Marc Andreessen：AI 是“八十年一夜之间的成功”Marc Andreessen: AI Is an 'Eighty Year Overnight Success'
核心 takeaway：我们正经历 AI 的“八十年一夜成功”——数十年的基础研究终于爆发，推理、Agent 和自我改进三大突破让这一次真正不同以往。
a16z 联合创始人 Marc Andreessen 从 1980 年代就开始研究 AI，他指出神经网络架构在 70 年争议后被证明正确，从 2017 年 Transformer 到 o1 推理模型、OpenClaw Agent 和 RSI（递归自我改进），技术已真正落地。
他强调：“这是八十年来工作的结晶，现在它终于成真。”Pi 和 OpenClaw 将 LLM 与 Unix shell 思维结合，让 Agent 能自我修改、迁移环境、自主扩展功能，完全独立于底层模型。
Andreessen 看好 scaling law 继续生效、供应链瓶颈会缓解，也高度肯定开源（包括 DeepSeek 等中国实验室）的贡献。但他提醒，真实世界有 80 亿人、法规和机构，落地会非常复杂，GDP 影响可能滞后。
这段洞见特别适合正在构建产品的你：既要抓住真实技术进步，也要为实际部署中的混乱做好准备。
a16z 联合创始人 Marc Andreessen 从 1980 年代就开始研究 AI，他指出神经网络架构在 70 年争议后被证明正确，从 2017 年 Transformer 到 o1 推理模型、OpenClaw Agent 和 RSI（递归自我改进），技术已真正落地。
他强调：“这是八十年来工作的结晶，现在它终于成真。”Pi 和 OpenClaw 将 LLM 与 Unix shell 思维结合，让 Agent 能自我修改、迁移环境、自主扩展功能，完全独立于底层模型。
Andreessen 看好 scaling law 继续生效、供应链瓶颈会缓解，也高度肯定开源（包括 DeepSeek 等中国实验室）的贡献。但他提醒，真实世界有 80 亿人、法规和机构，落地会非常复杂，GDP 影响可能滞后。
这段洞见特别适合正在构建产品的你：既要抓住真实技术进步，也要为实际部署中的混乱做好准备。
The Takeaway: We are living through an "eighty-year overnight success" in AI—decades of foundational research have finally exploded into breakthroughs in reasoning, agents, and self-improvement, making this time truly different.
a16z co-founder Marc Andreessen, who has worked on AI since the 1980s, explains that the neural-network architecture was proven correct after 70 years of debate. From the 2017 Transformer to o1 reasoning models, OpenClaw agents, and RSI (recursive self-improvement), the technology is now delivering real-world results.
"This is the culmination of eighty years worth of work," he says. "This is the time it's becoming real." Pi and OpenClaw represent a profound architectural leap: marrying LLMs with a Unix-shell mindset so agents can self-modify, migrate environments, and extend their own capabilities independently of the underlying model.
Andreessen is bullish that scaling laws will continue, supply constraints will ease, and open source (including Chinese labs like DeepSeek) will accelerate progress. Yet he cautions that the real world of 8 billion people, regulations, and institutions makes adoption messy and GDP impact may lag.
Essential reading for builders: seize the genuine technical progress while preparing for the practical chaos of deployment.
查看原文 →
a16z co-founder Marc Andreessen, who has worked on AI since the 1980s, explains that the neural-network architecture was proven correct after 70 years of debate. From the 2017 Transformer to o1 reasoning models, OpenClaw agents, and RSI (recursive self-improvement), the technology is now delivering real-world results.
"This is the culmination of eighty years worth of work," he says. "This is the time it's becoming real." Pi and OpenClaw represent a profound architectural leap: marrying LLMs with a Unix-shell mindset so agents can self-modify, migrate environments, and extend their own capabilities independently of the underlying model.
Andreessen is bullish that scaling laws will continue, supply constraints will ease, and open source (including Chinese labs like DeepSeek) will accelerate progress. Yet he cautions that the real world of 8 billion people, regulations, and institutions makes adoption messy and GDP impact may lag.
Essential reading for builders: seize the genuine technical progress while preparing for the practical chaos of deployment.