Reading Guide
The division of labor in AI product teams can first be discussed around three questions: who identifies user problems, who turns ideas into verifiable prototypes, and who is responsible for launch quality. The different positions collected in this article emphasize overlapping responsibilities, but this should not be taken to mean that every company should eliminate role divisions.
Read the consensus in Section 1 and the disagreements in Section 2 first, then read the product, design, or engineering perspective according to your current role. Section 9 provides a structured guide for choosing collaboration paths by role, which is suitable for discussing actual team constraints rather than copying a particular company's organizational structure.
You can validate this with your next small feature review: write down the user problem, prototype owner, acceptance criteria, and final responsible person in advance. The historical interviews, domestic job requirements, and independent product practices in the original text each have their applicable scenarios and are not a universal conclusion for all AI teams.
Recommended further reading:
When to open: You are transitioning to AI PM / working as Design / Eng at an AI company / want to understand how Anthropic+OpenAI operate internally. This wiki integrates 4 schools of perspective and explicitly marks 3 core disagreements.
One-sentence core takeaway:There is a universal consensus inside Anthropic + OpenAI that "role boundaries are disappearing" (PM/Eng/Design merging), but the domestic / classic school / Lao Huang independent product school disagree on this — the classic school insists on the binary division of "startup PM vs professional PM"; the domestic AI interview school emphasizes that "AI PMs must understand algorithms + data closed loops".
0. Positions of the 4 schools on "PM in the AI era"
| Perspective | Source | Position | One-sentence core takeaway |
|---|---|---|---|
| Anthropic+OpenAI internal school | Lenny's Newsletter / 2026 Three-Person Talk on Organizational Operations in the AI Era (Anthropic + OpenAI) | Role boundaries are disappearing / ship features weekly | "PMs, engineers, and designers are merging"(Cat Wu) |
| Domestic AI PM transition school | AI Product Manager Interviews and Transitions / Compilation of 12 Resources(12 resources) | Transitioning from internet PM to AI PM requires understanding algorithms + data closed loops | "6 Major Differences Between AI PMs and Internet PMs" |
| Independent product school | newtype · Vibe Coding Independent Product Practice / newtype mindset · Super-individual methodology(Huang Yihe) | One person covers PM/Eng/Design entirely | "Three Unfair Advantage capabilities / use strengths as the skeleton" |
| Classic product methodology school | Yitang / Wang Huiwen's Tsinghua Product Class 66-page Complete Notes(Wang Huiwen) | Binary of startup PM vs professional PM | "Product manager is the position closest to the CEO / related to one's own worldview/values/way of thinking" |
1. Cross-author consensus (5 points)
1.1 "Role boundaries are disappearing" (unanimous inside Anthropic+OpenAI)
| Company | Statement |
|---|---|
| Anthropic | "PMs, engineers, and designers are merging." (Cat Wu)<br>"Designers are becoming more and more like PMs and Engineers now" (Jenny Wen) |
| OpenAI | "Boundaries blurring. PMs feel more empowered — they can use Codex themselves to answer questions / build prototypes / do data analysis" (Embiricos) |
Embiricos confirms:Codex's impact on 3 roles:
| Role | Impact |
|---|---|
| Engineers | Improved efficiency in writing complex code + reduced burden of Code review/validation + shift from manual coding → higher-level thinking |
| PM | Cross communication boundaries, answer questions yourself / build prototypes / do data analysis / compressed talent stack |
| Designer | Directly build animations + APP prototypes → quickly go from idea to product |
1.2 Scarcity in the AI era ≠ coding ability
Lenny's Newsletter / 2026 Three-Person Talk on Organizational Operations in the AI Era (Anthropic + OpenAI) §I.2 Key quotes:
| Quote | Source |
|---|---|
| "As code becomes much cheaper to write, the thing that becomes more valuable is deciding what to write." | Cat Wu |
| "Product taste > engineering ability" | Cat Wu |
| "Execution + understanding customers > the idea itself" | Embiricos |
| "AI can weigh in, but it can't necessarily solve disputes between people about what should go into a feature." | Jenny Wen |
Translation:The cheaper writing code becomes, the more valuable "deciding what to write" is;AI can give opinions, but when people argue about "what a feature should do", AI cannot resolve it.
1.3 Just do things / cross team boundaries (shared product philosophy)
| Philosophy | Proposer |
|---|---|
| Just do things(Just do it) — first principles + cross team boundaries + reject rigid job roles | Cat Wu |
| Be kind and candid(Be kind and candid) — directness is a form of kindness | Embiricos |
| It is what it is(It is what it is, accept it) — stay calm in the face of the uncontrollable, don't be discouraged | Jenny Wen |
Lao Huang confirms(newtype mindset · Super-individual methodology):JUST DO IT —— Don't wait until you finish learning to start doing.
1.4 Product taste / aesthetic judgment is core
| Perspective | Position |
|---|---|
| Cat Wu | "Hire many PMs with engineering backgrounds because of taste; taste = aesthetic judgment / discernment / judgment" |
| Wang Huiwen(Yitang / Wang Huiwen's Tsinghua Product Class 66-page Complete Notes) | "Product management has nothing to do with major,it is related to one's own worldview/values/way of thinking. If your worldview doesn't match the field, you won't achieve much" |
| Xiaohongshu vibecoding | "Models don't matter, aesthetic judgment is the primary productive force"(See Aesthetic judgment / democratization of creation (cross-author theme synthesis) Batch 9) |
| Jobs(as recounted by Wang Huiwen) | "Whether a product is good is ultimately determined by taste" |
1.5 Failures are far more common than successes (consistent from the classic school to internal AI teams)
| School | Statement |
|---|---|
| Wang Huiwen | "Failures are 30 times more common than successes" / 5 failure-success pairs (Nokia/iPhone, MPMan/iPod, Kik/WeChat, Smart ABC/Sogou, IE/Chrome) |
| Cat Wu | "Remove crutches" —— Cloud Code's early to-do list was built to help models do large refactors; when new models no longer needed it, it was removed |
| Karpathy(Scaling Law / AGI timeline / Software 3.0 (cross-author theme synthesis)) | "Demos work well on carefully selected inputs; products must not break on all real-world inputs / 1000x difficulty" |
2. Explicit disagreements (3 points — not smoothed over)
2.1 Disagreement 1: Must AI PMs understand algorithms?
| Perspective | Position |
|---|---|
| Domestic AI PM transition school(AI Product Manager Interviews and Transitions / Compilation of 12 Resources) | Must understand —— Among 100 interview questions, technical questions across 5 categories (NLP / CV / speech / large models / data) make up the majority |
| Cat Wu(Anthropic) | Don't need to understand the underlying details, but must understand "model evaluation capability + first principles" —— One of the 5 key PM capabilities is Eval |
| Lao Huang(non-technical PM) | "Must have programming thinking, don't need programming experience" —— Learn while doing, but start from the most basic concepts |
| Wang Huiwen | "Product management has nothing to do with major; it is related to one's own worldview/values/way of thinking" —— Pichai studied metallurgical engineering / Zhang Xiaolong was an engineer by background, both worked |
Real disagreement:Domestic vs overseas AI PM paths:
- Domestic AI PM transition → must understand algorithms/data closed loops/speech/CV (to prepare for 100 interview questions)
- Overseas Anthropic/OpenAI internal PM → Taste + Eval + first principles + cross boundaries to fill gaps
- Independent product non-technical PM → must have programming thinking, no need to memorize specific syntax
2.2 Disagreement 2: Should Designers write code?
Jenny Wen's core proposition:"The Design Process is Dead"(The design process is dead).
3 characteristics of the old process:
- Extensive research / discovery
- Fixed divergence-convergence cycle
- Beautiful long-term decks
Why must it be abandoned?:
- Engineers can now quickly build features(build features quickly, relying on Cloud Code)
- Designers blocking engineers = unnecessary bottleneck
- Vision timeline has shortened to 3-6 months(prototypes instead of long-term decks)
Dramatic shift in Designer time allocation:
| Task | A few years ago | Now (2026) |
|---|---|---|
| Mocking / Prototyping | 60-70% | 30-40% |
| Collaborate directly with engineers | 20% | 30-40% |
| Implement / Polish in code | Almost 0 | 20-30% |
| Cross-team coordination | 10% | 10% |
3 types of valuable Designer prototypes:
| Prototype | Characteristics |
|---|---|
| Strong Generalist(Strong generalist) | Strong in multiple core skills(similar to T-shape but more "block-shaped") — easily adapts to expanding PM + Eng boundaries |
| Deep Specialist(Deep specialist) | Deeply focused on one domain (software engineering / visual design / icon design) — differentiated product |
| Craft New Grad(Craft-minded new grad) | Early career,blank slate + eager to learn — not bound by traditional processes, brings new perspectives |
Real disagreement:Designer transition directions:
- Anthropic position:Designers must be able to write code to polish 20–30% of the time(Jenny Wen)
- Classic school position: Designers don't write code, only draw mocks
- Lao Huang independent product position:Keynote replaces v0.dev(Go Fundamental mindset)
2.3 Disagreement 3: PM team structure — 30 people vs 2000 people?
| Perspective | Position |
|---|---|
| Cat Wu(Anthropic) | 30-40 person PM team / Research preview model / ship features weekly |
| Wang Huiwen(Meituan) | 2000 PMs, over 300 in the food delivery business / binary of startup PM vs professional PM |
| Lao Huang(independent product) | 1 person = PM + Eng + Design + Marketing / Unit Economics mindset |
Anthropic PM team structure:
| Team | Function |
|---|---|
| Research PM(Research PM) | Led by Diane, customer feedback + model releases |
| Cloud Developer Platform | API |
| Cloud Code team | Core code product |
| Enterprise(Enterprise) | Enterprise adoption |
| Growth(Growth) | Overall product suite growth |
Real disagreement:Company size determines PM structure:
- Anthropic 30-40 people → AI company speed + tight coupling
- Meituan 2000+ → platform company + fine-grained professional PM division of labor
- 1-person independent product → full-stack non-technical PM
3. Cat Wu's perspective: Detailed explanation by Anthropic Cloud Code PM Head
3.1 Evolution of the PM role (before AI vs after)
| Dimension | Before AI | AI era (2026) |
|---|---|---|
| Time horizon | Long-term planning (6-12 months) + cross-team coordination | Ship features every week / as fast as 1 day in extreme cases |
| Core skills | Roadmap / cross-team / PRD | Get to users fast + Taste + Eval |
| Write PRDs | Must | Occasionally writes |
| Product convergence speed | 6 months → launch | 1 month → launch in 1 day |
3.2 5 Key Capabilities of PMs in the AI Era
- Set clear goals:Who are the core users? What is the most important problem? (Example: reduce permission prompts)
- Build a repeatable shipping process:Cloud Code uses research preview(Research Preview) mode to lower the commitment threshold, shipping new features in 1–2 weeks
- Cross-team framework(Engineering / Marketing / Docs are tightly coupled; marketing and docs can respond immediately when a feature is ready)
- Eval capability(Model evaluation) — witness weekly changes in models
- First-principles thinking(First principles) + cross-boundary gap filling
3.3 Anthropic Product Matrix + Use Cases
| Product | Applicable scenarios | Key features |
|---|---|---|
| Cloud Code Terminal Edition | One-off coding tasks | New features land here first |
| Cloud Code Desktop Edition | Frontend work + task overview + non-technical users | Real-time preview of web apps |
| Cloud Code Web / Mobile | "Kicking off tasks while out" | Start tasks even when not at a computer |
| Cowork(Claude with Hands) | Non-code tasks(Slack 0 / Inbox 0 / writing docs / Slide decks) | Connect Google Calendar / Slack / Gmail / Drive, synthesize information → 20-page deck |
3.4 Cowork Case Study: 20-Page Slide Deck
Input:Connect data sources (Google Calendar + Slack + Gmail + Drive) + prompt + a narrative passage Output:Generate a 20-page draft in a few hours Note:PMs still make the final decision. Give Cowork a standardized deck to follow the design system.
3.5 Two Success Factors for Anthropic
- Unified mission:Bring safe AGI to all humanity —— this is cited in every decision
- Focus:Overall goals > individual product lines / teams sacrifice individual priorities for the whole company / prioritize first-party + API
4. Embiricos' Perspective: OpenAI Codex Lead Explains in Detail
4.1 Current State of Codex (2026-05)
| Metrics | Data |
|---|---|
| Growth after GPT-5 (starting August) | 20× |
| Weekly tokens served | Trillions |
| API ranking | #1 (most served coding model) |
4.2 Codex Vision — Software Engineering Teammate
"It's a bit like this really smart intern that refuses to read Slack, doesn't check data dog unless you ask it to."
Translation: It's a bit like a very smart intern who refuses to check Slack messages and won't look at monitoring unless you tell them to
Now:A smart intern who needs guidance Future:Proactive / collaborative partner across the entire development cycle (ideation → maintenance)
4.3 Real Acceleration Cases (Organizational Level)
| Project | Data |
|---|---|
| Sora Android App | 18 days to finish the internal version,28 days public launch, live on the App Store |
| Atlas Browser | Originally 2–3 weeks / 2–3 engineers → 1 week / 1 engineer |
4.4 AGI Limiting Factor (Most Important Insight from This Podcast)
"The current underappreciated limiting factor is literally human typing speed or human multitasking speed."
Translation: The currently severely underestimated limiting factor is human typing speed, or multitasking speed
Outlook:
- Next year (2026) — early adopters see agents become self-sufficient
- In a few years — large companies follow suit
- Reaching AGI — when productivity growth feeds back to AI labs
4.5 Mixed-Initiative Design Philosophy
"Tesla Autopilot is a good example of mixed-initiative software — empowering humans while providing assistance."
Codex is designed in the same direction — not autonomy max, but humans + AI synergy.
4.6 Practical Advice for Using Codex
| Advice | Details |
|---|---|
| Give it the hardest tasks | Don't start with simple tasks; instead,give it the hardest bugs |
| Build trust | First let it understand the codebase + write a plan → gradually increase tasks |
| Try in parallel | Run multiple tasks at the same time |
| Treat it like a new employee | Build trust gradually |
5. Jenny Wen's Perspective: Anthropic Design Head Explains in Detail
5.1 Two Tiers of New Design Work
| Type | Description |
|---|---|
| Support implementation + execution | No longer making polished mocks (engineers can quickly run them) — directly help engineers polish + guide direction |
| Create short-term vision | 3–6 months ahead — in prototype form, not long-form decks |
5.2 Priority Traits for Designer Hiring
- Resilience(Resilience) + Adaptability
- Build actual things and share(Make real things and share them — have a record of project work)
- Excited about technology + willing to be at the frontier of design —— build real things within 4–6 months
5.3 Advice for New Designers
| Advice | Notes |
|---|---|
| Build actual things and share them | More important than portfolio |
| Don't pursue a perfect portfolio | Learning comes from shipping real products |
| Tech literacy is essential | Tailwind / Figma + code polish |
6. Domestic AI PM Transition (Exclusive Complete Version)
AI Product Manager Interviews and Transitions / Compilation of 12 Resources —— 12 resources + 100 interview questions:
6.1 6 Major Dimensions: AI PM vs Internet PM
| Dimension | Internet PM | AI PM |
|---|---|---|
| Core capabilities | User experience + business design | Algorithm understanding + data closed loop + business integration |
| Uncertainty | Medium (user behavior) | High(model effects are unpredictable) |
| Iteration speed | Weeks / months | Months / quarters (data collection is slow) |
| MVP approach | Quickly run through the process | Balance three factors: effect + speed + cost |
| Collaboration targets | Engineers + designers | + algorithm engineers + data annotators |
| Cost of failure | Low | Extremely high(sunk data / compute costs) |
6.2 Distribution of 100 Interview Questions
| Category | Examples of high-frequency questions |
|---|---|
| AI general knowledge | What is AI? Supervised vs unsupervised? Pre-train vs Fine-tune? |
| NLP | What is NER? How to do RAG? How to choose embeddings? |
| CV | Object detection vs classification? Segmentation vs detection? CNN principles? |
| Speech | Difficulties in ASR? VAD? TTS evaluation metrics? |
| Large models | How to use large models? Prompt engineering? How to design Agents? |
| Product methodology | Differences between AI PMs and Internet PMs? How to write AI product requirement documents? |
| Data | Data annotation process? Data closed loop? Cold start? |
| Business | How to find PMF for AI products? How to do commercialization? |
| Management | AI team structure? Algorithm PM vs application PM? |
| Industry | Landscape of OpenAI / Anthropic / domestic large model companies? |
6.3 Systematic Thinking = Looking at Problems Using Principles of Physics
- Use "first principles"(First Principles)
- Think with "models" (simplify, then expand)
- Train with "counterintuition" (break fixed mindsets)
7. Wang Huiwen's Classic Perspective (Exclusive)
Yitang / Wang Huiwen Tsinghua Product Class 66-Page Complete Notes —— 2020 Tsinghua Product Class 66 pages:
7.1 Failure: 30 Times More Common Than Success
| Failure | Success |
|---|---|
| Nokia (peak in 2000) | iPhone(2007) |
| MPMan (Korea's first MP3 player) | iPod |
| Kik / Miliao | |
| Smart ABC | Sogou Input Method |
| IE | Chrome |
Core rule: In almost every major industry, the ultimate winner is usually not the first company.
7.2 5 Great PMs
| PM | Key products |
|---|---|
| Neil McElroy(Procter & Gamble) | The first product manager in human history |
| Steve Jobs | PC + graphical interface + iPhone |
| Zhang Xiaolong | Foxmail + QQ Mail + WeChat (domestic legend, engineer background) |
| Pichai | Chrome (metallurgical engineering major) |
| Mark Zuckerberg / Shawn Parker / Mark Pincus |
7.3 Startup PM vs Professional PM
| Type | Focus |
|---|---|
| Startup PM | Overall business success or failure / strategy / business choices / Strategy |
| Professional PM | Specific features / 4 major documents (product requirement document/prototype/PRD) |
"Product manager is the role closest to the CEO,but only if you have the mindset and thinking of a startup product manager. "
7.4 5 Must-Read Books
| Book | Value |
|---|---|
| The Paypal Wars | Elon Musk is portrayed negatively in the book |
| The Tipping Point | Vivid examples, explaining product characteristics + why iterations are accepted by consumers |
| The Lean Startup | Shorter iteration cycles than others = huge competitive advantage |
| The Innovator's Dilemma | Companies disrupted in the hard drive industry failed not because of technology, but because of organizational decisions |
| The Philosophy of Retail(Suzuki Toshifumi) | Readable from beginner to expert / covers 7-Eleven's decision-making process |
7.5 The Real Two Core Competencies (Wang Xing answered "courage")
- Ability to discover opportunities
- Ability to keep learning and improving
8. Lao Huang: Solo PM for Independent Products (Exclusive)
newtype · Vibe Coding Independent Product Practice §1 + newtype Mindset · Super Individual Methodology:
8.1 3 Mindsets on Unit Economics
| Approach | Content |
|---|---|
| Weigh your capabilities and resources | 1 person + non-technical + spend a little money → don't touch large/complex projects |
| Start from a business angle | Can you turn technical problems into business problems? |
| Start from your strengths | If marketing/operations/product are your strengths, use them as the framework to structure the whole project |
8.2 4 Things Lao Huang Did for the Full-Stack Web Version of Prompt House
(The rest of the time Cursor was writing code while Lao Huang was watching short videos):
- Register accounts on various platforms
- Configure Google OAuth
- Configure Supabase database
- Configure Stripe (API Key, Webhook Key)
8.3 6 Mindsets from Lao Huang
| Mindset | Content |
|---|---|
| Three Unfair Advantage capabilities | "What do I have that others cannot copy?" |
| Just Do It | Don't wait until you finish learning to start |
| Digging for treasure vs finding faults | Two basic modes of looking at things |
| Go Fundamental | Return to the foundational layer (Keynote replaces v0.dev for design) |
| How vs What | Focus on How, not What |
| 5 core values | Independent thinking / long-termism / first principles / continuous learning / creating value |
9. How to Choose — Collaboration Paths by Identity
| Who you are | Primary method | Starting point |
|---|---|---|
| AI company PM (overseas) | Cat Wu Anthropic path | Research preview + weekly ship + Taste + Eval + first principles |
| AI company engineer | Embiricos OpenAI Codex path | Software engineering teammate + Mixed-Initiative + give hardest tasks to build trust |
| AI company Designer | Jenny Wen Anthropic path | 20–30% code polish + 3 types of prototypes / Build & Share |
| Domestic AI PM transition | 12 resources for domestic AI PM transition | 100 interview questions + 6 major dimension differences + required systematic thinking course |
| Solo PM for independent products | Lao Huang non-technical path | Unit Economics 3 Mindsets + 6 Mindsets + 4 Web Things + Mac App Validation Model |
| Classic Product Methodology | Wang Huiwen's Tsinghua Product Course | 5 Must-Read Books + 5 PM Gurus + Startup PM vs. Career PM |
General Discipline:
- The cheaper writing code becomes, the more valuable "deciding what to build" is(Cat Wu)
- AI cannot resolve interpersonal disagreements about "what a feature should do"(Jenny Wen)
- Don't be an "Agent Demo Addict" —— The 1000x difficulty gap between a Demo and a real product
- Build actual things and share them —— More important than a portfolio (Jenny Wen)
- Cross boundaries to fill gaps(Cat Wu first principles)
- Failure is 30 times more common than success(Wang Huiwen)
- Product taste > Engineering capability(Cat Wu + Wang Huiwen + Steve Jobs)
- Build Mixed-Initiative systems, don't max out autonomy(Embiricos)
10. Reference List
Main original author wikis:
- Lenny's Newsletter / 2026 AI-Era Organization Operations Roundtable (Anthropic + OpenAI) —— Cat Wu Anthropic PM Head / Embiricos OpenAI Codex / Jenny Wen Anthropic Design / Blurring role boundaries / Cowork
- AI Product Manager Interviews and Career Transitions / Compilation of 12 Resources —— 12 PDFs + 100 interview questions + 6 key dimensions of AI PM vs. Internet PM + Required course on systematic thinking
- newtype · Vibe Coding Independent Product Practice —— Lao Huang's independent products / Unit Economics 3 mindsets / 4 Web things
- newtype Mindsets · Super Individual Methodology —— Lao Huang's 15 mindsets / Unfair Advantage / 6 mindsets
- newtype Knowledge System and Full Evolution of Life OS —— Three layers of knowledge systems / Personal AI OS
- newtype Essay Collection (Industry Analysis + Philosophy + Startup Stories) —— Dan Koe creator story ($3K→$4.2M) / Weekly newsletter omnibus
- Yitang / Wang Huiwen Tsinghua Product Course 66-Page Complete Notes —— Wang Huiwen Tsinghua Product Course / Failure 30 times more common / 5 PM gurus / 5 must-read books / Wang Xing quotes
Related topic wikis (this series):
- Claude Code Workflow (cross-author topic synthesis) —— Claude Code Workflow (Batch 1) / Cowork
- Startup Methodology (0 to 1) (cross-author topic synthesis) —— Startup Methodology (Batch 1) / Wang Huiwen school
- AI Agent Architecture and Multi-Agent Orchestration (cross-author topic synthesis) —— Agent Architecture (Batch 2)
- Scaling Law / AGI Timeline / Software 3.0 (cross-author topic synthesis) —— Scaling Law / Karpathy Iron Man Suit (this batch Batch 4)
History
- 2026-05-18:Topic-dimension wiki first release. Synthesized and rewritten from 7 author wikis. Explicitly notes 3 points of disagreement: Must an AI PM understand algorithms? / Should designers write code? / PM team size: 30 vs. 2000 vs. 1
Sources and Related Materials
- Lenny's Newsletter / 2026 AI-Era Organization Operations Roundtable (Anthropic + OpenAI)
- AI Product Manager Interviews and Career Transitions / Compilation of 12 Resources
- newtype · Vibe Coding Independent Product Practice
- newtype Mindsets · Super Individual Methodology
- newtype Knowledge System and Full Evolution of Life OS
- newtype Essay Collection (Industry Analysis + Philosophy + Startup Stories)
- Yitang / Wang Huiwen Tsinghua Product Course 66-Page Complete Notes
- Claude Code Workflow (cross-author topic synthesis)
- Startup Methodology (0 to 1) (cross-author topic synthesis)
- AI Agent Architecture and Multi-Agent Orchestration (cross-author topic synthesis)