Tools·Tools
Autonomous coding agents: CLI and cloud
Terminal-first and cloud agents that run on repos, issues, and PRs with minimal human-in-the-loop.
This article is AI-assisted and co-authored by Xesca Alabart, co-founder of EasySpecs.
Autonomous coding agents optimize for long-horizon work: subagents, hooks, sandboxes, and issue → PR flows. They are closest to a software factory spine when paired with a validation harness—otherwise you get fast, plausible diffs without proof.
Tools in this category
| Tool | Notes |
|---|---|
| Claude Code | Terminal-first; subagents, hooks, large context; widely adopted. |
| OpenAI Codex | Cloud sandbox, desktop multi-agent app; strong benchmark positioning. |
| Devin 2.0 (Cognition) | Cloud autonomous agent, planning, wiki. |
| OpenHands | Open source, Docker sandbox, model-agnostic, multi-agent. |
| SWE-Agent | Academic ACI focus—how LLMs use computer interfaces. |
| Aider | CLI; repository map (tree-sitter + PageRank) influenced the field. |
| Factory.ai | Issue assignment → PR automation; SDLC traceability story. |
Factory takeaway
These tools consume whatever context you put in the repo. Seed manufacturing and requirements gathering remain upstream—agents do not remove the need for explicit intent.
Related
References
- Claude Code — Anthropic
- OpenAI Codex — Cloud agent
- Devin — Cognition
- OpenHands — Open source
- SWE-Agent — ACI
- Aider — CLI + repo map
- Factory.ai — Droids