吴恩达 (Andrew Ng)，人工智能革命中的杰出人物，做客 “No Pires” 播客，讨论了人工智能的现状和未来发展方向，特别关注了 “具身人工智能 (agentic AI)” 的概念。吴恩达强调，人工智能能力的进步将来自多个维度，包括具身工作流、多模态模型和新技术，而不仅仅是通过大型公司通过大量公关宣传所强调的模型规模化。 吴恩达提出了 “具身人工智能 (agentic AI)” 这个术语，以避免关于什么是“代理”的争论，而是专注于构建具有不同程度自主性的系统。虽然围绕具身人工智能的营销炒作迅速加速，但实际的业务进展也在增长，尽管没有营销宣传所暗示的那么快。实施真正的 AI 代理的最大障碍不是技术，而是人才。他强调在构建有效的代理时，系统性的错误分析和评估至关重要，而这往往是经验不足的团队所缺乏的技能。 吴恩达认为，通过具身工作流可以自动化大量工作，但他强调，构建此类工作流需要通常锁在人们头脑中的外部知识。目前，人类工程师和产品经理在做出上下文相关的决策方面至关重要，而这些决策是 AI 代理目前还无法复制的，特别是在处理专有或非通用知识时。 在吴恩达看来，代理的最强例子是 AI 编码代理。他强调编码代理是产生经济价值的两个明确类别之一，另一个是回答人们的问题。他认为这些代理在规划和执行多步骤流程以构建软件方面具有高度自主性。他相信编码的经济价值推动了大量资源投入到构建有效的编码代理中。吴恩达更喜欢“AI辅助编码”而非“感觉编码 (vibe coding)” 这个说法，因为后者暗示任务比实际情况更简单。使用 AI 辅助使编码成为一项深刻的智力活动。 吴恩达指出，AI 辅助编码正在通过加速编码速度和降低成本来改变创业公司的性质。这会将瓶颈转移到产品管理，要求产品经理依赖直觉和客户共情。他对尝试自动化产品管理某些方面的工具，例如使用 AI 代理进行市场调研，持谨慎乐观态度。但他仍然认为，在取代人类产品经理方面，它们不如编码工具有效。 吴恩达认为，对 AI 技术有深刻理解的创始人更有可能在当前快速发展的环境中取得成功。创始人应该掌握新兴的 AI 技术，否则他们很难领导公司。以技术为导向的产品领导者比以业务为导向的领导者更有可能成功。他强调，对于那些想要改变世界的人来说，努力工作和挑战现状至关重要。 他观察到，创始人主要有两种类型：一种痴迷于他们的企业获胜，另一种痴迷于他们的客户获胜。吴恩达更看重以客户为中心。他强调需要根据对客户的深刻理解做出果断且快速的决策。 吴恩达还谈到了由于人工智能的快速发展，他的观点和工作流程是如何变化的。在招聘工程师时，他非常重视 AI 技能。软件工程师是其他学科的先驱，因为软件工程领域的工具更加先进。此外，他认为未来的工作性质可能涉及更小、高度熟练的团队，在 AI 的辅助下，其表现优于更大、成本更低的团队。 吴恩达相信，在未来五年内，由于 AI 的集成，人们将更加有力量和能力。拥抱 AI 的个人将比大多数人意识到的更加高效和有能力。

Andrew Ng, a prominent figure in the AI revolution, joined the "No Pires" podcast to discuss the current state and future direction of AI, particularly focusing on the concept of "agentic AI." Ng emphasizes that progress in AI capabilities will come from multiple vectors, including agentic workflows, multi-modal models, and new technologies, and not solely from scaling models as heavily emphasized by large companies with significant PR influence. Ng introduced the term "agentic AI" to move away from debates about what constitutes an agent and instead focus on building systems with varying degrees of autonomy. While marketing hype around agentic AI has accelerated rapidly, real business progress is also growing, albeit not as quickly as the marketing suggests. The biggest obstacle to implementing true AI agents is not technology but talent. He emphasizes the importance of systematic error analysis and evaluation in building effective agents, a skill often lacking in less experienced teams. Ng believes that a significant amount of work can be automated through agentic workflows but emphasizes that building such workflows requires external knowledge often locked in the heads of people. Currently, human engineers and product managers are crucial in making context-aware decisions that AI agents cannot yet replicate, particularly when dealing with proprietary or non-general knowledge. According to Ng, the strongest example of agency is AI coding agents. He highlights coding agents as one of the two clear buckets of economic value that include answering people's questions. He finds these agents highly autonomous in planning and executing multi-step processes to build software. He believes the economic value of coding has driven significant resources towards building effective coding agents. Ng prefers the term "AI-assisted coding" over "vibe coding" because the latter implies the task is simpler than it is. Using AI assistance makes coding a deeply intellectual exercise. Ng notes that AI-assisted coding is changing the nature of startups by accelerating coding speed and reducing costs. This shifts the bottleneck to product management, requiring product managers to rely on gut instinct and customer empathy. He is cautiously optimistic about tools that attempt to automate aspects of product management, such as market research using AI agents. Still, he believes they are not yet as effective as coding tools in replacing human product managers. Ng argues that founders with a strong understanding of AI technology are more likely to succeed in the current rapidly evolving landscape. Founders should be on top of emerging AI technologies, or they struggle to lead the company. The technology-oriented product leaders are more likely to succeed than business-oriented leaders. He emphasizes the importance of hard work and a willingness to challenge the status quo for those seeking to change the world. He observes that there are two main types of founders: those obsessed with their business winning and those obsessed with their customers winning. Ng values customer obsession. He emphasizes the need for decisiveness and rapid decision-making based on a deep understanding of the customer. Ng also addresses how his views and workflows are changing due to AI's rapid progress. When hiring engineers, he places a high value on AI skill set. Software engineers are the harbinger for other disciplines because tools are more advanced in software engineering. Furthermore, he suggests that the future nature of work might involve smaller, highly skilled teams with AI assistance outperforming larger, lower-cost teams. Ng believes that in the next five years, people will be more empowered and capable due to AI integration. Individuals embracing AI will be significantly more productive and capable than most realize.