What is a background (autonomous) coding agent?

An agent you hand a task to and leave alone. It plans, edits, runs tests, and opens a pull request. No human sits in the loop between steps. You review the result rather than each action. It differs from an IDE autocomplete or chat assistant, which work with you turn by turn.

Do these open a pull request or commit directly?

The ones here open a pull request for review by default. That is the safer pattern. You see the diff, the tests, and the reasoning before anything merges.

Which autonomous coding agents are open source?

OpenHands and SWE-agent are MIT-licensed. You can self-host both. The rest listed are hosted services.

Top Autonomous & Background Coding Agents (2026)

An autonomous or background coding agent takes a task, such as an issue or a spec. It works without you in the loop between steps. It usually opens a pull request for review. Some run in the cloud. Some you self-host. This list covers the agents built for that async, hands-off flow.

1
Devin
Cognition's hosted autonomous "AI software engineer." It takes tasks and ships pull requests. It reacts to review feedback and CI.
Model:
Hosted SaaS
Runs:
Cloud + desktop + CLI
2
OpenAI Codex (cloud)
OpenAI's cloud software-engineering agent. It runs coding tasks in parallel, each in its own sandbox, and opens pull requests. Also available via CLI, IDE, and ChatGPT.
Model:
Hosted (OpenAI)
Runs:
Cloud sandboxes; also CLI + IDE
3
GitHub Copilot coding agent
An async agent you assign issues to. It works in a secure GitHub Actions cloud environment. It opens a pull request for review.
Model:
Hosted (GitHub)
Runs:
GitHub Actions cloud env; opens PRs
4
Google Jules
An async coding agent powered by Google's Gemini models. It clones your GitHub repo to a cloud VM. It plans, then opens pull requests.
Model:
Hosted (Google)
Runs:
Cloud VM; opens GitHub PRs
5
Factory
An agent platform whose "Droids" handle engineering tasks on their own. You delegate a task and review the diff. Then you merge or open a PR from the app or terminal.
Model:
Hosted SaaS
Runs:
Desktop app + CLI + web
6
OpenHands
An open-source (MIT) control center you host yourself. Its coding agents automate dev tasks. It runs locally, in Docker, on a VM, or in OpenHands Cloud. Formerly OpenDevin.
License:
MIT
Runs:
Local / Docker / VM / cloud
7
SWE-agent
An open-source (MIT) research agent framework from Princeton and Stanford. It gives a language model of your choice the tools to resolve GitHub issues on its own. Self-hosted. Works with any supported model.
License:
MIT
Runs:
Self-hosted / CLI, any supported LM
8
Syndai
Hosted coding-as-a-service. You submit a plain-language spec. You get a pull request in your repo. The work runs in an isolated sandbox. Test, build, and validation artifacts come attached. Tagline: "Prompt in. Pull request out. Evidence attached."
Model:
Hosted service
Runs:
Isolated sandbox; opens a PR with evidence

Reviewed quarterly. This is the fastest-moving list here. Rebrands and buyouts are frequent. So surfaces and framing are the most likely to drift. Durable facts only. Each row carries its last-checked date and source.

Common questions

What is a background (autonomous) coding agent?: An agent you hand a task to and leave alone. It plans, edits, runs tests, and opens a pull request. No human sits in the loop between steps. You review the result rather than each action. It differs from an IDE autocomplete or chat assistant, which work with you turn by turn.
Do these open a pull request or commit directly?: The ones here open a pull request for review by default. That is the safer pattern. You see the diff, the tests, and the reasoning before anything merges.
Which autonomous coding agents are open source?: OpenHands and SWE-agent are MIT-licensed. You can self-host both. The rest listed are hosted services.