cannot be chosenQuestions already put to support frame a reference model better than questions written coldA reference model is accepted by replaying the questions that framed it, without reopening the codeA link pointing to a card that does not exist is the work queue, not an error to fixA source that does not come from the version being described is not a sourceTwo parts of a product that contradict each other at the same version signal an inconsistency, not a lagA software object is only described once it has been read in the core code, the database, the screens and the translationsA rule that exists only in the screen is almost always a gapThe names of technical error messages state business rules the rest of the code formulates nowhereDividing software into modules approximates the business, dividing it into tables does notThe cohesion of a set of cards and the coverage of what they describe are two distinct checksWe pick a stopping criterion we can satisfy rather than one that proves the work was doneA stopping criterion looked for at the end is picked from among those you are already sure of satisfyingThe signal in a coverage measurement is the disagreement between two angles of enumeration, not the score of any one angleA meaning-based search engine always answers and never proves an absenceOne and the same rule copied into three documents only diverges on the day it changesA badly framed term count produces false data that looks like dataAn agent enumerates better than a human and declares the work finished sooner than oneA calculation with no single home is rediscovered by every new feature, at its own expenseAn object the user sees on screen without it existing in the product manufactures an entire family of complaintsInterviews say what customers want to build, tickets say what breaksAn estimation error of one order of magnitude is enough for a piece of work never to be startedA reference model complete over one scope serves every function working on it, a half-finished one over five serves nobodyAn append-only log distinguishes a change in the product from a change in our understandingA "what the method doesn't say yet" section is what makes a method document honestA maturing reference model moves through mergings and splittings of objects, movements no tool tells the story of The most useful moment in writing…A deliverable carries the result…Between sources and deliverables,…AI makes an intermediate layer pa…A well-phrased answer proves noth…The value of persistence lies in…Slowness blamed on the tool often…Handing an ancillary task to a to…"To be prioritized later" is a re…As long as the loss isn't named,…An arbitration reopened as soon a…The code is the only artifact tha…Clean code stops being a technica…Rebuilding a specification from t…A test written before the code tu…A coherence tool is worth only wh…An arbitration with no written me…Producing more code produces neit…A single file cannot be organised…Forcing a generated answer to cit…A coding assistant doesn't invent…
  • AI & context
  • Product management
  • Knowledge & notes
  • Business & management
  • No category