A coding agent is an AI tool that does programming work rather than only suggesting code: it reads the files in a project, makes changes, runs commands such as tests, and keeps going until the task is done or it needs you. Claude Code, OpenAI’s Codex, Google’s Gemini CLI and GitHub Copilot’s cloud agent are examples.
Think of a leaking sink. Autocomplete is a helper who hands you the next tool. A chatbot is a friend on the phone telling you how to fix it. A coding agent is the plumber you let into the house: it does the job itself, room by room. You still look at the work before you call it finished.
In a sentence: “I asked the coding agent to fix the broken sign-up form, and it found the bug, changed two files and ran the tests before showing me what it had done.”
How is a coding agent different from a chatbot?
The plain point: the difference is who carries out the steps.
| Tool | What it does |
|---|---|
| Code autocomplete | Differs: suggests the next few lines as you type. You do every step yourself. |
| Chatbot | Differs: writes code into the chat when you ask. You copy it, run it and come back with the error. |
| Coding agent | Differs: works inside the project. It opens files, edits them, runs commands and checks the result in a loop. |
| Vibe coding | Differs: not a tool but a way of working: accepting AI-written code without reading it. You can do it with or without an agent. |
Anthropic’s guide to its own agent puts the difference this way: “Unlike a chatbot that answers questions and waits, Claude Code can read your files, run commands, make changes, and autonomously work through problems while you watch, redirect, or step away entirely.”1 It is a kind of AI agent, one whose job is software.
What can a coding agent actually do?
The plain point: read, change, run, report, with each vendor describing it in much the same words.
Anthropic describes Claude Code as “an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with your development tools.”2 OpenAI calls Codex CLI “a coding agent from OpenAI that runs locally on your computer.”3 Google calls Gemini CLI “an open-source AI agent that brings the power of Gemini directly into your terminal.”4 A terminal is the text window where you type commands to the computer.
GitHub says its Copilot cloud agent can “research a repository, plan and make code changes, and create pull requests for you to review.”5 A repository is the folder that holds a project’s code; a pull request is a bundle of proposed changes that a person reviews before it is merged in.
Where does a coding agent run?
The plain point: on your own computer, where it can see your files, or in the cloud, where it works on a copy and hands the changes back.
Some run in both places. Anthropic says the web version of Claude Code lets you “kick off long-running tasks and check back when they’re done”, and OpenAI points people who want a cloud agent to Codex Web.63
Why does it matter who checks the work?
The plain point: a tool that can change your files can also break them, so the checks are part of the job.
You can decide which steps need your yes. Claude Code’s documentation, for example, says it “uses a tiered permission system to balance power and safety”, and in its manual mode editing files needs your approval.7 Permissions for AI agents explains what to allow.
AI-written code also needs reading. In an older study of GitHub Copilot’s code suggestions, “approximately 40%” of programs written for security-sensitive test tasks were vulnerable.8 That study tested a suggestion tool from 2021, not today’s agents, but the lesson carries over: read the changes and run the tests before anything goes live.
A coding agent does the steps itself: read, edit, run, check. That makes it fast and also makes your review the safety net. Decide what it may touch, and read what it changed.
Questions people ask
Do you need to know how to code to use a coding agent?
Not to start, but someone has to be able to check the result. In this site’s reading, an agent is safe for throwaway projects without code knowledge, and anything holding logins, payments or other people’s data needs a person who can read the code; vibe coding explained sets out why.Reasoning, 19 Oct 2026 — this site’s advice, drawing on the security study cited above.
How do you give a coding agent standing instructions?
Most read an instruction file kept in the project. Anthropic says Claude Code reads a CLAUDE.md file at the start of every session, and “If your repository already has an AGENTS.md for other coding agents, Claude Code can read that in place of a CLAUDE.md.”6 How to prompt AI agents covers what to put in it.
Are any coding agents open source?
Some are. Gemini CLI’s page lists “License: Apache License 2.0”, an open-source licence.4 The AI model behind the tool usually still runs on the company’s servers, so check each tool’s own page.
Where to go next
- How to brief one: how to prompt AI agents.
- What to let it touch: permissions for AI agents.
- The tools around it: the AI coding stack, explained.
- The wider idea: what is an AI agent? and agentic AI.
- The words: the AI glossary.
QUICK CHECK · 3 QUESTIONS
Pick an answer to see if you got it. Nothing is recorded.
1. What does a coding agent do that a chatbot does not?
- It writes code into the chat for you to copy and run
- It works inside the project, editing files and running commands itself
- It only suggests the next line while you type
Show the answer
Answer: It works inside the project, editing files and running commands itself. The page says a chatbot answers and waits, while a coding agent opens files, edits them, runs commands and checks the result in a loop.
2. Is vibe coding the same thing as a coding agent?
- No, vibe coding is a way of working, not a tool
- Yes, they are two names for the same tool
- Yes, but only when the agent runs in the cloud
Show the answer
Answer: No, vibe coding is a way of working, not a tool. The page says vibe coding means accepting AI-written code without reading it, which you can do with or without an agent.
3. Why does the page say your review matters so much?
- Coding agents cannot run tests
- Reviews make the agent work faster
- A tool that can change your files can also break them
Show the answer
Answer: A tool that can change your files can also break them. Because the agent does the steps itself, checking what it changed is the safety net, and the page’s study shows AI-written code can carry flaws.
Related terms: Agentic workflow · API · Benchmark · Embedding · Eval · Function calling
- 1Anthropic, Best practices for Claude Code, Claude Code docs, read at source 13 Oct 2026 (as cited on this site’s page on prompting AI agents): “Unlike a chatbot that answers questions and waits, Claude Code can read your files, run commands, make changes, and autonomously work through problems while you watch, redirect, or step away entirely.”↑
- 2Anthropic, Claude Code overview, Claude Code docs, read at source 19 Oct 2026“Claude Code is an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with your development tools. Available in your terminal, IDE, desktop app, and browser.” ↑
- 3OpenAI, openai/codex, GitHub README, read at source 19 Oct 2026“Codex CLI is a coding agent from OpenAI that runs locally on your computer.” ↑a“If you are looking for the cloud-based agent from OpenAI, Codex Web, go to chatgpt.com/codex.” ↑b
- 4Google, google-gemini/gemini-cli, GitHub README, read at source 19 Oct 2026“Gemini CLI is an open-source AI agent that brings the power of Gemini directly into your terminal.” ↑a“License: Apache License 2.0”. ↑b
- 5GitHub Docs, GitHub Copilot cloud agent, read at source 19 Oct 2026“Find out how Copilot can research a repository, plan and make code changes, and create pull requests for you to review.” ↑
- 6Anthropic, Claude Code overview, read at source 19 Oct 2026“Run Claude Code in your browser with no local setup. Kick off long-running tasks and check back when they’re done”. ↑a“CLAUDE.md is a markdown file you add to your project root that Claude Code reads at the start of every session.” and “If your repository already has an AGENTS.md for other coding agents, Claude Code can read that in place of a CLAUDE.md.” ↑b
- 7Anthropic, Configure permissions, Claude Code docs, read at source 19 Oct 2026“Claude Code uses a tiered permission system to balance power and safety.” Its table lists “File modification” with approval required “Yes” in Manual mode. ↑
- 8Pearce et al., Asleep at the Keyboard? Assessing the Security of GitHub Copilot’s Code Contributions, arXiv, revised 16 Dec 2021, read at source 1 Oct 2026 (as cited on this site’s vibe coding page): “Of these, we found approximately 40% to be vulnerable.”↑
Pearce et al., Asleep at the Keyboard?, arXiv, revised 16 Dec 2021, read at source 1 Oct 2026.
The comparisons and examples on this page (the plumber, the sign-up form) are this site’s own ways of explaining, not claims from these sources.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK