Who this is for
This guide is for developers choosing an agent for an existing project, engineering leads setting guardrails, and teams evaluating a tool before standardising on it.
Decision criteria
- Repository context and multi-file editing
- Terminal, IDE, CI, and cloud workflow fit
- Approval, sandbox, diff, and human-review controls
- Provider choice, self-hosting, and data handling
- Ability to reproduce and test changes outside the product
The shortlist
These tools represent different operating models, from editor and terminal agents to managed cloud workers and self-hosted platforms.
Claude Code
Agentic coding tool from Anthropic that reads your codebase and makes edits across the terminal, IDE, desktop app, or browser.
Strong fit: Excellent reasoning · Strong multi-file edits
Consider: Requires a paid plan
Cursor
AI coding agent and editor built on VS Code, offering codebase chat, multi-file editing, autonomous agents, and integrations across CLI, Slack, and GitHub.
Strong fit: Best multi-file context · VS Code compatible
Consider: Separate app install required
OpenAI Codex
Coding agent for delegating software tasks, exploring repositories, and reviewing implementation work.
Strong fit: Strong task decomposition · Good repository awareness
Consider: Plan availability varies
Windsurf
AI-first editor with an agentic Cascade workflow for understanding and changing entire codebases.
Strong fit: Strong agent workflow · Fast onboarding
Consider: Usage limits on free tier
Cline
Open-source coding agent available as a VS Code extension, CLI, and SDK that can edit files, run commands, and automate workflows.
Strong fit: Open source · Strong user control
Consider: Requires model setup
Roo Code
Open-source autonomous coding agent for VS Code with configurable modes and model providers.
Strong fit: Highly configurable · Open source
Consider: Setup is technical
Devin
Autonomous software engineer for planning, implementing, debugging, and shipping development tasks.
Strong fit: Handles long-running work · Useful for repetitive tickets
Consider: Enterprise-focused
OpenHands
Open platform for cloud coding agents that plans, edits code, runs commands, tests changes, and automates engineering workflows across tools like GitHub, Slack, and Jira.
Strong fit: Powerful agent loop · Self-hosting options
Consider: Setup is technical
How to evaluate an agent
Start with a small, representative issue rather than a toy prompt. Give each candidate the same repository snapshot, ask it to explain its plan, require tests, and inspect the resulting diff. Record setup friction, unnecessary edits, recovery after a failed command, and how easily a teammate can audit the work.
Privacy and control
Cloud agents and editors may process code on vendor infrastructure; the catalog identifies cloud processing or self-hosting where the listed profile does. Check each official policy and workspace setting for your repository before uploading proprietary code. Open source does not by itself mean private, and a local model does not remove the need for secrets hygiene.
Relevant tool profiles
Compare the shortlist
Sources
These are the official URLs listed in the directory profiles. Confirm changing prices and policies there.
Frequently asked questions
Are coding agents safe to run on production repositories?
Treat an agent as an untrusted collaborator until reviewed. Use a branch or disposable checkout, least-privilege credentials, approval gates, and tests; never grant production secrets merely because a tool can run a command.
Which agent should a solo developer try first?
An editor-based agent can reduce setup friction, while a terminal agent is a better fit for developers who already work from the shell. Compare both against one real maintenance task before committing.
Do free labels mean the whole workflow is free?
No. The catalog labels the product's current pricing model, but model usage, hosted automation, or provider accounts may be separate. Verify changing terms on the official page.