Coding agents: 7 AI coding tools

A coding agent takes a task in plain language and works on it with some independence. It reads the relevant parts of the repository, writes a plan, edits files, runs tests or builds to check its work, and hands back a set of changes, often as a branch or a pull request. You stay in the loop by reviewing what it produced instead of typing every line.

Agents come in a few shapes. Some run in your terminal on your own machine, such as Claude Code. Some run in a cloud sandbox and work on several tasks in parallel, such as OpenAI Codex and Jules. Others are platforms built around a team's tickets and pull requests, such as Devin and Factory, and some live inside an IDE, such as Junie. OpenHands is an open-source option you can host yourself.

Agents do best on tasks with a clear definition of done: a failing test to fix, a well-described bug, a small feature with an example nearby, or a mechanical change repeated across many files. They struggle more with vague goals, large design decisions and code without tests, because they have no reliable way to check themselves.

Good habits make the difference. Write the task as you would for a new teammate, point the agent to the files and commands that matter, and keep instructions for your repository in a file the agent reads every time. Limit what it can reach, especially secrets and production systems, and review its pull requests with the same care as anyone else's.

Featured AI coding tools: coding agents

SponsoredCoding agents This position is open Put your tool at the top of the coding agents page. Paid spots are always labeled Sponsored. From $99 a month. Get featured

What a featured listing costs · from $99 a month

7 AI coding tools in coding agents

Claude Code

by Anthropic

Anthropic's coding agent that works in your terminal, IDE or browser, reading the codebase, editing files and running commands.

OpenAI Codex

by OpenAI

OpenAI's coding agent, available in ChatGPT, a CLI and IDE extensions, that works on tasks in a sandbox and proposes changes.

Devin

by Cognition

Autonomous software engineering agent that takes tickets, works in its own environment and opens pull requests for review.

Jules

by Google

Google's asynchronous coding agent that clones a GitHub repository, works on a task in a cloud VM and proposes a pull request.

OpenHands

by All Hands AI

Open-source platform for software development agents that can write code, run commands and browse, hosted or self-run.

Open source
Can run on your own servers or machine

Factory

Platform of AI agents, called Droids, for coding, review and incident work across a team's tools and repositories.

Junie

by JetBrains

JetBrains' coding agent that plans and carries out tasks inside JetBrains IDEs, running tests and checking its own changes.

Works in JetBrains IDEs

Cost and speed are the other practical questions. Agents run many model calls per task, so usage-based plans can add up, and a task may take minutes rather than seconds. Cloud agents that run several tasks at once suit a queue of small chores, while a terminal agent suits a developer working through one problem interactively.

Start with a narrow trial. Pick a repository with good tests, give the agent a list of small, real issues, and track how many of its pull requests merge with light edits. That number tells you more than any demo, and it shows where your codebase needs better tests or clearer instructions before agents can help much.

Expect the first few weeks to be about learning, for the team as much as the tool. People get better at writing tasks an agent can finish, and the repository gets better instructions and tests along the way.

Questions about coding agents

What kind of tasks suit a coding agent?

Well-defined ones: bug fixes with a reproduction, tests for existing code, dependency upgrades, small features that follow an existing pattern, and repetitive edits across many files. Big design choices still need a person.

Do coding agents push code straight to production?

Not by default. Most agents return a branch, a diff or a pull request for review. Keep it that way with branch protection, and do not give an agent credentials that can deploy.

Where does the agent run?

It depends on the tool. Terminal agents such as Claude Code run on your machine. Cloud agents such as OpenAI Codex and Jules work in a remote sandbox. Self-hosted platforms such as OpenHands run wherever you install them.

How do I give an agent context about my project?

Most agents read a project instructions file in the repository, plus the files you point them to. Describe build and test commands, code style and anything the agent must not touch, and keep that file up to date.

Coding agents