When to open:You need to judge "how the overall AI track will play out / who will win between OpenAI vs Anthropic vs Google / what to buy now / how to view the SaaS crisis / can Apple still catch up". This wiki integrates the perspectives of 7 author wikis on the theme of "industry judgments", and explicitly marks 4 core disagreements.
One-sentence core:All authors agree that LLM = OS / models = applications / general and vertical models coexist, but there are clear disagreements on the three camps of Scaling Law, the SaaS crisis, and Apple's critical vulnerabilities. Each author's judgments have been repeatedly verified by time, and are worth reading in comparison.
0. The "AI industry" from 6 perspectives
| Perspective | Source | Core pattern | One-sentence stance |
|---|---|---|---|
| 3-year in-depth observation camp | Wang Kai's Observations on AI Industry Judgments and Large Company Competitive Landscape(Wang Kai, 790 posts) | Judgments + verification table + missed opportunities table | "Cognition that is not monetized equals no cognition / view the present from the perspective of AGI" |
| Methodology + immutable anchor camp | newtype's Observations on AI Business, Industry and Investment(Huang Yihe, 452 posts + BIG NEWS) | 7 immutable anchors + LLM=OS + Context Matters | "Don't build a house on someone else's foundation / models are eating applications, position first" |
| Market data camp | AI Agent Market Investment Trends 2025 (Data Report)(2026-01 research report) | $87 billion / 191 unicorns / king-making strategy data | "Sequoia Agent Economy three stages / 25× revenue multiple / bubble warning" |
| Model evolution camp | newtype Model Evaluation, Fine-tuning and Hardware Evolution(505 posts, 2024-2026) | Complete 2024-2026 timeline + real hardware testing | "Models are no longer scarce; what is scarce is the ability to engineer and integrate them" |
| Anthropic+OpenAI insider camp | Lenny's Newsletter / 2026 Three-Person Talk on Organizational Operations in the AI Era (Anthropic + OpenAI) | Cat Wu(PM)+ Embiricos(Codex)+ Jenny Wen(Design) | "AGI bottleneck = human typing speed / design processes are dead / remove the crutch" |
| Shensi Circle VC camp | Shensi Circle Complete Author Collection Portal(659 articles) | Three key terms: US dollars + financing + a16z | "Provides in-depth interpretation 1-2 weeks earlier than other domestic self-media" |
1. Cross-author consensus (7 points)
1.1 LLM = OS, not a chat box
Lao Huang newtype's Observations on AI Business, Industry and Investment §1.3 (2024-05 Post 227 / Classic):
| OS component | LLM equivalent |
|---|---|
| Memory management | Context window |
| File system | Conversation history + knowledge base |
| Drivers | Function Call |
| User interface | Natural language interaction (replacing CLI / GUI) |
"If you string together OpenAI's updates along this path,it truly follows the logic of an OS. "
Lao Huang further advanced this in 2025-03:"Models are evolving into complete applications, eating up various vertical tracks and traditional positions" — an extension of LLM = OS.
Wang Kai corroborates (Wang Kai's Observations on AI Industry Judgments and Large Company Competitive Landscape §10): In the internet era Page Rank → in the AI era Agent Rank. Intermediary-style monopolies end.
1.2 The Big Three landscape of OpenAI / Anthropic / Google is clear
| Giant | Wang Kai's judgment | Lao Huang's judgment | Valuation (2025) |
|---|---|---|---|
| OpenAI | Replicates Google's ecosystem route + maintains model lead → strongest independent large model in the long term | 2024-10 "chose the wrong core users" (code not as good as Claude / Canvas not as good as Artifacts / search not as good as Perplexity / GPTs are dispensable) | $500 billion → $830 billion |
| Anthropic | Pure AI route + best aligned with the token economy, but needs to bind to ecosystems such as Google to counter OpenAI's platformization | Five major stacks: Claude 4 / Sub-agents / Skills / Cowork / Code, main force shifting to OpenCode in 2026 | $183 billion → $350 billion |
| Product orientation > technical orientation / No. 1 in the world in AI implementation / ecosystem advantages emerging | 2024-05 "Google's launch event was more useful than OpenAI's" / Gemini 2.5 Pro is god-tier / Stitch + Veo + Genie | Full-stack ecosystem |
Consensus:
- Beyond the model layer,both OpenAI / Anthropic are becoming ecosystem players(GPTs / Skills / MCP)
- Google's "full-suite context" is a long-term advantage
- Startups can no longer build "super large models" (Kai-Fu Lee 2025-01 + Lao Huang consensus)
1.3 "Don't build a house on someone else's foundation"
Lao Huang newtype's Observations on AI Business, Industry and Investment §1.1 (2024-03 Post 4 / Zhu Xiaohu):
"I always tell domestic entrepreneurs,never build a house on someone else's foundation. "
Domestic large model companies want to do both models and applications, so entrepreneurs are even more afraid to use them.
Wang Kai corroborates ("Four Great Mountains for AI Entrepreneurs"Wang Kai's Observations on AI Industry Judgments and Large Company Competitive Landscape §2):
- Don't build things on the short Siri path
- Don't build things that phone manufacturers' systems will definitely include
- Find niche points that "hardware + systems + platforms + large companies" are all unwilling to touch
Lao Huang upgraded this in 2025-05:"Don't compete with giants over the 'input box'" — ChatWise-style Chatbots are doomed.
1.4 LLMs are consumables; data / context / tool integration are the moats
Lao Huang newtype's Observations on AI Business, Industry and Investment §1.7 (2025-08 Post 1516):
"Claude models are hard currency right now.Whoever provides more model quota will have a hit product(Kiro). Whoever can't bear the cost and starts to scale back will go downhill (Cursor)."
Lao Huang newtype's Observations on AI Business, Industry and Investment §1.5 Context Matters:
"Pro Users definitely use multiple tools, so Prompt management is a real need". Context is the real moat.
Wang Kai corroborates (Wang Kai's Observations on AI Industry Judgments and Large Company Competitive Landscape §3.4) "Context engineering perspective":
- Google holds the most high-quality context (Search / Gmail / Drive / YT)
- Alibaba / ByteDance / WeChat are also massive context holders
- In the future, users' GPT/Claude will look identical on the surface, but with different connected data/tools, efficiency will differ vastly
1.5 5 levels of AI (OpenAI 2024-07 / adopted industry-wide)
graph TD
L1[L1 Chatbot<br/>对话能力<br/>2022-2023] --> L2[L2 Reasoner<br/>专家级推理<br/>2024 o1/r1]
L2 --> L3[L3 Agent<br/>多步任务自主<br/>2025 Devin/Claude Code]
L3 --> L4[L4 Innovator<br/>原创性突破<br/>?]
L4 --> L5[L5 Organization<br/>完整组织<br/>?]
Wang Kai's current positioning:L2 → L3 transition period — "Consumer-side Agents are still one 90% drop in token prices away from breakout".
Embiricos(Lenny's Newsletter / 2026 Three-Person Talk on Organizational Operations in the AI Era (Anthropic + OpenAI)) adds:"AGI bottleneck = human typing speed" — not GPU compute / model intelligence, but the speed at which humans input prompts + manually verify AI work.
1.6 General + vertical coexistence (2025 industry consensus)
| Camp | Stance |
|---|---|
| Cohere CEO Aidan(2024-10) | Build a general prototype first → do vertical fine-tuning after it works |
| Kai-Fu Lee(2025-01) | AI applications need to be "small enough, fast enough, cheap enough, good enough" — not necessarily the largest |
| Lao Huang(newtype's Observations on AI Business, Industry and Investment §1.4) | "General at the bottom + vertical at the top, use whichever is good or cheap; truly optimized small vertical models become moats" |
| Wang Kai(firsthand experience) | Multi-model hybrid / Claude for decision-making + Gemini for large context + Sonnet/Flash for execution |
1.7 Models are no longer scarce; engineered integration is scarce
newtype Model Evaluation, Fine-tuning and Hardware Evolution one sentence:
"Models are no longer scarce; what is scarce is 'the ability to engineer and integrate them'. "
2024→2026 evolution corroborates:
- Single RTX 3060 runs 7B (2024)→ Mac mini M4 cluster runs Qwen 32B at 18 tok/s (2024-11)→ M3 Ultra dual 512G runs full DeepSeek at 20 tok/s (2025-02)→ OpenCode black card $200 to use any model(2026-01)
Models have shifted from "scarce" to "infrastructure",true differentiation lies in engineered integration layers such as Prompt House / Super Analyst Skill / Multi-Agent Orchestrator.
2. Explicit disagreements (4 points — not smoothed over)
2.1 Disagreement 1: The three-way Scaling Law debate
| Camp | Claim | Representatives |
|---|---|---|
| Brute-force aesthetics camp | More data + more compute = higher intelligence = AGI | OpenAI / Anthropic / Google / Wang Kai leans this way |
| Application-layer camp | Large model upgrades will not eat up opportunities; application scenarios are the core | Zhu Xiaohu (2024-03) / domestic VCs |
| Ceiling camp | Scaling Law is already showing diminishing marginal returns | Kai-Fu Lee (2025-01) / Cohere CEO Aidan (2024-10) |
Full original quote comparison: Zhu Xiaohu vs Kai-Fu Lee vs Cohere
| Camp | Key quote | Time |
|---|---|---|
| Zhu Xiaohu(Application-layer camp) | "Right now OpenAI's technology iteration curve is still relatively steep, and open source is definitely one year or even a year and a half behind closed source. But once the closed-source technology iteration curve slows down, open source will catch up" | 2024-03 |
| Zhu Xiaohu(ibid.) | "Before controllable nuclear fusion is achieved,I don't believe the Earth can achieve true AGI" | 2024-03 |
| Cohere Aidan(Pragmatist camp) | "Scaling Law still holds and will remain valid for a considerable period. But the reality is,most companies' businesses still cannot support GPT-4-level deployment costs" | 2024-10 |
| Kai-Fu Lee(Evolutionist camp) | "Scaling Law is slowing down. Only big companies can continue to build super large models; startups have an extremely low probability of success. The models AI applications need are small enough, fast enough, cheap enough, good enough" | 2025-01 |
| Kai-Fu Lee(ibid.) | "2025 accelerates AI commercialization and also accelerates elimination.When opportunities come, make decisions bravely; when opportunities disappear, do the same"" | 2025-01 |
Lao Huang's comprehensive judgment (2025-03-27 Post 1040)
"Many people in China believe that even if models are not good enough, you can still compete on applications. But the current situation is,models are evolving into complete applications, eating up various vertical tracks and traditional positions. "
→ Lao Huang stands between the "brute-force aesthetics camp + anti-application-layer camp".
Wang Kai's stance (Wang Kai's Observations on AI Industry Judgments and Large Company Competitive Landscape §5)
Leans toward the brute-force aesthetics camp (firmly believes AGI will arrive before 2035), but acknowledges the application layer still has space when it can solve "details that AI is not good at".
Real disagreements are not smoothed over:
- If you do models/compute/infrastructure → bet with the brute-force aesthetics camp
- If you build applications/vertical/SaaS → work with Cohere/Kai-Fu Lee on "small enough and cheap enough"
- If you don't build AI and build products directly → bet with Zhu Xiaohu on "applications reign supreme"
2.2 Disagreement 2: How to view the SaaS crisis?
| Perspective | Position |
|---|---|
| February 2026 market panic logic(short-term crash) | AI replaces human labor → layoffs → SaaS charges per seat → revenue declines → valuation collapses (CRM/NOW/ADBE plunge) |
| Wang Kai's rebuttal(Wang Kai's Observations on AI Industry Judgments and Large Company Competitive Landscape §8) | "No problem with certainty within 3 years" / The real risk is not the present, but AI changing reading and writing(everyone wears AI glasses and no longer opens CRM) / The present is a "good asset, good volatility" opportunity |
| Lao Huang(newtype Observations on AI Business, Industry and Investment February 2026 Post 2373) | "The end of intermediary monopolies" / Page Rank → Agent Rank / The end of SaaS intermediary monopolies |
| Consensus | SaaS giants have proprietary data + channels + services, and still hold an advantage in the AI era; but intermediary-type players (that only do "information distribution" without proprietary data) will definitely die |
Real disagreement:How to define "SaaS":
- Lao Huang defines "SaaS = intermediary monopoly" → will definitely die
- Wang Kai defines "SaaS = data + workflow" → still holds an advantage
Key to judgment:Whether this SaaS has "proprietary data / proprietary workflow / proprietary customer relationships"?
2.3 Disagreement 3: Apple's path — ultimate winner or slow-executing loser?
| Perspective | Position |
|---|---|
| Wang Kai(2024-02) | "Apple is bound to be the biggest winner in the AI era" — integrated software-hardware ecosystem + brand equity |
| Wang Kai(2024-06) | "Siri is a certain ticket to the AI era, eating all shallow-interaction apps" |
| Wang Kai(actual verification in 2026) | ⚠️ Apple executes slowly — has already exited Apple positions |
| Lao Huang(newtype Observations on AI Business, Industry and Investment §4.6) | "GUI Agent is a platform's job; for example, Apple and Google will definitely do it. For developers and startups, there is no advantage in going this direction" |
| Lao Huang(newtype Observations on AI Business, Industry and Investment §4.7-4.8) | "On-device AI is significantly behind the broader industry / mobile permissions + computing power, two key points, are held in the hands of phone manufacturers" |
Real disagreement:Whether Apple has already missed the first year of AI:
- Wang Kai's position:Still a winner in the long run(brand + ecosystem + software-hardware integration), but cannot be bet on in the short term
- Lao Huang's position:On-device AI is doomed, investing in phone manufacturers' AI is meaningless
- Consensus: Do not enter the GUI Agent track (it is platform work)
2.4 Disagreement 4: The controversy over large model companies "choosing the wrong core users"
| Perspective | Position |
|---|---|
| Lao Huang(October 2024 Post 562 / classic diss) | OpenAI chose the wrong core users — code is not as good as Claude / Canvas is not as good as Artifacts / search is not as good as Perplexity / GPTs are dispensable |
| Wang Kai(Wang Kai's Observations on AI Industry Judgments and Large Company Competitive Landscape §3.1) | OpenAI is copying Google's ecosystem route —— GPTs / Store / Custom GPTs / Apps in ChatGPT |
| Embiricos(Lenny's Newsletter / 2026 Three-Person Talk on Organizational Operations in the AI Era (Anthropic + OpenAI)) | Codex is 20× growth after GPT-5 / serving trillions of tokens per week trillions of tokens / No. 1 in API — OpenAI Coding counterattacks Claude |
| Market data(AI Agent Market Investment Trends 2025 (data report)) | OpenAI $500 billion → $830 billion / Cursor in November $29.3 billion(3x increase in 5 months) / Anthropic $183 billion → $350 billion |
Real disagreement:Whether OpenAI is a ChatBot company or an OS platform company:
- Lao Huang's position: OpenAI wants to be a platform but has not found its core users
- Wang Kai's position: OpenAI is copying Google's ecosystem, with leading models + building the platform slowly
- Embiricos' counterattack: Codex is already No. 1 in API,$500 million → $8.3 billion valuation multiplied 16 times
Judgment: 2026 situation = OpenAI counterattacks in Coding / Anthropic holds Cloud Code + Skills / Google's full suite slowly penetrates.A tripartite standoff, not "OpenAI dominance".
3. Lao Huang's 7 Immutable Anchors (exclusive, most stably verified 2024-2026)
newtype's Observations on AI Business, Industry and Investment §1 — Lao Huang's two years of business judgments:
| # | Immutable anchor | Fulfillment status |
|---|---|---|
| 1 | Don't build a house on someone else's foundation(March 2024 / Zhu Xiaohu) | ✅ Domestic large model companies build both models and applications, leaving entrepreneurs with no way out |
| 2 | Models are eating applications; position first then wait for the wave(2024-05) | ✅ Anthropic Claude Code eats Coding Agent / OpenAI 4o eats Browser Use |
| 3 | LLM = OS(May 2024 Post 227) | ✅ Models already = complete applications, eating vertical tracks |
| 4 | General foundation + vertical upper layer(2024-08) | ✅ Industry consensus / Lao Huang builds Super Analyst Skill |
| 5 | Context Matters(July 2025 Post 1495) | ✅ Prompt House monthly active users verified / Wang Kai's "context engineering perspective" |
| 6 | Pro Users definitely use multiple tools; Prompt management is a real demand(2025-11) | ✅ Prompt House / Skill ecosystem explosion |
| 7 | Data is the moat, models are consumables(2025-08) | ✅ Cursor unsubscriptions / Kiro popular / Claude hard currency |
4. Wang Kai exclusive: 13 time anchors + admitted wrong judgments in Section 1.2
Wang Kai's Observations on AI Industry Judgments and Large Company Competitive Landscape §13 + §1.2:
4.1 13 time anchors (the biggest AI industry event each year)
| Year | Landmark event | Wang Kai's observation |
|---|---|---|
| 2022-11 | ChatGPT launch | "AI unlocks opportunities wave by wave" |
| 2023-03 | GPT-4 + domestic large model free-for-all | "China will definitely produce multiple large models, no leader" ✅ |
| 2023-11 | OpenAI internal strife | "Sam is irreplaceable" ✅ |
| 2024-05 | GPT-4o | "Multimodality, camera input" |
| 2024-09 | o1 reasoning model | "Democratization of slow thinking" |
| 2025-01 | DeepSeek v3/R1 | "Reassessment of China-US AI competition" |
| 2025-03 | Manus | "Beginning of the Agent era, sentiment > technology" ✅ |
| 2025-08 | GPT-5 | "thinking mode as default" |
| 2025-11 | Claude Code popularization | "AI Coding is a basic skill" ✅ |
| 2026-02 | SaaS plunge | "The market has already begun to value using AI's endgame replacement" |
4.2 Three judgments Wang Kai admitted getting wrong (self-reflection)
| Judgment | Reason for error | Lesson |
|---|---|---|
| Firmly believed in AI in December 2022, but did not buy NVDA | Did not convert cognition into investment returns | "Cognition not converted into returns equals no cognition" |
| Search would be immediately impacted after ChatGPT came out | Underestimated the lag effect | "Change is cumulative, not overnight" |
| Smartphones would definitely replace feature phones immediately | Same as above | Nokia still has a $40 billion market cap now |
Why these 3 reflections are valuable:
- "Immediately" is the most common bias in judging AI's impact
- Lag effect: technology → user habits → business model, requires 3-5 years of accumulation
- Do not make "immediate impact" judgments before AI has "spread to ordinary people"
5. Complete model evolution timeline (exclusive to Lao Huang's newtype-model-bench)
newtype Model Evaluation, Fine-tuning and Hardware Evolution — 505 posts from March 2024 to May 2026:
5.1 Key inflection points (only the most important per quarter listed)
| Time | Key model | Inflection point significance |
|---|---|---|
| 2024-04 | Cohere Command R+ | First time "knowledge base first choice" |
| 2024-05 | GPT-4o + Claude 3.5 Sonnet | Dual inflection point of multimodality + Coding |
| 2024-07 | OpenAI AGI 5 levels | Industry standard |
| 2024-09 | OpenAI o1 | Slow thinking era |
| 2024-11 | Mac mini M4 cluster runs Qwen 32B at 18 tok/s | Local large model turning point |
| 2025-01 | DeepSeek R1 APP goes live | National-level / China-US AI reassessment |
| 2025-02 | Claude 3.7 Sonnet | No. 1 in AI Coding |
| 2025-03 | Gemini 2.5 Pro | QR code scanning artifact |
| 2025-04 | Cursor valuation $9.9 billion → $29.3 billion in November (3x in 5 months) | Typical "Kingmaking Strategy" |
| 2025-05 | Claude 4 Sonnet / Opus | Proactively uses Sequential Thinking MCP |
| 2025-08 | GPT-5 + Claude Code Sub-agents | thinking default + multi-Agent |
| 2025-10 | Anthropic Claude Skills | Birth of Super Analyst / Super Writer |
| 2025-11 | Kimi K2 Thinking + Gemini 3 Pro + Claude Opus 4.5 | Domestic practicality inflection point + tripartite coexistence |
| 2025-12 | GPT-5.2(Instant/Thinking/Pro three tiers) | "Reached human expert level for the first time" |
| 2026-01 | OpenCode black card $200 sold out in seconds + Anthropic Cowork + NVDA $20 billion acquisition of Groq | Multi-Agent commercialization inflection point |
| 2026-04 | Anthropic requires identity verification + GitHub suspends Copilot registrations | Signal that computing power cannot hold up |
| 2026-05 | Claude Opus 4.7 + Perplexity Personal Computer ($200-tier investment tool) | Model application-ification |
5.2 Lao Huang's current (May 2026) model portfolio
开发主战场:Claude Code(Skills + MCP + Slash + Sub-agents)
代码评审:Claude Code + Ultra think
代码搜索:Gemini 2.5 Pro(大上下文扫码)
API / 模型组合:Claude(规划)+ Gemini(大上下文)+ Sonnet/Flash(执行)
本地服务器:Mac mini M4(OLLAMA_KEEP_ALIVE=-1,搭配 Enchanted iOS)
6. AI Agent market data (exclusive to newtype research report)
AI Agent Market Investment Trends 2025 (data report):
6.1 Overall scale 2025
| Indicator | Data |
|---|---|
| Global GenAI VC investment | $87 billion |
| AI Agent dedicated financing | $6.4 billion / 451 deals(as of October) |
| Year-over-year growth vs 2024 | 65% |
| Mega-round share | Top 4 account for total 30%+ |
| Share of $500 million+ rounds | 58% |
6.2 Valuation evolution of the three giants
| Company | 2025 valuation |
|---|---|
| OpenAI | $5,000→$830 billion |
| Anthropic | $1,830→$350 billion |
| xAI | Cumulative $42.1 billion |
6.3 "Kingmaking Strategy"
The most anomalous phenomenon in 2025: startups complete two consecutive rounds of financing within 27-60 days, with valuations doubling or even tripling without new data support.
Typical case Cursor:
| Time | Valuation | Remarks |
|---|---|---|
| June | $9.9 billion | $900 million financing |
| November | $29.3 billion | $2.3 billion Series D (3x growth in 5 months) |
| Annualized revenue | $1 billion | Sales revenue grew 100 times |
6.4 Sequoia "Agent Economy" three-stage theory
- Stage 1:AI Assistants(Assistants) — current
- Stage 2:Agent clusters(Agent Swarms)— 2026-2028
- Stage 3:Agent economy(Agent Economy)— 2028+
6.5 a16z bets
- Raised $15 billion new fund in 2025 (accounting for 18% of total US VC fundraising)
- Led investment in Thinking Machines Lab ($2 billion seed round)
- Focus: AI infrastructure + Agent application layer + Voice AI
7. Internal perspectives from Anthropic + OpenAI (Lenny three-person talk)
Lenny's Newsletter / 2026 Three-Person Talk on Organizational Operations in the AI Era (Anthropic + OpenAI) —— A rare "internal PM/Eng/Design perspective" in external materials:
7.1 Cat Wu(Anthropic Cloud Code PM Head)
- Claude Code multiple forms (terminal / desktop / web / mobile / Cowork)
- Ship features weekly / 30-40 person PM team / Research preview mode
- Remove crutches — early to-do lists helped models do major refactoring; remove them when new models no longer need them
- "PMs, engineers, and designers are merging"
7.2 Embiricos (OpenAI Codex lead)
- After GPT-5 20× growth / serving trillions of tokens per week trillions of tokens / No. 1 in API
- Compaction is the key innovation — allows models to work continuously beyond the context window
- Vision:Software engineering teammate
- Design philosophy:Mixed-Initiative(mixed initiative)
- "AGI limiting factor = human typing speed / multitasking management speed, not GPU computing power"
7.3 Jenny Wen(Anthropic Head of Design Claude)
- "The Design Process is Dead" — the "design process" instilled in designers is dead
- Designer time allocation:Mocking 60-70% → 30-40% / Code polish 0 → 20-30%
- 3 types of valuable Designer prototypes: Strong Generalist / Deep Specialist / Craft New Grad
8. Shensi Circle VC Perspective (Exclusive)
Entry:Full entry to Shensi Circle authors —— VC + corporate moves Deepest perspective
Topic wiki(Added 2026-05-18):
- Shensi Circle / Leading AI company moves (OpenAI/Anthropic/Google/Apple/Perplexity) —— Moves of leading companies: OpenAI / Anthropic / Perplexity / Google / Apple / Microsoft
- Shensi Circle / AI industry trends + new categories + predictions —— Five major trend sections: AGI progress + new categories + legacy SaaS replacement + infrastructure + leading predictions
- Shensi Circle / AI primary market VC perspective (a16z + Sequoia + YC) —— Perspectives from three VCs: a16z + Sequoia + YC
8.1 Key judgments (latest as of 2026-05)
| Source | Judgment |
|---|---|
| Sequoia US 2026 annual shareout | "AGI has arrived, redefining the AI revolution" |
| a16z prediction | Computer Use lets AI Agents work like humans,surpassing human efficiency in 18 months |
| a16z prediction | AI upends Moore's Law / reaches tipping point in 2028 |
| Anthropic Jaya Gupta(2026-05) | "The scarcest thing in enterprise AI is shifting from 'intelligence' to 'permissions'" — moat view |
| a16z platform company playbook | Playbooks of Google / Microsoft / AWS / Palantir → accelerating in the AI era (capability as threat + the more used, the better understood) |
8.2 Legacy SaaS replacement wave (milestone events 2024-2026)
| Milestone | Implication |
|---|---|
| Peter Thiel invests $35 million in sales-focused Claude Code | "Ending the Salesforce era" |
| Sequoia US invests $25 million in AI-native ERP | Killing traditional finance software |
| Sequoia invests in legal: don't sell tools, directly AI-rebuild law firms | Paradigm leap |
| Intercom itself starts selling AI models | Legacy customer service giant transforms |
| 2024 = Year one of enterprise AI (16 B2B trends) | Shensi Circle framing |
9. AI hardware wave (Wang Kai vs Lao Huang divergence)
9.1 Wang Kai's position (Wang Kai's observations on AI industry judgments and big-tech competitive landscape §7)
- Phone makers will definitely participate: AI phones are the third revolution of the phone industry
- Apple is the ultimate winner: all-around advantages in hardware + OS + ecosystem + brand, but execution is slow
- AI hardware review/media ecosystem: currently a blank period (analogous to early phone review bloggers)
- Physical AI: the next wave as described by Jensen Huang
Wang Kai investment targets:
- Apple (core) → later rotated out
- ISRG(Intuitive Surgical medical robotics) → sold after 2-month holding surged 20%
- Tesla (FSD + robotics) → ongoing adjustments
- Do not buy OPPO / Xiaomi (insufficient conviction, not monopolies)
9.2 Lao Huang's position (newtype observations on AI business, industry, and investment §4.6-4.8)
- GUI Agent is a platform's job(Apple and Google will definitely do it), no room for startups
- On-device AI significantly lags the broader industry —— mainstream phones run 3B, Qualcomm can reach 7B
- Two key bottlenecks: permissions + compute, held by phone makers
- NVDA acquires Groq for $20 billion (2026-01-24) —— "When giants acquire tech heretic companies, it signals an inflection point of the era"
9.3 Lao Huang's 10-year NVDA holding logic (Post 2076 highlight, 2025-12-02)
- Accelerated computing
- Digital twins
- AI for Science
10. How to use — judgment paths by identity
| Who you are | Primary judgment framework | Starting point |
|---|---|---|
| Investor / focused on US stocks | Wang Kai three-layer framework (tech stack / value chain / competitive landscape) | Only place heavy bets at monopoly/oligopoly layers (Apple / NVDA / Google) |
| AI entrepreneur / independent developer | Lao Huang 7 immovable anchors | Don't build a house on someone else's foundation / models are eating applications, position first / Context Matters |
| Focused on frontier models / tool stacks | newtype model evolution timeline + BIG NEWS | Current stack (Claude primary + Gemini large context + Sonnet/Flash execution) |
| Focused on Agent valuations / VC perspective | newtype-agent-market + Shensi Circle | Sequoia Agent Economy three stages / 25× revenue multiple / "kingmaking strategy" |
| Focused on AI product implementation / internal perspective | Lenny three-person talk | Cat Wu removing crutches / Embiricos typing speed bottleneck / Jenny Wen design process is dead |
| Domestic AI industry / regulation | Wang Kai analysis of domestic AI regulation and localization compliance(not included in this wiki) + Lao Huang LLM=OS | China will produce multiple large models with no clear leader / domestic AI products can no longer charge fees |
General judgment framework(for any identity):
- Pass the 7 immovable anchors(Lao Huang): for every judgment first ask "does it meet these 7 rules" → if not, question it
- Look at the three Scaling Law camps: which camp you bet on determines direction
- Look at the 5 levels of AI: what L is your product at? Will it still exist after L+1?
- Look at context moats(Wang Kai): do you have proprietary data / proprietary workflows / proprietary customer relationships?
- Look at time anchors(Wang Kai's 13): which time anchor does your judgment correspond to? Has the situation changed after that anchor?
- Look at admitting mistakes(Wang Kai 1.2): are you again making an "immediate impact" judgment?
11. Reference list
Main original author wikis:
- Wang Kai's observations on AI industry judgments and big-tech competitive landscape —— Wang Kai 790 posts, 3 years of observations / 16 judgments + verification tables / 3 admissions of error / 13 time anchors / three Scaling Law camps / SaaS crisis
- newtype observations on AI business, industry, and investment —— Lao Huang 452 posts + BIG NEWS timeline / 7 immutable anchors / LLM=OS / Context Matters / startup methodology / individual business experience
- AI Agent Market Investment Trends 2025 (data report) —— $87 billion market / 191 unicorns / kingmaking strategy / Cursor valuation tripled in 5 months / Sequoia Agent Economy / a16z $15 billion
- newtype model evaluation, fine-tuning, and hardware evolution —— 505 posts on 2024-2026 model evolution / Mac mini M4 cluster / M3 Ultra full DeepSeek run / OpenCode black card $200
- AI Agent Technical Architecture Evolution 2025 (in-depth research report) —— Agent technical architecture 2025 / real deployments at Morgan Stanley/Klarna
- Lenny's Newsletter / 2026 Three-Person Talk on Organizational Operations in the AI Era (Anthropic + OpenAI) —— Anthropic + OpenAI internals / Cat Wu PM / Embiricos Codex / Jenny Wen Design / AGI bottlenecks
- Full entry to Shensi Circle authors —— Shensi Circle 659 articles / VC perspective / three key terms: USD + financing + a16z / in-depth interpretation 1-2 weeks early
Related topic wikis (this series):
- AI Agent Architecture and Multi-Agent Orchestration (Cross-Author Topic Synthesis) —— Agent architecture (this batch Batch 2)
- AI Coding Tool Stack Evolution (Cross-Author Topic Synthesis) —— AI Coding tool stack (this batch Batch 2)
- AI Investment and Valuation (Cross-Author Topic Synthesis) —— AI investment valuation (Batch 4)
- Scaling Law / AGI Timeline / Software 3.0 (Cross-Author Topic Synthesis) —— Scaling Law / AGI timeline (Batch 4)
- Startup Methodology (0 to 1) (Cross-Author Topic Synthesis) —— Startup methodology (Batch 1)
History
- 2026-05-18: Topic-dimension wiki first release. Synthesized and rewritten from 7 author wikis. Explicitly marks 4 divergences: three Scaling Law camps / SaaS crisis / Apple path / whether OpenAI "chose the wrong core user"
Sources and related materials
- Wang Kai's observations on AI industry judgments and big-tech competitive landscape
- newtype observations on AI business, industry, and investment
- AI Agent Market Investment Trends 2025 (data report)
- newtype model evaluation, fine-tuning, and hardware evolution
- AI Agent Technical Architecture Evolution 2025 (in-depth research report)
- Lenny's Newsletter / 2026 Three-Person Talk on Organizational Operations in the AI Era (Anthropic + OpenAI)
- Full entry to Shensi Circle authors
- Shensi Circle / Leading AI company moves (OpenAI/Anthropic/Google/Apple/Perplexity)
- Shensi Circle / AI industry trends + new categories + predictions
- AI Agent Architecture and Multi-Agent Orchestration (Cross-Author Topic Synthesis)
- AI Coding Tool Stack Evolution (Cross-Author Topic Synthesis)
- AI Investment and Valuation (Cross-Author Topic Synthesis)
- Scaling Law / AGI Timeline / Software 3.0 (Cross-Author Topic Synthesis)
- Shensi Circle / AI primary market VC perspective (a16z + Sequoia + YC)