Kotlin creator's new language: a formal way to talk to LLMs instead of English

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:

• Non-Deterministic Outputs: The same prompt can yield different results, breaking automated workflows.
• Poor Maintainability: "Prompt engineering" often involves tweaking fragile text strings without a clear audit trail.
• Hidden Complexity: Multi-step reasoning or strict data formatting is incredibly cumbersome to describe in plain English.
• Integration Challenges: It's difficult to seamlessly weave a natural language prompt into a deterministic software process, a key requirement for a platform like Mewayz where modules must interoperate reliably.

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.

"The core idea is to make prompts composable, testable, and manageable... It's about making this interaction with the LLM a proper software engineering artifact." – Andrey Breslav on the motivation behind the new language.

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.

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.