Skip to main content
ModelTerms

Agents & Tools · intermediate

Agentic Coding (coding agent)

Agentic coding is an LLM-driven workflow where the model reads code, plans changes, edits files, runs commands, and iterates against feedback — autonomously closing tasks rather than just suggesting code.

Explanation

Old-style AI coding: autocomplete in your editor. Agentic coding: hand the model a task ("fix this bug", "ship this feature") and it edits files, runs tests, reads tracebacks, and tries again.

Claude Code, Cursor's Composer/Agent, Aider, Devin, and OpenHands are agentic coding tools. Under the hood they're ReAct loops with file-edit, shell-execute, and read tools plus careful prompt engineering and safety guardrails.

Quality is rising fast: SWE-bench scores went from <5% in 2023 to >60% by mid-2025 for the strongest agents.

Examples

  • Claude Code closing a GitHub issue end-to-end.
  • Cursor Agent refactoring a module across many files.

When to use agentic coding

Whenever a task is well-scoped, has objective success criteria (tests pass, types check), and would take a human 15+ minutes.

Frequently asked

What is Agentic Coding?

Agentic coding is an LLM-driven workflow where the model reads code, plans changes, edits files, runs commands, and iterates against feedback — autonomously closing tasks rather than just suggesting code.

What is an example of agentic coding?

Claude Code closing a GitHub issue end-to-end.

How is Agentic Coding related to Agent?

Agentic Coding and Agent are both agents & tools concepts. An AI agent is an LLM-driven system that decides which actions to take, executes them via tools, observes the results, and iterates until a goal is met.

When should I use agentic coding?

Whenever a task is well-scoped, has objective success criteria (tests pass, types check), and would take a human 15+ minutes.

Is Agentic Coding considered intermediate?

Agentic Coding is generally considered intermediate-level material in the AI and LLM space.

AgentAgents & Tools

An AI agent is an LLM-driven system that decides which actions to take, executes them via tools, observes the results, and iterates until a goal is met.

Tool UseAgents & Tools

Tool use is when an LLM can call external functions — APIs, code interpreters, databases, web fetchers — and read their results. The mechanism that turns chat into action.

ReActAgents & Tools

ReAct is a prompting pattern that interleaves reasoning ("Thought:") with actions ("Action:") and observations ("Observation:"). It is the foundation of most tool-using agents.

Model Context ProtocolAgents & Tools

MCP is an open standard for connecting LLMs to external tools and data sources. It defines a JSON-RPC protocol so any client (Claude Desktop, Cursor, IDE plugins) can talk to any MCP server.

SWE-benchEvaluation

SWE-bench is a benchmark of ~2.3K real GitHub issues from popular Python repos. The model must read the codebase, understand the bug, and write a patch that passes the existing tests.

Side-by-side comparisons

Sources