What is an AI Agent (Sub-agent)?
A specialized worker that handles one focused job well.
An agent (often a 'sub-agent') is an AI configured with a specific role, instructions, and tools to handle a focused task — like reviewing code, mapping a codebase, or drafting research. Instead of one general assistant doing everything, you delegate to specialized agents that each excel at their job.
How sub-agents work
A sub-agent is typically defined by a Markdown file describing its role, its system prompt, which tools/MCP servers it can use, and which model to run on. A coordinating agent can spawn several sub-agents — even in parallel — and combine their results.
Why delegate
Specialization improves quality and keeps context focused. A 'code reviewer' agent tuned for security review will outperform a generalist, and running agents in parallel (e.g. mapping a large codebase) is far faster.
Where they run
Sub-agents are common in Claude Code and multi-agent frameworks, and increasingly work across Codex, Cursor, and other harnesses.
Common examples
- A code-review agent that leaves line-level PR feedback
- A cartographer agent that maps and documents a codebase
- An architect agent that interviews you and designs a system
Explore agents
Browse the directory of agents — all linking to GitHub.
Popular agents
Code Reviewer
A specialist sub-agent that performs thorough, standards-aware pull-request reviews.
Equity Research Agent
Produces institutional-grade equity research from a ticker or thesis.
Cartographer
Maps and documents codebases of any size using parallel AI subagents.
Agent — FAQ
How is an agent different from an MCP server?+
An MCP server provides tools/data. An agent is the AI worker that uses those tools to accomplish a task.
Can agents call other agents?+
Yes — a coordinating agent can delegate to multiple sub-agents and merge their outputs.