Skip to content
10 toolsPublic bookmarkCurated by Bookmarkit TeamUpdated September 2026

AI coding agents to shortlist

An AI coding agent reads a repository, edits files, runs commands, and opens pull requests. This shortlist is for teams choosing a daily driver after the September 2026 model wave, not for people who only want a chat sidebar.

What is an AI coding agent?

An AI coding agent is software that can search a codebase, change more than one file, run tests or commands, and come back with a diff or a pull request. A chat model that only pastes snippets is not an agent. An IDE with tab-complete is not enough either. You judge these products on repo awareness, tool allowlists, logs, and whether you can stop a run.

This shortlist includes Claude Code, Cursor, Google Antigravity, OpenAI Codex, Devin Desktop, Devin, and 4 more.

Who this shortlist is for

  • Engineering managers comparing a Claude, OpenAI, or Google-native agent
  • Teams leaving Windsurf and needing the Devin Desktop name on the page
  • Platform owners who need a terminal agent, an IDE, or a cloud worker, and who refuse to treat those as the same buy

How we picked

Bookmarkit added Claude Code, Google Antigravity, OpenAI Codex, OpenCode, and Devin Desktop as first-class profiles in September 2026. Cursor, GitHub Copilot, Devin, and Aider were already in the catalog. We leave out prompt-to-app builders such as Lovable unless the job is a greenfield prototype.

  1. Can edit a real repository, not only generate a snippet
  2. Has a documented surface: terminal, IDE, desktop, or cloud worker
  3. Public pricing, a ChatGPT/Claude/Gemini seat, or an open-source install path
  4. A Bookmarkit profile you can save on a shortlist

The shortlist

Open a profile for pricing, alternatives, and a product page.

Cursor logo
Cursor4.4(120)0

AI-native code editor that pairs large language models with your codebase.

ProprietaryDeveloper ToolsWebFreemium · See website
Devin logo
Devin4.4(120)0

Cognition operates Devin, the first autonomous software engineer.

ProprietaryAI AgentsWebFreemium · See website
Aider logo
Aider4.4(120)0

Aider lets you pair program with LLMs to start a new project or build on your existing codebase.

ProprietaryAI AgentsWebFreemium · See website
Cursor Agent logo
Cursor Agent4.4(120)0

This element contains an interactive demo for sighted users showing multiple Cursor interfaces: Cursor Desktop interface, Cursor.

ProprietaryAI AgentsWebFreemium · See website

How to choose a coding agent

Decide the surface first. Terminal agents meet you in git. IDEs keep the editor. Cloud workers run while you are away. Mixing two is normal. Buying three without a log and a kill switch is not.

If this is the jobLook at
You live in the terminalClaude Code, OpenCode, or Aider. Confirm how they call git, tests, and MCP tools.
You want an editor plus an agent managerCursor, Devin Desktop, or Google Antigravity.
The org already pays OpenAI or GitHubOpenAI Codex or GitHub Copilot before you add a second seat tax.
The job is unattended repo workDevin Cloud (the Devin profile), not only Devin Desktop.

Related research

Frequently asked questions

What is an AI coding agent?

It is software that can read a repo, change files, run commands, and open a pull request. Claude Code, Cursor, Codex, Antigravity, and Devin Desktop all claim that loop. A chat window that only suggests code is not the same product.

Is Devin Desktop the same as Windsurf?

Yes. Cognition renamed Windsurf to Devin Desktop on 2 June 2026 and kept existing plans. Use the Devin Desktop profile for the IDE. Use the Devin profile for the cloud worker.

Should I pick Claude Code or Cursor?

Pick Claude Code if the team already pays for Claude and wants a terminal-first agent that also plugs into VS Code and JetBrains. Pick Cursor if the editor is the product and you want model choice inside one IDE. Many teams run both for a quarter and then drop one.

Where does Google Antigravity fit?

Antigravity is Google’s agent-first IDE, CLI, and manager. Google pointed Gemini 3.8 Flash developers at it in September 2026. Shortlist it when the org is already on Gemini or Gemini Enterprise.

What should I ask before connecting an agent to git?

Ask where diffs are logged, whether writes need approval, which MCP servers it can reach, and how you stop a run if the process hangs. Then run the same messy ticket on two vendors. Do not score the demo.