Series 4 of 4 · 30 chapters · 3 levels ·

AI PM
OPERATING SYSTEM

The SaaS playbook broke.
Per-seat pricing is dying.
Build the operating system.

Start with Topic 01 — Why AI PM ≠ SaaS PM → See the full map ↓
30 chapters 3 levels
Evidence · Selected

Three receipts. One operating system.

01
$7,225
Cursor's one-user invoice.

What happens when the SaaS pricing model meets the inference treadmill. The bill arrives before the seat does.

02
−14%
Replit's margins.

Negative unit economics on a flagship AI feature. Margin compression is the new churn.

03
95%
Pilots that never hit P&L.

Most AI initiatives die in PoC. Not because the model failed — because nobody priced the data prep tax.

L1 · FOUNDATIONS10 chapters

Act 01The Structural Rupture

By the end: you can defend why AI PM ≠ SaaS PM, kill bad initiatives before they eat a quarter, and put a unit cost on the first page of every PRD.

T01 Why AI PM ≠ SaaS PM — and why the harness is your highest-leverage skill The bridge from engineering to strategy. The litmus question that defines the rest of the series. READ T02 Why most AI products fail — three traps and the margin death spiral Cursor's $7,225 invoice. Replit's −14% margins. The cost of shipping AI on SaaS assumptions. READ T03 The Agentic PMF Standard — what changes when the user delegates A new PMF lens. The Indispensability Index. 7-Fits adapted for agents. The Agent Tax anti-pattern. READ T04 When AI is the right answer — Boring AI and the rules-vs-API-vs-RAG decision Say no when rules win. The build-or-buy call that saves teams from custom-model debt. READ T05 Taste at Speed — the death of the traditional PRD One-week prototype sprints. The 5-Lens scorecard. Pretotyping. Anthropic's Boris Cherny shipping 20–30 PRs a day. READ T06 Pressure-testing AI initiatives — the Cagan risk taxonomy Five tests — Value, Usability, Viability, Feasibility, Ethics — that kill bad initiatives before they eat a quarter. READ T07 AI Economics 101 — the Inference Treadmill Fuel dropped 280×. Bills grew 320%. The token tsunami at P90 and the cost cliff at 500–5,000 users. READ T08 Cost in Every PRD — Day-1 measurement and AI Cost per Output The new unit metric. The 20–40% data-prep tax most teams forget to budget. READ T09 Why Per-Seat Pricing Dies — and the four pricing models replacing it The SaaSpocalypse. Why willingness-to-pay comes before the roadmap, not after. READ T10 The Value Model — Measurement × Adoption = AI ROI A 7-step value model. The four ROI calculations CFOs actually use. Why 95% of pilots never hit P&L. READ
L2 · OPERATING DISCIPLINES10 chapters

Act 02Ship What the CFO Will Fund

From PoC to production with positive unit economics — and a board narrative that survives scrutiny.

T11 The 4D Strategic Framework — Discover, Design, Develop, Deploy The execution spine that takes any AI product from PoC to production. READ T12 Building Compounding Moats — Data, Distribution, Dogfooding, Design The four moats that survive commoditisation. Why Single-Trick Pony products die in ninety days. READ T13 Product Architecture as a Strategy Decision — Augmentation, Copilot, Agent A product taxonomy, not a technical one. Intelligence versus Judgment as the architectural abstraction. READ T14 AI Pricing Models — the five-band pricing spectrum Four archetypes plus outcome. The consumer-vs-API split. How Intercom's Fin hit $343M ARR at $0.99 a resolution. READ T15 Inference FinOps for PMs — the four disciplines and the five PM-owned levers Distillation. Semantic caching (90% off). Loop pruning (50–70%). Continuous batching. Cost levers in CFO language. READ T16 Evals as the New PRD — what changes about the PM workflow Data-Task-Scores. The offline-to-online flywheel. The HHH framework. Why Braintrust waited for Claude 3.7. READ T17 Stakeholder Translation — turning harness signals into board narratives Translate Context Durability, Intervention Rate, and eval scores into language CFOs, GCs, COOs, and CHROs can act on. READ T18 Privacy + Enterprise Readiness as Competitive Moat — the Apple-Google bet HITL, HOTL, HOOTL. The CAIR equation. Apple's Private Cloud Compute as masterclass. READ T19 The Outcome-Based Pricing Playbook — eight-or-nine-word operational definitions Outcome Measurement Agreements. The multi-touch attribution problem. Why only 5–17% of vendors ship outcome pricing successfully. READ
L3 · THE STRATEGIC LAYER10 chapters

Act 03Compounding Moats

Self-improving products. Multi-model orchestration. The golden quadrant where service becomes software.

T01 Reading the Harness as a PM — the seven CONTEXT layers Where the PM owns the harness, and where to delegate. What to specify, what to leave alone. READ T02 Compounding Feedback Loops as Your Moat — the Karpathy Loop pattern Micro, Meso, Macro layers. Agentic feedback loops as first-class events. Why Context Rot hits at 32K tokens. READ T03 Multi-Model Orchestration as a Cost-Strategy Decision — the 80/20 routing pattern The Cost / Capability / Speed triangle. SLMs (Phi-4) with frontier escalation. Mixture of Experts. READ T04 AI PM Team Structure for the Agentic Era — Builder PMs, Eval Engineers, the new shape How AI PM teams diverge from SaaS PM teams. The cross-functional product trio. Hiring criteria. Career path. READ T05 GTM-AI Fit + Agent Distribution — being the canonical answer inside other AI surfaces MCP, AGENTS.md, and the 5-Question Agent Audit. Stripe and Cloudflare patterns. Agentic.Market: $50M and 480K agents in seven days. READ T06 The Self-Improving Moat — the only moat that appreciates with time Karpathy's Autoresearch. Shopify's 53% render speedup from 93 automated commits. Self-correcting eval loops. READ T07 The Golden Quadrant — outcome pricing at scale and the Service-as-Software pivot The Autonomy × Attribution matrix. The SaaSpocalypse repricing event. Pricing-led enterprise transformation. READ T08 Vendor Strategy + AI Portfolio Governance — Trust vs Lock-In Multi-model strategy. Apple's Project Campos as the privacy-as-architecture case (Stateless AI, PCC, Gemini → Ferret-3 bridge). READ T09 Real ROI at Scale — the case study lens (XPO, Intercom Fin, ServiceNow, Shopify) JPMorgan COiN: 360,000 lawyer-hours saved. Klarna: $60M, 853 FTEs. Intercom: $343M ARR. The 5–20% capturing most of AI's value. READ
Bonus After the main thirty · Strategic-decision companions

The practitioner depth.

The main thirty stay vendor-neutral on principle. These companions answer the platform, vendor, and career questions your team will actually ask — each one a decision framework, not a reference doc. Together they are the operating manual the main thirty imply.

BONUS-T01 The AI PM Career Path The arc from SaaS PM to AI PM. What compounds, what decays, and the next rung worth optimising for. READ BONUS-T02 The Bridger Archetype By 2026 the structural shape of the AI PM role. Engineering depth + design discovery + business model + governance — in one operator. READ BONUS-T02·b The Dinner-Table / Boardroom Translator Five questions about a new hire — memory, trust, growth, economics, risk — that open the whole 2026 AI conversation for non-technical stakeholders. READ BONUS-T03 Salesforce Agentforce — Certified-Practitioner Deep Dive Mechanics, the hidden 2–3× TCO rule, the Data Cloud Mirror, the by-segment verdict, and the adoption numbers. The bet for or against Salesforce as the AI platform. READ BONUS-T03·b AI PM in the Indian Market Where the playbook bends — pricing power, data realities, distribution, and the talent compression that makes Bridgers scarce and decisive. READ BONUS-T04 Agentforce vs Microsoft Copilot — The CPO Comparison Which platform owns which workflows, where the boundaries fall, and why most mature enterprises run both. The Semantic Layer as the structural lens. READ BONUS-T04·b The Agent as Buyer When the customer of your product is another agent, not a human. What pricing, UX, and trust look like when machines transact on behalf of people. READ BONUS-T05 How the 5% Actually Operate The rituals, artefacts, meetings, and metrics that compound in the 5% of enterprise AI deployments that succeed. The operating manual the main thirty imply. READ BONUS-T05·b The Future of the AI PM Where the role lands in three years. The disciplines that compound, the ones that decay, and the bets worth making now. READ

The best AI PMs ship faster because they treat pricing, costs, and evals as part of the product — not afterthoughts.

Start with Topic 01 → See the full map ↓
Fin · Series 04 of 04 · Thank you for reading
Continue the syllabus. · Three next-hops, in order.
Read next Evals L1·T01 — Benchmarks ≠ Evals Then Harness L1·T01 — Why Your Agent Fails Or jump to Agentic L1·T01 — Prompt vs. Context