以下是内容的中文翻译： 在2026年9月28日录制的一期播客节目中，TypeSafe创始人兼领导者Diogo与Martin和另一位主持人一起，讨论了他的公司产品JEV，以及该产品在软件开发中对人工智能的革命性应用方法。Diogo因其独特融合了人工智能研究、系统专业知识和编程敏锐度而被誉为“英雄”，他通过发问“他妈的所有自动化到底在哪儿？”来挑战当前关于人工智能的主流论述。 Diogo遗憾地指出，尽管人工智能拥有“令人难以置信的智能”能力，但除了聊天机器人和编码代理之外，它在“所有其他方面都几乎毫无用处”，未能兑现广泛自动化的承诺。他认为，虽然当前的AI工具，如Cloud Code或Codex（被Gary Tan称为“即时软件”），在生成代码方面表现出色，但它们仅仅是复制人类会写的东西，通常缺乏真正的语义理解或架构洞察力。 另一方面，TypeSafe的JEV旨在通过扩展软件本身的能力来创建“智能软件”。Diogo将JEV描述为一种“新原语”或“软件内部的智能层”，其功能类似于一个库，允许开发者用自然语言描述意图，然后该意图与状态机交互，以具有置信水平的方式做出决策。他解释说，这种方法使那些*应该*可以自动化的事情，*真正*变得可以自动化，超越了人机交互，去构建“真正的软件”。尽管承认JEV是一个“分类器”，Diogo强调其设计是为了实用性，甚至可能超越2019年的机器学习工程（MLE）团队。 Diogo的个人经历，从屡获殊荣的数学竞赛选手到一位偶然进入AI研究领域的计算机科学家，塑造了他的务实愿景。他在Kaggle竞赛中获胜并非依靠复杂的数学，而是通过广泛的自动化，这促使他采取了系统优先的AI方法。他的职业生涯使他先后在Google Brain和OpenAI工作，并在那里深入参与了早期GPT的开发。 TypeSafe哲学的核心是其座右铭：“我们构建产品（prod），而不是上帝（God）。”Diogo强烈反对关于单一全能AI的“单模型迷魂汤”（mono model Kool-Aid），相反，他倡导一个工作得到增强、世界变得“更好得多”的未来。他对“与现实的脱节”感到沮丧，即人工智能拥有巨大潜力，但基本任务（如OpenAI自2020年以来一直试图自动化的客户服务）却仍未实现自动化。 JEV的一个关键区别在于其对**可靠性**的关注。Diogo解释说，这意味着确保“每次都具备相似的智能”（鲁棒性），而不仅仅是正常运行时间或严格的确定性。最终目标是让开发者无条件信任JEV，使他们能够进入“永久心流状态”，而无需测试示例查询。 Diogo预测JEV将对软件行业产生变革性影响。他认为，虽然编码代理可能自动化语法，但它们通常在架构和语义方面表现不足。相比之下，JEV在软件*内部*提供了新功能，从而产生本质上更好的应用程序。他指出，最初关于“SaaS末日”（即AI编码代理会降低SaaS公司价值）的担忧，已被采用JEV的SaaS公司的“反向SaaS末日”或“SaaS盛典”（Sassapalooza）所取代，因为JEV极大地提升了他们产品的实用性。Diogo展望了一个多项选择表单消失，所有技术都“做我所想”的未来。 最终，Diogo认为JEV弥合了长期以来人工智能和软件之间“夜晚的船只”（指彼此擦肩而过、缺乏有效连接）的鸿沟。以前，人工智能的自然语言输出常常由于软件无法直接处理，而不得不反馈给人类或其他大型语言模型。JEV能够将人工智能直接映射到状态机，提供了一种富有成效的方式，将智能深入集成到系统“内部”，为概率编程开辟了新时代，并实现了软件系统的根本性重建。

In a podcast recorded on 2026-09-28, TypeSafe founder and leader Diogo joined Martin and another host to discuss his company’s product, JEV, and its revolutionary approach to AI in software development. Diogo, described as a "hero" for his unique blend of AI research, systems expertise, and programming acumen, challenged the prevailing narrative around AI by asking, "Where the f*** is all the automation?" Diogo lamented that despite AI's "unbelievably smart" capabilities, it remains largely "useless at all other stuff" beyond chatbots and coding agents, failing to deliver on the promise of widespread automation. He argued that while current AI tools like Cloud Code or Codex (dubbed "just-in-time software" by Gary Tan) are excellent at generating code, they merely replicate what a human would write, often without true semantic understanding or architectural insight. TypeSafe's JEV, on the other hand, aims to create "smart software" by expanding what software itself can do. Diogo described JEV as a "new primitive" or an "intelligent layer inside the software," functioning like a library that allows developers to describe intent in natural language, which then interacts with a state machine to make decisions with confidence levels. This approach, he explained, makes things that *should* be automatable, *actually* automatable, moving beyond human-in-the-loop interactions to build "real software." While acknowledging JEV as a "classifier," Diogo emphasized its design for practical utility, potentially outperforming even 2019 Machine Learning Engineering (MLE) teams. Diogo's personal journey, from award-winning mathlete to a computer scientist who fell into AI research, informed his pragmatic vision. He won a Kaggle competition not through sophisticated math but by automating extensively, pushing him towards a systems-first approach to AI. His career path led him through Google Brain and OpenAI, where he was deeply involved in early GPT developments. Central to TypeSafe's philosophy is the motto, "we build prod, not God." Diogo vehemently disagrees with the "mono model Kool-Aid" of a single all-powerful AI, advocating instead for a future of enhanced jobs and a "way better world." He expressed frustration over the "discordance with reality" where AI has immense potential but basic tasks (like customer service, which OpenAI has been trying to automate since 2020) remain unautomated. A key differentiator for JEV is its focus on **reliability**. Diogo explained that this means ensuring "similar intelligence every time" (robustness), not just uptime or strict determinism. The ultimate goal is for developers to trust JEV implicitly, allowing them to achieve a "perma flow state" without needing to test example queries. Diogo projected that JEV would have a transformative effect on the software industry. He argued that while coding agents might automate syntax, they often fall short on architecture and semantics. JEV, by contrast, provides new capabilities *within* the software, leading to inherently better applications. He noted that the initial "SaaS-pocalypse" fears (that AI coding agents would devalue SaaS companies) have been replaced by an "inverse saspocalypse" or "Sassapalooza" for SaaS companies adopting JEV, as it dramatically enhances their product's utility. Diogo envisions a future where multi-choice forms disappear, and all technology "does what I mean." Ultimately, Diogo sees JEV as bridging the long-standing gap where AI and software were like "ships in the night." Previously, AI's natural language outputs often had to be fed back to humans or other LLMs due to software's inability to process them directly. JEV's ability to map AI directly to a state machine offers a productive way to integrate intelligence deeply into system "guts," opening up new eras for probabilistic programming and enabling a fundamental rebuilding of software systems.