The best coding agents, according to hackers
41 coding agents recommended in Hacker News comments over the past year, ranked by how many separate people recommend each one.
- Most recommendedGitHub Copilot571 people recommend it
- Least regretAutolith14 recommend, 0 criticize
- Free and lovedoh-my-pi104 people recommend it
All coding agents
- GitHub CopilotRecommended as a cheap way to use many AI models with smooth VS Code integration and autocompletion, though critics resent tightening request limits, pricing changes and ads in PRs.571recommend
- oh-my-piA batteries-included fork of the Pi coding harness with sensible defaults, easy provider switching and built-in search and context tools, though critics find it bloated and token-hungry.104recommend
- Grok Build CLIFast, smooth terminal coding agent that people find capable and recommend; critics say it uploads whole directories, including secrets, to xAI.66recommend
- DeepSeek HarnessRecommended as an open-source agent harness with good observability, a plugin system and a desktop app. Critics mostly question its plugin design and worry about data collection.63recommend
- Claude Code VS Code extensionBrings Claude Code into VS Code for reviewing and editing code in a GUI. Main caveat: a lookalike extension on the Cursor marketplace reportedly contained malware.31recommend
- Antigravity CLIRecommended as a fast, responsive step up from Gemini CLI, with many praising its speed. Critics point to low usage quotas and login problems.31recommend
- MakiRecommended as a fast, lightweight Rust terminal coding agent that runs well on small servers and parses shell permissions; the most common caveat is its curl-to-shell install script.30recommend
- Hax coding agentA tiny, MIT-licensed terminal coding agent in C with minimal tools, favored over bloated harnesses. Critics note weak context handling and slow-provider behavior.28recommend
- Roo CodeFree, open-source VS Code extension for AI coding agents, liked for agent modes, customisation and task delegation; critics say results depend heavily on the model and default prompts.26recommend
- Qwen CodeRecommended as a free-tier coding agent that works well with Qwen models and uses fewer tokens than Claude Code; the common caveat is failed tool calls.25recommend
- Kimi CodePeople recommend it as a capable, cheap coding agent that rivals Claude Opus in planning, though critics say usage limits run out fast.25recommend
- ZerostackSmall, fast Rust coding agent with a minimal design, low memory use and built-in tools. Critics say it crashes without stacktraces and adds little beyond prompts.25recommend
- MiMo CodeRecommended as a free open-source coding agent from Xiaomi's MiMo models, praised for its UI, speed and no-account use. Common caveat: critics dispute whether the subscription beats API pricing.24recommend
- Google JulesPeople like its isolated cloud runs that you review as pull requests and steer from a phone; the common caveat is weak output quality.23recommend
- Augment CodeRecommended for a context engine that understands large codebases and finds where changes belong; the most common caveat is the move to per-token pricing.20recommend
- fx coding agentFree, minimal, embeddable Zig coding agent harness with a built-in CLI, repo map and JSON viewer; critics say it requires Vercel AI Gateway and feels unfinished.19recommend
- ContinueOpen-source AI coding assistant people recommend for its configurable checks, autocomplete, and local model support with Ollama. Common caveat: many reports of buggy plugin behavior in VS Code and JetBrains.18recommend
- GitHub Copilot coding agentPeople recommend it for running coding tasks in an isolated container and reviewing results in one PR. Main caveat: critics say it changes silently and edits PRs unprompted.16recommend
- Bullet coding agentRecommended for fast coding-agent execution that gets projects started quickly. Critics object to default chat sharing, self-attribution in commits and hidden code, and unverified benchmarks.15recommend
- JetBrains AIRecommended for its integration with Claude Code and other agent tools and for switching models inside JetBrains IDEs. Critics say suggestions and output are often inaccurate.14recommend
- AutolithA free coding agent built in Common Lisp that runs inside a live Lisp image, letting the agent inspect and modify running code and its own source.14recommend
- P0 AI coding harnessPeople recommend P0 for running feature development as a structured, autonomous pipeline that ships working code into real codebases. The most common caveat is its price, which one commenter called too high for hobbyists.14recommend
- HopliteCloud coding agents with per-thread VMs for isolation, no markup on token costs; $99 per seat monthly is steep for personal use.13recommend
- ShelleyPraised as a capable web-based coding agent that runs in a Linux VM, can use a browser to test sites, and keeps searchable conversation history. Critics mention it is hard to get running.12recommend
- Cursor Cloud AgentsPraised for easy repository automation and a pleasant experience; the main caveat is cost, and it uses Cursor's own harness.12recommend
- Google agy CLIRecommended as a generous, capable coding agent harness with skills and sub-agents. Critics say it asks for too many permissions and handles complex problems poorly.12recommend
- DiracOpen-source coding agent people recommend for token efficiency and fast refactoring on large codebases. Common caveat: it still feels rough, with bugs and missing integrations.11recommend
- GitHub Copilot ChatPeople recommend it for in-editor agentic coding in VS Code, with good results on whole-file generation. Critics say it lags in large sessions and feels worse after recent updates.10recommend
- Zed AgentRecommended for live pair programming and self-hosted models; critics say it breaks on Windows with WSL or SSH remotes.10recommend
- MystiPeople recommend it for running Claude, Codex and Gemini in one interface, where they can brainstorm or debate code together. Critics note its earlier BSL license and no published pricing.10recommend
- BroccoliOpen-source coding agent that runs in your own cloud with no third-party control plane, and multiple rounds of review on its output.10recommend
- Gemini Code AssistPeople recommend it as a free, decent AI coding assistant that finds real errors. The most common caveat is confusing naming and weak tool calling.9recommend
- mini-swe-agentMinimal coding agent running an LLM in a loop with only terminal access; praised for simplicity. Critics note it now exceeds its original 100 lines.9recommend
- Kimi CLIFree open-source coding agent CLI that people pair with Kimi models and find pretty good. The most common caveat is that it leaks memory.9recommend
- GitHub Copilot in Visual StudioRecommended for an agent that runs code, reads the errors and keeps going, plus fast autocomplete. Critics say suggestions are hit or miss and the integration is buggy.8recommend
- AI2 Open Coding AgentsRecommended as an open coding agent with its training pipeline available and reproducible for about $400; critics say only part of the training data is open.8recommend
- OpenHandsFree, MIT-licensed open-source AI coding agent that works with any model. Critics call its container-per-agent design heavy.6recommend
- Juno by InfiniaXPeople recommend Juno, also published as juno-code, for fixing LLM formatting errors and for speed, and one commenter calls it a clear fix for known problems.6recommend
- 3codeA free, cross-platform coding agent written in Nim and built for token efficiency, which people say uses fewer tokens than Claude Code.5recommend
- IBM BobRecommended for enterprise cost control and IBM-specific tooling. Critics say it is a VS Code fork with weak security defaults and is not production-ready.5recommend
- OpenGravityZero-install vanilla JS clone of Antigravity that uses your own API key, with local-model runs praised for speed and memory use.5recommend