What they have in common
Cursor Cloud Agents (formerly Background Agents) [1], Claude Code on the web [4], OpenAI Codex cloud [5] and GitHub Copilot cloud agent — which GitHub previously called the coding agent [7] — share a shape. Each runs in an isolated cloud environment with a copy of your repository, works on its own branch, and hands back changes for you to review. The differences are in where you start them, how they bill, and how much control you get over the environment.
Last checked: October 3, 2026 · these products change monthly
The comparison table
Every cell links to the vendor's own documentation. Last checked October 3, 2026.
| Cursor Cloud Agents | Claude Code on the web | OpenAI Codex cloud | GitHub Copilot cloud agent | |
|---|---|---|---|---|
| Where it runs | Its own sandboxed VM per agent [1] | A cloud environment per session, with its own network and setup settings [4] | An OpenAI cloud environment with your repos and tools [5] | An ephemeral environment powered by GitHub Actions [7] |
| Ways to start it | Cursor Desktop, web, iOS, Slack, GitHub/Bitbucket, Linear, API [1] | Browser, the Claude mobile app, the Desktop app (select Cloud), Slack, the terminal (claude --cloud) and routines [4] | Web, the desktop app, mobile and the Codex CLI [5] | GitHub issues, Copilot Chat, Slack and Teams [7] |
| Issue trackers | Linear (assign to @Cursor), GitHub issues [8] | Not documented on the cloud page [4] | GitHub, GitLab (beta), Linear integrations listed in docs [5] | GitHub issues; Linear, Jira and Azure Boards integrations listed in docs [7] |
| Parallel runs | "As many agents as you want in parallel" [1] | Each --cloud command is its own session; share your account's rate limits [4] | Within your ChatGPT plan's usage limits [6] | Each assigned task runs as its own session, using Actions minutes and AI credits [7] |
| What it hands back | A PR with screenshots, videos and logs [3] | A branch / PR; can auto-fix CI failures and review comments [4] | Results to review; you commit or open a PR [5] | Changes on a branch; optionally a PR [7] |
| How it's billed | "API pricing for the selected model" [1] | Included in Pro, Max, Team and eligible Enterprise plans; shares your plan's rate limits; "no separate compute charge for the cloud VM" [4] | Included in ChatGPT plans, shared usage limits [6] | GitHub Actions minutes plus AI credits [7] |
| Environment setup | Per-environment repos, secrets and network access [1] | Network access, environment variables, setup scripts [4] | Cloud environments, internet access, env vars and secrets [5] | Runs your tests and linters in its Actions-powered environment [7] |
What the research says
The best public evidence so far is a study of 7,156 pull requests from five agents — OpenAI Codex, GitHub Copilot, Devin, Cursor and Claude Code — in the AIDev dataset, first published February 2026. [9] Its headline finding is that task type matters more than the agent: documentation tasks were accepted 82.1% of the time versus 66.1% for new features. Codex had consistently high acceptance across task categories (59.6%–88.6%), "however, no single agent performs best across all task types: Claude Code leads in documentation (92.3%) and features (72.6%), while Cursor excels in fix tasks (80.4%)." [9]
Treat these as a snapshot. Acceptance also depends on who uses each agent and for what, and every one of these products has shipped new models since the data was collected.
How to choose
By where your team works
In Cursor already? Cursor Cloud Agents. In the terminal with Claude? Claude Code on the web. In ChatGPT? Codex. Living in GitHub issues? Copilot cloud agent.
By billing model
Per-token API pricing (Cursor), included in a chat subscription with shared limits (Claude, Codex), or Actions minutes plus credits on a GitHub plan (Copilot).
By environment control
How much you need to configure — setup scripts, secrets, network access, multiple repos — and who on the team will maintain it.
By issue tracker
If your team plans in Linear, check how each agent picks up work from it — see the table.
Mixing tools
Mixing is fine, but each tool has its own config — rules files, hooks, environment setup. Keep the shared parts in the repository, not in any one tool: tests, pre-commit checks and an AGENTS.md everyone reads. Then a new tool inherits your standards on day one. More on that in our guide to instant feedback for coding agents.
Where Aerunit fits
To be clear about where we stand: Aerunit dispatches to Cursor Cloud Agents today, and only to Cursor Cloud Agents. If one of the other tools fits your team better, use it.
If you pick Cursor, Aerunit adds a layer on top. It writes the stories, reading every repository a story touches. It follows Linear's blocks / blocked-by relations, queuing the next issue when its blocker's PR merges. It reviews every agent PR and can send findings back to the session. And it shows every running session in one list, each linked to its Linear issue. More on that in our Cursor background agents guide.
Cursor today
Aerunit works with Cursor Cloud Agents only.
On top of Cursor
Stories, dependency-aware dispatch and automatic PR review.
Pick by workflow
Choose the agent that fits where your work already lives — tracker, repo host, chat.
Common questions
Which cloud coding agent is best?
None wins everywhere. A study of 7,156 agent pull requests found task type mattered more than the agent, and that no single agent performed best across all task types. Pick by where your team already works, how you want to pay, and how much control you need over the environment. [9]
Can I use more than one?
Yes. They all work through git and pull requests. Keep the shared parts — tests, pre-commit checks, AGENTS.md — in the repository so every tool follows the same rules.
Does Aerunit work with all of these?
No. Aerunit currently works with Cursor Cloud Agents only. The others are included here so you can make a fair choice.
Sources
- [1]Cursor Docs — Cloud Agents
- [2]Cursor Docs — Automations
- [3]Cursor Docs — Cloud Agent capabilities
- [4]Claude Code Docs — Claude Code on the web
- [5]OpenAI Developers — Codex Cloud
- [6]OpenAI Developers — Codex pricing
- [7]GitHub Docs — About Copilot cloud agent
- [8]Linear Changelog — Cursor background agents
- [9]Pinna et al. — Comparing AI Coding Agents: A Task-Stratified Analysis of Pull Request Acceptance
