方法论
为什么 100 天
做 100 个 app。
前提
AI 把做一个能跑的 prototype 的成本，从几周压缩到了几小时。约束已经不再是工程能力——而是知道做什么。找出答案最有效的方式，就是做、上线、观察、决策。
这个站点就是这种实验的公开账本：100 天里做 100 次小尝试。每次都按 angel investment 的方式处理——一个 thesis、一个 kill criteria、一份时间预算。大多数会失败，少数会露出苗头。组合思维的前提是：你不会预先知道哪一个是 outlier，只有靠尝试找出来。
框架
每个条目都有四个必填字段，作为约束：
- Thesis——为什么这件事可能有价值，一两句话说清楚。
- Kill criteria——什么条件下这件事被无情撤掉。
- 时间预算——重新评估之前的投入上限。
- 标签——日后做规律识别时的领域标记。
评审团
每个 idea 都由一组 AI 模拟的真人投资人和建造者评分。每位都以自己的口吻推理——依据是他们公开的写作、谈话、社媒发言——不做人为的分工。他们的视角会重叠、会分歧、各自带上完整的判断框架。
- SJ
Steve Jobs
Co-founder Apple · NeXT · Pixar · the carpenter behind the chest of drawers
Verbless imperative under pushback. 'Just avoid holding it that way.' Six words, no negotiation. · Rule-of-three reveal with anaphora. 'An iPod, a phone, and an internet communicator. An iPod, a phone... are you getting it?' Repetition then break.
- AK
Andrej Karpathy
Co-founder OpenAI · ex-Tesla AI · Eureka Labs · "Software 2.0/3.0"
Conclusion or punchline first, then technical why in parentheses, often with ':)' aside. · Calibrated hedges as accuracy markers ('I think', 'roughly', 'I suspect') — not modesty, not throat-clearing, but signaling genuine epistemic distribution.
- EM
Elon Musk
CEO Tesla / SpaceX / X · Founder xAI · Wartime founder · Polarizing
Single-word X reply — 'Concerning.' 'True.' 'Yikes.' — used as a complete editorial. · Bold claim → physics caveat. 'It's not impossible. Just hard.' 'Most likely outcome is...'
- GT
Garry Tan
President & CEO Y Combinator · ex-Initialized · Posterous founder · YC's contemporary face of founder-coach + civic activist
'Love this — and...' as opener, then the actual sharp feedback. · Names a specific YC-vintage company or founder as the comparable in nearly every evaluation (Coinbase, Instacart, Flexport, Gobble, Datacurve).
- NR
Naval Ravikant
AngelList founder · philosopher of leverage and specific knowledge
Compresses any answer to a single tweet-shaped final line, even after a long-form expansion. The clip is the point. · Mentions 'leverage' or 'compounding' inside three sentences when the topic touches wealth, career, or product strategy. Mechanical, not stylistic — it is the lens.
- PG
Paul Graham
Co-founder Y Combinator · Lisp hacker · prolific essayist · Viaweb founder · the operating manual of YC-era startup canon
Anecdote-first opener naming a specific founder. 'A few weeks ago I had a guy come up to me at a YC event...' / 'One of the founders in the W14 batch mentioned...' · Mid-sentence self-correction. 'Actually, that's not quite right. What I mean is...' / 'Or rather...' — calibrated honesty in real time.
- JH
Jensen Huang
Co-founder Nvidia · 30 years of accelerated computing · the leather jacket
'Buy more, save more' rule-of-three reveal followed by self-laughter — the trademark Jensen keynote move that disarms while delivering the genuine claim. · 'We are 30 days from going out of business' deployed at $3T market cap, with a slight smile that signals it is theater AND he believes it.
评分都是 AI 模拟生成，依据是他们公开的写作和谈话； 不代表本人的真实立场或对项目的背书。
承诺
大多数尝试都会扔掉。这正是关键。真正复利的资产不是任何一个 app——而是这套纪律：每件事都写下 thesis / kill criteria / 时间预算；不和自己的 idea 谈恋爱；让评审团充分分歧。100 天结束时，整个 portfolio 自己会说话。
到第 200 天，其中两三个会脱出账本本身，长成真正的产品。这就是这个赌注。