Kotlin 创建者的新语言：与法学硕士交谈的正式方式，而不是英语

作为程序的提示：人工智能时代的新语言

多年来，与大型语言模型 (LLM) 的交互感觉就像是用外语进行高风险对话。我们精心制作英语提示，希望得到最好的结果，但经常面临不一致、含糊不清和令人沮丧的缺乏精确性。 Kotlin 的创建者 JetBrains 的 Andrey Breslav 正在带头解决这个问题。他的团队正在开发一种新的编程语言，暂定名为“Kira”，其设计目的不是与传统意义上的计算机对话，而是与人工智能对话。这一举措标志着范式的转变：从非正式的提示到正式的、可执行的规范。对于在 Mewayz 等平台内利用人工智能的企业来说，这种演变可以将混乱的人工智能实验转变为模块化业务操作系统的可靠、版本控制的组件。

为什么英文提示不符合业务逻辑

虽然自然语言很直观，但对于复杂的任务来说它是非常不精确的。像“分析本季度的销售额并创建摘要”这样的指令留下了巨大的解释空间。什么指标？什么格式？与什么基线相比？这种模糊性导致：

非确定性输出：相同的提示可能会产生不同的结果，从而破坏自动化工作流程。

可维护性差：“即时工程”通常涉及在没有明确审计跟踪的情况下调整脆弱的文本字符串。

隐藏的复杂性：用简单的英语描述多步骤推理或严格的数据格式非常麻烦。

集成挑战：将自然语言提示无缝地融入确定性软件流程是很困难的，而这是像 Mewayz 这样的平台的关键要求，其中模块必须可靠地互操作。

可靠人工智能交互的正式规范

Breslav 与 Kira 的愿景是创建一种语言，开发人员可以为法学硕士编写清晰、结构化且可测试的规范。您可以编写一个正式的指令集来定义任务的约束、预期的输出结构，甚至后备行为，而不是暗示性的段落。可以将其视为从向快递员提供模糊指示，转变为提供带有明确检查点的精确、机器可读的 GPS 路线。这种形式化方法可确保在 Mewayz 工作流程中执行任务的人工智能代理（例如，从电子邮件中提取发票数据并填充数据库）以可预测的结构执行，使其输出成为链中下一个模块的可信输入。

“核心思想是使提示可组合、可测试和可管理......这是为了使与 LLM 的这种交互成为适当的软件工程工件。” – Andrey Breslav 谈新语言背后的动机。

对 Mewayz 等模块化业务系统的影响

对商业操作系统平台的潜在影响是深远的。在 Mewayz 中，不同的业务功能（CRM、项目管理、计费）作为互连模块存在，人工智能可以充当智能结缔组织。法学硕士的正式语言将允许将这些人工智能增强的连接构建为强大的、可部署的组件。例如，开发人员可以编写一个“Kira”脚本来定义客户支持分类代理：指定如何对工单意图进行分类、提取关键实体以及为工单模块格式化结构化 JSON 有效负载。该脚本成为 Mewayz 生态系统中受版本控制、可调试的资产，而不是隐藏在提示框中的神奇咒语。它将人工智能从一个富有创造力但不可靠的合作伙伴转变为一个正式的、可操作的引擎。

未来：从快速制作到人工智能编程

虽然仍处于早期开发阶段，但这一概念预示着“人工智能编程”将成为一门独特学科的未来。目标不是消除自然语言——它永远是一个很好的起点——而是在人类意图和机器执行之间提供严格的桥梁

Frequently Asked Questions

The Prompt as a Program: A New Language for the AI Age

For years, interacting with large language models (LLMs) has felt like a high-stakes conversation in a foreign language. We craft elaborate English prompts, hoping for the best, but often face inconsistency, ambiguity, and a frustrating lack of precision. The creator of Kotlin, JetBrains' Andrey Breslav, is spearheading a radical solution to this problem. His team is developing a new programming language, tentatively called "Kira," designed not for talking to computers in the traditional sense, but for talking to AI. This move signals a paradigm shift: from informal prompting to formal, executable specification. For businesses leveraging AI within platforms like Mewayz, this evolution could transform chaotic AI experimentation into a reliable, version-controlled component of a modular business operating system.

Why English Prompts Are Failing Business Logic

While natural language is intuitive, it's notoriously imprecise for complex tasks. An instruction like "analyze this quarter's sales and create a summary" leaves vast room for interpretation. What metrics? What format? Compared to what baseline? This ambiguity leads to:

Formal Specifications for Reliable AI Interactions

Breslav's vision with Kira is to create a language where developers can write clear, structured, and testable specifications for an LLM. Instead of a suggestive paragraph, you'd write a formal instruction set that defines the task's constraints, expected output structure, and even fallback behaviors. Think of it as moving from giving vague directions to a courier, to providing a precise, machine-readable GPS route with defined checkpoints. This formal approach ensures that an AI agent tasked within a Mewayz workflow—for instance, extracting invoice data from emails and populating a database—executes with predictable structure, making its output a trustworthy input for the next module in the chain.

Implications for Modular Business Systems Like Mewayz

The potential impact on business OS platforms is profound. In Mewayz, where different business functions (CRM, project management, billing) exist as interconnected modules, AI can act as the intelligent connective tissue. A formal language for LLMs would allow these AI-augmented connections to be built as robust, deployable components. A developer could, for example, write a "Kira" script that defines a customer support triage agent: specifying how to classify ticket intent, extract key entities, and format a structured JSON payload for the ticketing module. This script becomes a version-controlled, debuggable asset within the Mewayz ecosystem, not a magical incantation hidden in a prompt box. It turns AI from a creative but unreliable partner into a formalized, operational engine.

The Future: From Prompt Crafting to AI Programming

While still in early development, the concept heralds a future where "AI programming" is a distinct discipline. The goal isn't to eliminate natural language—it will always be a great starting point—but to provide a rigorous bridge between human intent and machine execution. For businesses, this means the powerful capabilities of LLMs can finally be integrated into core processes with software-grade reliability. Platforms that embrace this shift, like Mewayz, will enable their users to build not just with AI, but on top of AI, creating truly intelligent and automatable business systems where every interaction, even with a neural network, is defined with clarity and purpose.