AI Agents - Glossary

An autonomous AI system that can perform tasks, make decisions, and interact with environments or users with minimal human intervention. In coding contexts, AI agents can write, review, and modify code across multiple files.

Examples: Cline, Open SWE, Cursor's Composer

A unique identifier used to authenticate requests to AI services. Most AI coding tools require API keys from providers like OpenAI, Anthropic, or Google.

Security Note: Never commit API keys to version control.

Real-time code completion suggestions provided by AI models as you type. Also known as "inline suggestions" or "ghost text."

Tools: GitHub Copilot, Cursor, Windsurf

StackBlitz's AI-powered development environment for building full-stack applications directly in the browser with AI assistance.

Key Features: Instant deployment, AI code generation, full-stack templates

Conversational AI interface that allows developers to ask questions, request code changes, and get explanations in natural language.

Examples: Cursor Chat, GitHub Copilot Chat, Claude Code

An autonomous coding agent that can handle complex development tasks, manage files, run commands, and make decisions about code changes.

Capabilities: Multi-file editing, command execution, autonomous problem-solving

AI-powered feature that suggests code snippets, function implementations, or entire blocks of code based on context and comments.

Types: Inline completion, tab completion, function generation

AI feature that allows conversational interaction with an entire codebase, enabling questions about project structure, dependencies, and implementation details.

Tools: Cursor, Windsurf, some VS Code extensions

OpenAI's now-deprecated language model specifically trained on code. It was the foundation for GitHub Copilot and many early AI coding tools.

Legacy: Replaced by GPT-3.5-turbo and GPT-4 in March 2023

Cursor's AI agent feature that can edit multiple files simultaneously to implement complex changes across a project.

Use Cases: Refactoring, feature implementation, architecture changes

The amount of text (measured in tokens) that an AI model can process in a single request. Larger context windows allow for better understanding of large codebases.

Sizes: GPT-4: 8K-128K tokens, Claude: up to 200K tokens

AI-first code editor built on VS Code that integrates advanced AI capabilities like codebase chat, multi-file editing (Composer), and AI-powered code generation.

Key Features: Codebase chat, Composer, Apply feature, privacy controls

A view showing the differences between two versions of code, often used by AI tools to show proposed changes before applying them.

Context: AI tools often show diffs for review before making changes

Vector representation of code or text that captures semantic meaning. Used by AI tools to find relevant context and similar code patterns.

Applications: Code search, similarity detection, context retrieval

Process of training an AI model on specific data to improve performance for particular tasks or domains.

Context: Some tools offer custom models fine-tuned for specific codebases

Faded, inline code suggestions that appear as you type, which can be accepted or rejected. Also called "inline suggestions."

Interaction: Usually accepted with Tab key, rejected with Escape

AI pair programmer developed by GitHub and OpenAI that provides code suggestions and completions directly in your editor.

Versions: Individual, Business, Enterprise with different features

OpenAI's advanced language model used by many modern AI coding tools. Offers better reasoning, longer context windows, and improved code generation compared to earlier models.

Variants: GPT-4, GPT-4-turbo, GPT-4o

When AI models generate plausible-sounding but incorrect or non-existent code, APIs, or information.

Mitigation: Always verify AI-generated code, especially API calls and library usage

Code completions that appear directly in the editor as you type, typically shown as ghost text that can be accepted or rejected.

Also Known As: Ghost text, autocomplete, tab completion

AI model setting that ensures responses are formatted as valid JSON, useful for structured data generation and API responses.

Use Cases: Configuration generation, data transformation, API mocking

AI model trained to understand and generate human language and code. Examples include GPT-4, Claude, and Gemini.

Specialization: Some models are specifically trained or fine-tuned for code generation

Neural networks with billions of parameters trained on vast amounts of text and code data. Foundation of most AI coding tools.

Examples: GPT-4, Claude, Gemini, LLaMA

AI capability to make coordinated changes across multiple files in a project, understanding dependencies and relationships between files.

Tools: Cursor Composer, Cline, Windsurf Workflows

AI capability to generate code from human language descriptions and comments.

Example: "Create a function that sorts users by registration date" → Generated sorting function

Open-source AI software engineer that can handle complex development tasks autonomously, including planning, coding, testing, and debugging.

Capabilities: Autonomous task completion, multi-step problem solving

Open-source software engineering agent focused on repository analysis, code understanding, and automated modifications.

Strengths: Codebase analysis, automated refactoring

Development practice where two programmers work together. AI pair programming refers to AI tools acting as a coding partner.

AI Context: GitHub Copilot was marketed as an "AI pair programmer"

The practice of crafting effective prompts to get better results from AI models. Important for maximizing AI coding tool effectiveness.

Techniques: Clear instructions, examples, context specification

AI technique that combines retrieval of relevant information with generation, allowing models to access specific knowledge bases or codebases.

Application: Codebase-aware AI tools use RAG to provide relevant context

Replit's AI coding assistant integrated into their browser-based development environment.

Features: Code completion, explanation, debugging, chat interface

Search technique that understands meaning and context, not just keywords. Used by AI tools to find relevant code snippets and documentation.

Benefits: More accurate context retrieval, better code suggestions

AI-generated code proposals that developers can accept, reject, or modify. Can be inline suggestions or larger code blocks.

Types: Inline, multi-line, function-level, file-level

AI model parameter controlling randomness in output. Lower values (0.1-0.3) produce more deterministic code, higher values (0.7-1.0) more creative but less predictable results.

Code Generation: Lower temperatures typically preferred for coding tasks

Unit of text processing in AI models. Roughly equivalent to words or code symbols. Token limits determine how much context AI can process.

Estimation: ~1 token per 4 characters in English, varies for code

Database optimized for storing and searching vector embeddings. Used by AI tools for semantic search and context retrieval.

Examples: Pinecone, Weaviate, Chroma

AI-enhanced terminal that provides intelligent command suggestions, explanations, and workflow automation.

Features: Command completion, AI chat, workflow blocks

Codeium's AI-first code editor with advanced codebase understanding and workflow integration capabilities.

Strengths: Codebase awareness, workflow automation

Practice of providing relevant code context to AI tools while managing token limits and maintaining performance.

Techniques: File selection, relevant imports, focused prompts

Process of having AI systems review code for issues, improvements, and adherence to standards.

Focus Areas: Security, performance, maintainability, best practices

Technique of breaking complex tasks into smaller prompts that build on each other's outputs.

Benefits: Better results for complex tasks, easier debugging

AI systems that can independently complete entire development tasks with minimal human intervention.

Examples: Feature implementation, bug fixing, refactoring

Development approach that integrates AI assistance as a primary part of the coding workflow from the start.

Characteristics: AI-native tools, prompt-driven development, continuous AI collaboration

AI's ability to understand code structure, dependencies, patterns, and relationships across large codebases.

Applications: Smart refactoring, impact analysis, automated documentation