What is an AI coding agent?

An AI coding agent is software that uses artificial intelligence (usually large language models) to write, edit, or review code. Instead of only suggesting completions as you type, an agent can take a task (e.g. "add a login API" or "refactor this module") and produce or change code in your project.

How do AI coding agents work?

Agents use language models trained on code and natural language. You give them:

• A task or instruction (in plain language or structured form)
• Context such as file paths, repo structure, or existing code

The agent then generates or edits code, runs commands (e.g. tests, linters), and can iterate until the task is done or you stop it.

Examples of AI coding agents

• Claude Code – Anthropic's agent for writing and refactoring code
• Cursor – AI-powered editor; Cursor CLI can act as an agent
• GitHub Copilot – GitHub's AI pair programmer
• OpenAI Codex – Powers Copilot and other code-generation tools
• Gemini CLI – Google's Gemini model via CLI
• Others (Amp, Opencode, Qwen Code, Factory Droid, etc.)

Vibe Kanban supports many of these so you can run them in parallel and switch between agents. See supported agents.

Agents vs. simple autocomplete

Autocomplete (e.g. Copilot suggestions in the editor) suggests the next line or block as you type. An agent takes a higher-level task, works across files, and can run tools. Vibe Kanban is built for agents: it gives each agent its own workspace and lets you run several at once, then review their changes in one place.

Why run multiple agents?

Different agents are good at different things. Running them in parallel lets you:

• Tackle several tasks at once (e.g. frontend, backend, tests)
• Use the best agent per task
• Keep work isolated so agents don't overwrite each other

Parallel execution in Vibe Kanban →

Learn more

• Glossary – Terms like worktree, MCP, task
• Supported agents
• Use cases
• Get started with Vibe Kanban