Premiers principes et psychologie produit
Développez l'état d'esprit opérationnel, les modèles mentaux et le jugement conscient des biais derrière chaque décision PM.
Explainer
La gestion de produit commence par une manière de voir. Avant les frameworks, les feuilles de route ou les métriques, le travail du PM est de traduire l'ambiguïté en décisions qui se cumulent. Cela suppose de tenir en même temps la réalité de l'utilisateur, l'économie de l'entreprise et les contraintes du système, et de choisir ce qu'il faut apprendre ensuite sans esquiver l'écart entre ce que vous savez et ce que vous supposez.
Le PM comme traducteur, pas comme mini-PDG
Le cadrage 'PM = mini-PDG' est trompeur. Les PMs ont rarement l'autorité d'embauche/licenciement sur leur équipe et ne possèdent jamais unilatéralement la capacité d'ingénierie. Le cadre plus utile est celui du traducteur : entre la vérité utilisateur et les résultats business, entre la stratégie et les tickets, entre ce qui est possible aujourd'hui et ce qui le sera après le prochain pari. L'influence vient de la qualité de la traduction, pas du titre.
- Traduisez la douleur utilisateur en l'énoncé de problème le plus petit que l'on puisse apprendre.
- Traduisez les objectifs business en changements de comportement observables que l'équipe peut viser.
- Traduisez les contraintes techniques en arbitrages produit que l'équipe peut débattre.
Utilisateur vs client vs acheteur vs champion
La plupart des erreurs PM commencent en confondant quatre parties prenantes différentes en un seul mot. L'utilisateur vit le produit ; le client paye ; l'acheteur signe ; le champion défend en interne. En B2C et prosumer, ces rôles fusionnent souvent en une personne. En B2B, santé, secteur public et marketplaces, ce sont des parties prenantes distinctes aux incitations contradictoires et à la disposition très différente d'agir.
- Utilisateur : tire de la valeur ou subit la friction de l'usage quotidien ; supporte le coût de bascule.
- Client : paye la facture ; se soucie du ROI, de la gouvernance, de la sécurité.
- Acheteur : signe le contrat ou approuve le déploiement ; se soucie du risque et de la politique.
- Champion : pousse l'adoption en interne ; se soucie de la crédibilité et des victoires visibles.
- Cartographiez-les par workflow avant de cartographier les fonctionnalités.
Le triangle produit : viabilité, utilisabilité, faisabilité
Toute décision produit s'inscrit dans un triangle : viabilité business (cela rapporte ou économise-t-il de l'argent ?), valeur utilisateur et utilisabilité (résout-il vraiment un vrai problème mieux que les alternatives ?) et faisabilité technique (peut-on construire, exploiter et faire évoluer cela avec notre équipe et notre stack ?). Les PMs solides circulent librement entre les trois lentilles ; les PMs faibles laissent l'une dominer par défaut — généralement celle avec laquelle ils sont le plus à l'aise.
- Viabilité : revenu, coût, économie de rétention, exposition réglementaire et contractuelle.
- Valeur et utilisabilité : demande démontrée, disposition à basculer, apprenabilité, accessibilité.
- Faisabilité : latence, fiabilité, sécurité, observabilité, maintenance, capacité de l'équipe.
- Quand les parties prenantes divergent, nommez quel coin du triangle chacune défend — la plupart des désaccords sont des chocs de coin, pas des faits.
Résultats vs livrables vs activités
La distinction de Marty Cagan est l'une des plus utiles en produit. Les livrables sont les choses que vous expédiez ; les activités sont les choses que vous faites ; les résultats sont les changements de comportement utilisateur ou de performance business que vous provoquez. La plupart des équipes mesurent des activités ('on a fait des entretiens découverte cette semaine'), certaines mesurent des livrables ('on a livré 8 fonctionnalités'), et seules les meilleures mesurent les résultats ('les éditeurs actifs hebdomadaires sont passés de 41 % à 49 % dans la nouvelle cohorte'). Les feuilles de route pilotées par le livrable semblent productives mais ne bougent rarement le business.
- Les activités sont les plus faciles à feindre et les plus courantes dans les rapports de statut.
- Les livrables ressemblent à du progrès mais ne comptent que si un comportement utilisateur change.
- Les résultats forcent les équipes à admettre quand le travail n'a pas bougé la métrique.
- Reformulez tout énoncé de livrable en demandant : 'Quel comportement utilisateur devrait changer à cause de ça, et comment le saurait-on ?'
Décisions de type 1 vs type 2
Distinction de Jeff Bezos : les décisions de type 1 sont des portes à sens unique — difficiles ou impossibles à inverser (choix d'architecture, changements publics de tarification, M&A). Les décisions de type 2 sont des portes à double sens — faciles à inverser (la plupart des lancements de fonctionnalités, changements de copy, expériences de prix sur une seule cohorte). Traiter les décisions de type 2 comme du type 1 est le frein le plus courant à la vitesse. Traiter celles de type 1 comme du type 2 est la source la plus courante de désastre.
- Privilégiez l'action sur les décisions de type 2 ; expédiez vite, apprenez vite, inversez si c'est faux.
- Ralentissez sur les décisions de type 1 ; convoquez les voix les plus fortes, faites un pre-mortem.
- Dans le doute, demandez explicitement : 'Comment inverserait-on cela en 7 jours si c'est faux ?'
Pensée par premiers principes
Raisonner par analogie est rapide et lossy ('faisons comme Stripe'). Raisonner par premiers principes est plus lent et plus durable : décomposez un problème jusqu'à des faits indiscutables, puis reconstruisez. Les PMs s'en servent pour challenger des hypothèses héritées ('on a toujours facturé au siège'), pour estimer à partir de zéro quand aucun benchmark n'existe, et pour repérer les stratégies copiées-collées qui ne collent pas aux contraintes réelles de l'équipe.
Modèles mentaux que les PMs devraient intérioriser
Les modèles mentaux sont des raccourcis de pensée réutilisables qui compressent l'expérience. L'idée n'est pas de les mémoriser mais d'en garder une demi-douzaine sur le bout de la langue pour pouvoir choisir le bon en pleine conversation.
- Inversion : au lieu de demander 'comment réussir ?', demandez 'comment garantir l'échec ?' et évitez ces voies.
- Pre-mortem : supposez que le lancement a échoué et écrivez le post-mortem avant l'expédition.
- Pensée de second ordre : 'et ensuite ?' Cartographiez les conséquences des conséquences.
- Steel-manning : énoncez la position adverse plus fort que l'adversaire ne le pourrait avant d'être en désaccord.
- Falsifiabilité : une croyance qu'on ne peut jamais réfuter n'est pas une croyance utile.
- Rasoir de Hanlon : n'attribuez jamais à la malveillance ce qui s'explique adéquatement par un manque de contexte.
- Barrière de Chesterton : ne supprimez pas une contrainte avant de comprendre pourquoi elle a été posée.
Archétypes de PM (et pourquoi ils comptent)
Le PM est l'un des rôles les plus larges en tech. Le quotidien d'un Growth PM est radicalement différent de celui d'un PM Plateforme. Savoir à quel archétype votre rôle se rattache vous aide à choisir les bonnes métriques, partenaires et compétences à investir.
- Product Lead PM : surfaces utilisateur final ; forte collaboration design ; métriques d'activation et de rétention.
- Growth PM : axé expérimentation ; centré entonnoir ; partenaire de marketing et data ; métriques AARRR.
- Platform / Infra PM : interne ou orienté développeurs ; ergonomie, fiabilité, adoption ; partenaire de l'ingénierie sur la roadmap des capacités.
- Technical PM : surface profondément technique (APIs, SDKs, produits data, systèmes ML) ; rédige souvent les specs en termes quasi-protocolaires.
- Data / ML PM : ground truth, évaluation, comportement modèle, droits sur les datasets ; partenaire de la recherche.
- Internal Tools PM : les employés sont les utilisateurs ; le ROI se mesure en heures gagnées et erreurs évitées ; la politique est la partie difficile.
Frameworks de décision : DACI, RAPID, RACI
La vitesse en produit tient surtout à la clarté sur qui décide quoi. DACI (Driver, Approver, Contributors, Informed) et RAPID (Recommend, Agree, Perform, Input, Decide) sont les plus utiles pour les décisions produit. RACI est plus courant pour les déploiements opérationnels. Choisissez-en un et utilisez-le de manière cohérente — le format compte moins que la discipline de nommer un seul Approver / Decide.
- Driver / Recommend : rédige la proposition ; généralement le PM.
- Approver / Decide : tranche ; une seule personne, nommée explicitement.
- Contributors / Input : apportent leur expertise ; leur rôle est d'être entendus, pas de mettre un veto.
- Informed : tenus au courant ; n'ont pas besoin de se prononcer.
Cartographie des parties prenantes
La carte des parties prenantes est l'artefact PM le plus sous-utilisé. Placez les parties prenantes sur une matrice 2x2 pouvoir / intérêt. Pouvoir élevé / intérêt élevé : co-auteurs du plan. Pouvoir élevé / faible intérêt : tenus informés mais pas consultés sur chaque détail. Pouvoir faible / intérêt élevé : généralement vos évangélistes. Pouvoir faible / intérêt faible : du bruit.
- Rafraîchissez la carte au début de chaque trimestre.
- Nommez l'enjeu personnel de chaque partie prenante — risque de carrière, charge d'équipe, impact sur la marque.
- Planifiez des points de contact délibérés, pas des updates de couloir fortuits.
Le système d'exploitation du PM
Les grands PMs fonctionnent à un rythme hebdomadaire — pas à coups d'héroïsme. Un rythme simple : revue des résultats le lundi (où en sont les métriques par rapport au plan ?), découverte et travail écrit asynchrone en milieu de semaine, session cross-fonctionnelle mercredi ou jeudi, note hebdomadaire le vendredi. Remplacez les réunions de statut par une note hebdomadaire d'une page ; remplacez les demandes ad-hoc par une file de triage.
- Privilégiez la communication écrite asynchrone ; réservez le synchrone aux décisions et désaccords.
- Maintenez un 'now / next / later' public pour que les parties prenantes s'auto-renseignent.
- Tenez un journal privé d''inconnues' : chaque hypothèse que vous n'avez pas encore testée.
- Bloquez du temps pour le travail profond ; le tetris d'agenda est un jeu de statut, pas une stratégie de productivité.
Framework atlas
Reference cards for each method in this mission
Expand a card for when to deploy it, misuse patterns, sequencing guidance, and (where relevant) shorthand formulas.
Decision-making · Atlassian / IntuitDACI(DACI)
A four-role decision framework that names exactly one Driver, exactly one Approver, a small set of Contributors, and a list of Informed parties for every meaningful decision.
When to use
- Cross-functional decisions involving 3+ teams.
- Decisions where you've felt 'no one is in charge' before.
- Recurring quarterly planning or roadmap calls.
When not to
- Truly trivial Type-2 decisions inside a single team.
- When the team is small enough that ad-hoc decisions are cheaper than the framework overhead.
How to apply
- Name the decision in one sentence with a deadline.
- Assign a single Driver who is responsible for moving the decision to a yes/no.
- Assign a single Approver — the person who will make the final call.
- List Contributors and explicitly note their expertise window.
- List Informed parties and how they will be told.
- Letting Approver be a list of people: that is a committee, not an approver.
- Treating Contributors as veto-holders.
- Skipping the Informed list — usually the source of last-minute drama.
Decision-making · Bain & CompanyRAPID(RAPID)
Recommend, Agree, Perform, Input, Decide. Heavier than DACI but useful when decisions cross legal/regulatory boundaries or have material financial consequences.
When to use
- Decisions with regulatory or contractual exposure.
- Decisions where 'Agree' (a soft veto, often legal/security) is meaningfully different from 'Input'.
When not to
- Day-to-day product trade-offs.
- Where DACI's simpler structure is enough.
How to apply
- Recommend: who frames the proposal.
- Agree: who must sign off (legal, security, finance).
- Perform: who executes the decision once made.
- Input: who provides expertise.
- Decide: who makes the final call.
- Confusing Agree with Decide; Agree is a veto on a specific dimension, not the call.
- Routing too many decisions through RAPID — the overhead defeats the purpose.
Risk surfacing · Gary KleinPre-Mortem
Imagine the launch failed, then write the post-mortem in advance. Surfaces the risks people are too polite or too anchored to raise during normal planning.
When to use
- Any Type-1 decision.
- Launches with reputational, regulatory, or financial exposure.
- When the team is unusually optimistic — 'this can't fail' is the alarm.
When not to
- Reversible micro-experiments where the cost of failure is one cohort and one week.
How to apply
- Set the scene: 'It's six months from now and the project has clearly failed.'
- Each person writes silently for 5 minutes: why did it fail?
- Cluster the failure modes; rank by likelihood and impact.
- Add the top 3-5 to your discovery and execution plans as risks to retire.
Personal prioritization · Dwight D. EisenhowerEisenhower Matrix
A 2x2 of urgent vs important. Urgent + important: do now. Important but not urgent: schedule. Urgent but not important: delegate. Neither: drop. Useful for personal triage as much as feature triage.
When to use
- Triage of an overflowing inbox or backlog.
- Differentiating reactive support work from strategic work in your week.
When not to
- Real product roadmap prioritization — use RICE/ICE/Kano instead, since 'important' is too vague at scale.
How to apply
- List every open ask or commitment.
- Tag each as urgent or not urgent, important or not important.
- Triage: do, schedule, delegate, drop.
- Treating 'urgent because someone asked loudly' as 'urgent because deadline'.
- Letting the 'schedule' bucket become a procrastination graveyard.
Team alignment · Andy Grove / Jeff BezosDisagree and Commit
Once a decision is made, even those who disagreed must commit fully. Surfaces dissent before the decision; produces unity after. The opposite of 'malicious compliance'.
When to use
- After a Type-2 decision where you've heard the dissent and made the call.
- When the team has a culture of relitigating decisions in side channels.
When not to
- When the dissent is about safety, ethics, or regulatory compliance — those should not be silenced.
How to apply
- Make the decision-making process clear up front (DACI).
- Explicitly invite dissent before the call.
- Once made, ask: 'Can you commit?' — accept yes or 'I disagree but will commit'.
- Set a review date so the disagreement has a future audience.
Mental model · G.K. ChestertonChesterton's Fence
Don't remove a constraint, process, or feature until you understand why it was put there. Many 'obviously dumb' systems are scar tissue from past failures.
When to use
- Inheriting a codebase, product, or process that 'should obviously be simplified'.
- Stakeholder pushback that seems irrational — find the original incident.
When not to
- When the original reason is documented and clearly no longer applies.
How to apply
- Identify the fence (rule, feature, ceremony).
- Talk to the longest-tenured person who remembers it.
- Search post-mortems and incident reports.
- If the original reason is gone, remove with a clear migration plan.
Product Psychology
Cognitive biases that distort product decisions
Confirmation Bias
The tendency to seek, interpret, and remember information that supports existing beliefs while discounting contradictory evidence.
Product Risk
Teams cherry-pick supportive interview quotes and ignore data that would invalidate the favored solution.
Research Countermove
Before discovery, write down what evidence would falsify the hypothesis. Track wins and losses for the hypothesis with equal rigor.
Survivorship Bias
Overweighting visible successes while ignoring the users, products, or companies that failed and dropped out of the dataset.
Product Risk
A team copies practices from breakout products without seeing the dozens of failed products that used the same tactics.
Research Countermove
Study churned users, failed experiments, lost deals, and abandoned workflows with the same rigor as power users and case studies.
Anchoring
The first number, framing, or example heard exerts disproportionate influence on subsequent judgments.
Product Risk
A loud stakeholder's number becomes the implicit forecast, and every estimate gets adjusted relative to it instead of from facts.
Research Countermove
Have estimators write numbers privately before discussion; use base rates and historical reference classes; explicitly call out the anchor in the meeting.
Sunk Cost Fallacy
Continuing investment in a course of action because of resources already spent rather than expected future return.
Product Risk
Multi-quarter initiatives keep getting funded because killing them would 'waste' the prior work, even when the forward expected value is negative.
Research Countermove
Run a 'fresh-eyes' review every quarter: 'If we were starting today with what we know, would we fund this?' If no, redirect.
IKEA Effect
People disproportionately value things they have built themselves. Internal teams systematically over-rate their own products and processes.
Product Risk
PM and engineering favor in-house tools, dashboards, or features over better-fitting third-party options because they remember the cost of building.
Research Countermove
Use blind comparisons in evaluations. Have someone outside the team rate the artifact. Include the maintenance cost forecast, not just the build cost.
Curse of Knowledge
Once you understand something, it becomes hard to imagine not understanding it; experts systematically overestimate how clear their explanations are.
Product Risk
PMs assume users will 'get' a workflow because the team gets it; copy and onboarding read fluently to the team and incomprehensibly to first-time users.
Research Countermove
Run unmoderated tests with users who have never seen the product. Treat 'I don't know what to do here' as the most valuable data of the week.
Loss Aversion
Losses feel roughly twice as painful as equivalent gains feel pleasurable. Users will do irrational work to avoid losing something they have.
Product Risk
Teams under-weight removal cost when sunsetting features; users churn over a small removed capability they barely used.
Research Countermove
When removing or migrating, frame the change as a gain users earn (more reliable, faster, less buggy). Provide transition periods. Communicate before the change, not after.
Availability Heuristic
Recent or vivid examples dominate judgment regardless of whether they're representative.
Product Risk
One angry executive escalation reshapes the roadmap because it's emotionally fresh, even though hundreds of users had a different problem.
Research Countermove
Aggregate evidence in a recurring view (support themes, NPS verbatims, churn reasons). Force the question: 'How big is this problem in the dataset, not in my inbox?'
Planning Fallacy
Systematic underestimation of how long tasks will take and how much they will cost, even when the estimator has experienced the same kind of task underrunning before.
Product Risk
Sprint commits, quarterly OKRs, and launch plans regularly slip; trust in the team's word erodes over time.
Research Countermove
Use reference-class forecasting: how long did the last 3 similar projects actually take? Multiply optimistic estimates by the historical slip factor. Plan for the unknown unknowns explicitly.
Action Bias
A preference for action over inaction, even when the available actions have negative expected value.
Product Risk
Teams ship features to 'be seen doing something' after a metric drop, before the cause is understood; the feature itself becomes the new variable to debug.
Research Countermove
When metrics drop, isolate the cause before shipping a fix. Track which 'fixes' actually moved the metric vs which were activity for activity's sake.
Authority Bias
Disproportionate weight given to the opinions of authority figures regardless of the underlying evidence.
Product Risk
An executive's casual comment becomes a quarterly goal because nobody pushes back on it.
Research Countermove
Separate the message from the messenger. Restate the executive's idea as a hypothesis, then evaluate it on the evidence the same way as any other hypothesis.
Bandwagon Effect
Beliefs and practices spread because others are adopting them, regardless of whether they fit the local context.
Product Risk
Teams adopt frameworks (OKRs, Shape Up, ICE) because peer companies do, then blame the framework when context-mismatch causes failure.
Research Countermove
Before adopting any practice, articulate the specific problem it solves for *you*. If you can't, you're cargo-culting.
Status Quo Bias
Preference for the current state of affairs; alternatives are viewed as losses relative to the status quo.
Product Risk
Legacy product behaviors are protected even when data shows they hurt new users; 'we can't break existing users' becomes an unconditional veto.
Research Countermove
Quantify the cost of the status quo. Often the supposed risk of change is dwarfed by the ongoing cost of stagnation.
Dunning-Kruger Effect
Low ability in a domain often correlates with overconfidence; competent practitioners are paradoxically more aware of what they don't know.
Product Risk
Junior PMs ship confident roadmaps in domains they barely understand; senior PMs hedge so much they look indecisive.
Research Countermove
For confident-feeling claims, ask: 'What's the strongest argument against this?' If you can't make one, you don't understand the area yet.
Rhyme-as-Reason Effect
Catchy or rhyming statements feel more truthful than equivalently precise but less elegant statements.
Product Risk
Memorable research summaries get repeated as roadmap-driving truths even when the underlying evidence is weak.
Research Countermove
Translate any catchy phrase into a falsifiable assumption tied to observed behavior or data.
Organizational anti-patterns
When ceremonies look like rigor but aren't
The Feature Factory
Roadmap progress is measured in features shipped per quarter. Outcomes are reported only when convenient. Discovery is whatever week 1 of the sprint is.
Output is easy to count; outcomes take time and honesty. Activity becomes a proxy for impact in performance reviews.
Re-base every roadmap item on a target metric movement. Kill the feature count metric in reviews. Celebrate killed bad bets the same as shipped good ones.
Stakeholder Ventriloquism
PM justifies decisions with 'the CEO wants it' or 'sales said'; no first-hand user evidence is presented; team feels like a delivery service.
PM is conflict-avoidant or under-informed and uses authority laundering to push items.
Translate every stakeholder ask into a user problem and an evidence list before bringing it to engineering. If the evidence is weak, push back upstream.
Solutioneering
Specs jump straight to UI mocks. The first slide is a screen, not a problem. Discovery, if any, happens after design starts.
Solutions are concrete and exciting; problems are abstract and uncomfortable.
Write the problem statement, the user, the current alternative, and the success metric on page 1 of every spec, before any UI.
The Roadmap as Wishlist
The roadmap has 40 items, all H2 priorities, no sequencing rationale. Every quarter, the team commits to all of it and finishes a third.
PM uses the roadmap to placate stakeholders rather than to make a bet.
Cap the roadmap at 'now / next / later' with no more than 3-5 themes per quarter. Force trade-off conversations up front, not at the end.
Discovery Theater
Team runs 'discovery' that consists of a few user interviews after the spec is written. Findings are summarized to confirm the existing plan.
Discovery has been adopted as a ritual rather than a learning loop.
Define the riskiest assumption and the falsification criterion *before* the interviews. Reserve the right to change the spec or kill the project based on findings.
Outcome Theater
Every project is suddenly tied to a vague outcome ('improve user experience'); nobody can agree on which metric, baseline, or threshold counts as success.
Teams have heard 'be outcome-focused' but lack rigor in defining and measuring outcomes.
Force the trio: metric, baseline, target, by-when. If you can't write all four for an initiative, it's an output dressed up as an outcome.
Worked examples
Walkthroughs translated from real trade-off rooms
Reframing 'Make signup easier'
A B2B SaaS PM gets a request from sales: 'We're losing trial users at signup. Make signup easier.'
- Resist the solution: 'easier' is not falsifiable.
- Define the user behavior: 'reach first valuable action within 10 minutes of email click.'
- Pull the data: 38% of trials drop off between email confirm and team invite step.
- Run 5 unmoderated tests on the existing flow; observe friction points.
- Propose two scoped experiments: lazy team invite (Type-2, ship in a week) vs. SSO at signup (Type-1, scope it out properly).
TakeawayThe feature request 'make signup easier' became two falsifiable bets attached to a measurable behavior, with one fast Type-2 experiment to learn before the Type-1 commitment.
Saying no to an executive ask using the Product Triangle
CEO wants a chat widget added to the marketing site this quarter to 'engage prospects'.
- Acknowledge the goal: more qualified pipeline.
- Walk the triangle: Viability — what's the expected lift on qualified pipeline? Usability — does the prospect actually want a synchronous chat right now? Feasibility — staffing chat in three time zones during a launch quarter.
- Present two alternatives: a smaller asynchronous form with reply SLA, or a chat widget gated to qualified accounts only.
- Frame the decision with DACI: Driver: PM; Approver: CEO; Contributors: Sales, CS, Design, Eng.
TakeawaySaid no to the literal ask, said yes to the underlying goal, and gave the CEO a real decision instead of a polite refusal.
Catching a confirmation bias in a discovery sprint
Team is convinced the next big bet is collaborative editing. After 8 interviews, the deck of quotes is overwhelmingly supportive.
- Pause and write the falsifying criterion: 'If fewer than 30% of users describe a recent collaborative-editing pain spontaneously, the demand is weaker than we think.'
- Re-tag the interview corpus by spontaneity: how many quotes were unprompted vs. prompted?
- Find that most supportive quotes were prompted; only 18% were unprompted.
- Reframe the bet: collaborative editing is desired in concept but not yet a top-of-mind pain. Move it from Q1 to Q3 and back-fill with a real top-of-mind pain.
TakeawayConfirmation bias quietly inflated the perceived demand. A simple falsification check rebalanced the roadmap.
Resources / Case Studies
Curated reading for this mission
Ben Horowitz
A 1996 memo on the operating standards expected of strong product managers vs the failure modes of weak ones.
Anchors the product mindset around responsibility for outcomes rather than activity or process theater.
Marty Cagan (SVPG)
The canonical text on modern product management. Covers product discovery, the product trio, outcomes vs outputs, and the role of the empowered product team.
Defines the vocabulary the rest of the industry uses. Required reading.
Marty Cagan & Chris Jones (SVPG)
Companion volume to INSPIRED. Focuses on what product leaders must do to enable empowered product teams to find and solve real problems.
Closes the gap between PM craft and the organizational system that lets that craft compound.
Lenny Rachitsky
Concrete decomposition of a PM's week, with time allocations for discovery, delivery, alignment, and management.
Demystifies the job for new PMs and gives experienced PMs a benchmark for where they may be over- or under-investing.
Shreyas Doshi
How high-agency PMs operate: refusing the false constraints, finding leverage, and translating ambiguity into action.
The single most concentrated talk on the operating mindset that distinguishes great PMs from competent ones.
Julie Zhuo
Practical, candid guide to the early years of management — applicable to senior PMs whose work is increasingly through other people.
Bridges the IC PM and Lead PM transition without the usual MBA fluff.
Annie Duke
Decision-making under uncertainty, drawn from poker. Separates outcome quality from decision quality.
Teaches PMs to evaluate decisions on the inputs available at the time, not on hindsight, which is the foundation of honest postmortems.
Daniel Kahneman
The definitive popular treatment of cognitive biases, dual-system thinking, and prospect theory.
Underwrites every bias section in this curriculum and most of behavioral economics in product.
Teresa Torres
Operationalizes discovery as a weekly habit rather than a phase. Introduces the Opportunity Solution Tree.
Sets a practical baseline for PM-design-eng trios that want to move from feature requests to demand-led product decisions.
Marty Cagan (SVPG)
Cagan's articulation of how product teams should be structured and held accountable: empowered teams, outcomes, and discovery as a continuous practice.
Crystallizes the operating model assumed in the rest of the curriculum.
Jeff Bezos (1997 Shareholder Letter)
The Type-1 / Type-2 distinction, in Bezos's own words, embedded in his 2015 letter to shareholders.
Provides the reusable frame for matching decision speed to reversibility — a daily PM tool.
Lenny Rachitsky
Long-form interviews with senior PMs and founders covering discovery, growth, monetization, and leadership.
The single best ongoing audio source for current PM practice across companies and stages.