Leadership

Leadership in the AI era spans people management, AI-augmented executive workflows, investor communication, and emerging AI governance. Keith Rabois's talent framework (expand scope until breaks, test small problems first, monitor desk traffic) applies equally to human and AI team management. CPO role predicted to vanish within five years as AI-native companies replace PM/design/engineering with "product builder" archetype spanning all three. CEOs use Claude Code as AI Chief of Staff doubling productivity: unifying inboxes, managing overnight todos, enriching contacts from transcripts, receiving strategic pushback. AI dissolves time barriers degrading leadership discipline—Friday reviews compress from hours to 12 minutes, 1:1 prep from hours to 5 minutes, removing procrastination excuses skipping high-leverage reflection. Hannah Stulberg (DoorDash) built shared repo where teams check call summaries, decision logs, analytics queries enabling 15-second natural-language queries returning full reasoning without pulling humans off work. Four independent implementations (DoorDash, Pendo, Google, solo) converged on same three-layer team-knowledge architecture—pattern is structurally robust, not idiosyncratic. One-command skill converts personal PM OS into team OS without leaking context, turning individual gains into team-wide compounding. Rippling PMs fix own copy errors rather than relying separate teams—ownership reduces coordination overhead. AI-native companies replacing standalone PM role with full-stack product builder (product+design+engineering IC) because standalone CPO creates coordination tax when ICs already blending. Career implication: PMs should develop fluency across product, design, engineering, analytics to become full-stack product builders; standalone specialization becomes overhead. AI-native company structures propagate to non-AI companies within 5 years. Agent governance should follow constitutional design: AI systems with distinct invested values (transparency, equity, due process) checking/balancing other AI systems; single concentration of intelligence should not self-regulate. SEC example illustrates gap: business school graduates with Excel combating AI-augmented trading is structurally inadequate—governments need AI-powered oversight matching AI-powered actors. Dorsey's four-layer AI-native org: Capabilities (hardware/models), World Model (unified vector DB company memory), Intelligence Layer (agent fleet deciding), Surfaces (human interaction)—any company can map this. DRI system applied to agents: temporary teams around specific 90-day goals, agents return to pool after, learnings absorbed into organizational brain. Agency/consulting new model: internal AI implementation becomes product—months compounded data and operational learnings become differentiation. Five moats surviving AI: compounding proprietary data (living, not static), network effects, regulatory permission, capital at scale ($20B chip fabs, $10B nuclear plants), physical infrastructure—all bottlenecked by time unparallelizable. Capital at scale underweighted moat—when bottleneck shifts software to atoms, financing/deployment at massive scale plus institutional trust becomes defining. Open question: does trust become moat when AI does more work? Institution bearing liability when things fail might become MORE valuable. Marek Siliski's clipboard trick (printed team photos with note space per person) lets exec ramp faster than any new peer seen, converting relationship-building into deliberate practice with checkpoints. Premortem prompt (Kahneman's most-valued decision technique, Google, Goldman Sachs, P&G; turned Claude pattern: "it's 6 months from now and this is dead") flips training-induced optimism because premise says already failed. Proper premortem returns: most likely failure, most dangerous failure, single biggest hidden assumption (often most valuable), revised plan closing gaps—counters confirmation bias on high-stakes decisions.