AI Lang

Artificial Intelligence Language

A new way to communicate with AI

A high-dimensional meta-language system designed to deliver human intent to AI systems with maximum precision

Goal of AI Lang

Take precise control over what AI creates. Not approximate predictions, but knowing exactly what you'll get—down to the details—and ensuring the results you want.

What is AI Lang?

AI Lang (Artificial Intelligence Language) is a sophisticated communication language system that goes beyond traditional natural language-based prompt engineering. It enables humans to interact with AI by understanding and utilizing the AI system's internal structure, knowledge access methods, agent collaboration patterns, and context management approaches to deliver human intent most efficiently.

In other words, it is an independent and structured language for AI that transcends conventional "prompt writing techniques".

"If a prompt is a sentence, AI Lang is more like a 'grammar'."

Why is AI Lang needed?

When humans write prompts in natural language, intermediate tools (chatbots, agents, etc.) mix the following elements and deliver them to AI:

System Prompts

Context (Memory, Session Information)

RAG (Retrieval-Augmented Generation)

Agent Chains (Agent, MCP, etc.)

Tool Usage Instructions

Format Constraints, Policies, etc.

Web Search and Internet Information

API Calls and External Service Integration

Vector Search (Embedding-based Search)

AI Optimization Settings (Temperature, Structured Input, etc.)

Model Selection and Routing

Humans cannot communicate directly with AI. Intermediate tools (chatbots, agents, etc.) mix all the above elements with the user's prompt and deliver it to AI, but humans cannot know what input is actually being delivered. The core concept here is to create 'AI Lang' so that humans can communicate better with AI and control AI behavior more precisely, even through these intermediaries.

Differences from Prompt Engineering

Category	Prompt Engineering	AI Lang
Foundation	Natural language optimization	Creating a new language between humans and AI
Goal	Writing better-readable prompts	Designing language optimized for AI's internal structure
Scope	Instructions for specific models	Entire system/context/tools/agents
Role	Good request writing method	Meta-language for conveying intent to AI most efficiently
Form	Natural language-centric	Natural language + Structured language + Rule-based language

Key Features

AI-Friendly Structuring

Structured AI-Friendly Language

Expresses commands, constraints, intent, context, and roles more explicitly than natural language.

AI System Layer Awareness

System Layer Awareness

Unifies RAG, tool calls, and multi-agent logic into a single grammar.

Extensible Meta-Language

A flexible structure based on sub-rules that can be applied even when models change.

Intent-to-Action Loss Minimization

Reduces ambiguity and compresses meaning into a form that AI can execute precisely.

AI Lang Grammar (YAML-based)

AI Lang uses YAML format to convey intent to AI in a structured way. It leverages YAML's concise syntax and readability to systematically express prompt engineering techniques.

Basic Structure

Basic Structure ailang

intent:   # CREATE, TRANSFORM, ANALYZE, EXPLAIN, DECIDE, VALIDATE

reason: |
  

persona:
  role: 
  expertise: 

context:
  domain: 
  language: 

current_state: |
  

reproduce:
  - 
  - 

expected_result: |
  

guidelines:
  - 
  - 

constraints:
  must:
    - 
  should:
    - 
  must_not:
    - 

output:
  type: 
  format: 

style:
  tone: 
  length: 

request: |

Core Keywords

intent grammar_key_intent_required

Declares the intent type (required)

reason

Explains why the current task is necessary

persona

Assigns role/expertise

context

Defines context information such as domain, language, framework

current state

Describes the current code, situation, or problem state

reproduce

Describes the steps to reproduce the error or bug

expected result

Specifies the desired final outcome or behavior

guidelines

Provides step-by-step work guidelines or rules to follow

constraints

Specifies must/should/must_not conditions

output

Specifies output format and structure

style

Specifies response tone and style

Intent Type Examples

intent: CREATE. Add a chat feature.

intent: UPDATE. Improve the existing login screen.

intent: TRANSFORM. Translate this document to Korean.

intent: ANALYZE. Analyze performance issues in this code.

intent: EXPLAIN. Explain the current authentication system.

intent: VALIDATE. Verify if this API response is correct.

intent: DEBUG. Fix the login failure bug.

intent: TROUBLESHOOT. Resolve the server connection error.

intent: MIGRATE. Migrate from MySQL to PostgreSQL.

intent: DECIDE. Which should I choose, React or Vue?

YAML Syntax Rules

1 Use indentation for hierarchy (2 spaces recommended)
2 Key-values are written in key: value format
3 Arrays are indicated with hyphens (-)
4 Multi-line text uses pipe (|) or angle bracket (>)
5 intent is required; other keywords are optional
6 Write request content in natural language under the request key

Prompt Engineering Techniques

In 2023, prompt engineering was a hot topic, but by 2026, it has become a fundamental skill. Still, knowing these techniques makes a clear difference. AI Lang systematizes these techniques in YAML syntax.

Core Concept: Humans cannot directly communicate with AI. We interact with tools (software) through devices, and the tools communicate with AI. Therefore, prompt engineering is about structuring requests so tools can understand them well.

What is a Prompt? It is a call to action to LLM (Large Language Model).

How LLM Generates Text: It generates one word at a time based on the question, then repeatedly generates the next word based on previously generated words. The response pattern is determined by the question format.

Persona (Role Assignment)

Assigning a clear role to AI to focus attention and narrow down data exploration scope

You are a [role]. Answer according to [role's criteria/goals].

System Prompt

Default prompt always sent to AI with user queries (e.g., CLAUDE.md, /command)

User Prompt

The actual user requirements in the prompt

Context

Essential data provided for problem solving - file paths, images, error messages, etc.

BreadCrumb Strategy

Technique to precisely indicate the desired modification location as a path

Mobile → Top Menu → Menu Open → 'Others' item

Output Format

Specify desired response format like JSON, YAML, Markdown, or provide examples

COT (Chain of Thoughts)

Technique to request step-by-step thinking for problem solving

1. Identify core issue → 2. Plan → 3. Execute → 4. Verify

TOT (Trees of Thoughts)

Technique to analyze from multiple perspectives (branches) and integrate

Branch 1: Svelte perspective / Branch 2: TailwindCSS perspective / Branch 3: Node.js perspective → Synthesize

Task Decomposition

Decomposing complex tasks into steps and solving sequentially

(1) Plan → (2) Complete UI Design → (3) Implement Logic

Structured Format (5W1H)

Clearly structure overview, goal, guidelines, current state, and expected results

Tone and Style

Specify response length, format, and tone for consistent output

Be concise, within 4 lines, key points only

Examples

Code Generation

Request for user authentication API implementation using Python FastAPI

Code Generation ailang

intent: Implement JWT token-based user authentication API

persona:
  role: FastAPI Development Expert
  expertise: Python Backend, Authentication Systems

context:
  domain: Software Development
  language: Python
  framework: FastAPI

constraints:
  must:
    - Ensure type safety
    - Support async processing
  should:
    - Include detailed documentation

output:
  type: code
  format: Python
  comments: detailed in English

style:
  concise: true
  max_lines: 100

request: |
  Please implement user authentication API endpoints.
  Use JWT token-based authentication.

Document Analysis

Request for software license agreement review

Document Analysis ailang

intent: Analyze risks in software license agreement

persona:
  role: Legal Document Analysis Expert
  expertise: Contract Review, Risk Analysis

context:
  domain: Legal
  document_type: Contract

constraints:
  focus:
    - Risk factors
    - Obligation clauses
  analysis_depth: detailed

output:
  type: structured report
  sections:
    - Summary
    - Risk Analysis
    - Recommendations
  language: English

request: |
  Please review the attached software license agreement.

Content Transformation

Request for English to Korean translation of technical documentation

Content Transformation ailang

intent: Translate React official documentation to Korean

persona:
  role: Technical Translation Expert
  expertise: React, Frontend Documentation

context:
  source_language: English
  target_language: Korean
  domain: Technical Documentation

constraints:
  preserve:
    - Code blocks
    - URL links
    - Document format
  tone: formal

output:
  type: markdown
  preserve_structure: true

request: |
  Please translate the React official documentation.

Standardization Proposal

AI Lang Specification v0.1.0

1. Overview

1.1 Purpose

AI Lang is a structured meta-language designed to optimize communication between humans and AI systems. It reduces the ambiguity of traditional natural language prompts and delivers intent in a form optimized for AI system internals.

1.2 Scope

This specification defines the syntax rules, semantics, and extensibility of AI Lang. It aims to be a universal language system not dependent on specific AI models or platforms.

1.3 Terminology

Directive: A core component of AI Lang starting with the @ prefix
Block: A collection of key-value pairs enclosed in curly braces
Intent: The type of task requested from AI

2. Core Concepts

AI Lang consists of four core concepts: Intent Declaration, Context Block, Constraints Specification, and Output Format Designation.

3. Syntax Rules

All directives begin with the @ symbol, and block structures use curly braces. Values support strings, arrays, and nested block formats.

4. Reserved Words

Reserved directives: @intent, @context, @constraints, @output, @task, @agents. Reserved intent types: CREATE, TRANSFORM, ANALYZE, EXPLAIN, DECIDE, VALIDATE

5. Extensibility

Domain-specific custom directives can be defined, and extension grammar is declared through the @domain directive.