Coding Tools & CLI Agents
Table of Contents
Two families here: IDE assistants (autocomplete + chat + in-editor agents living in your editor) and CLI/terminal agents (you talk to an agent in the shell. It reads, edits, runs, and tests across your repo). The 2025→2026 shift was from "smarter autocomplete" to "agentic". The tool plans, edits many files, runs commands, and iterates. Pricing has largely moved to usage/credits on top of a seat, so watch token burn.
IDE assistants
| Tool | Best for | Watch out | Price |
|---|---|---|---|
| Cursor | Whole-repo agentic edits (Composer), fast multi-file changes, predictive tab-completion | Agent runs burn credits; the pricing model has changed more than once | Free · Pro ~$20 · Business/usage 🔄 |
| GitHub Copilot | Ubiquity (VS Code, JetBrains, Neovim, Visual Studio); agent mode; choose Anthropic/OpenAI/Google models | Historically more conservative than Cursor; premium requests metered | Free · Pro ~$10 · Pro+ ~$39 · Business ~$19/user |
| Windsurf | Its Cascade agent flow and clean, agent-driven UX | 2025 ownership saga (now Cognition); watch roadmap/pricing continuity | Free · paid ~$15+/mo, credit-based 🔄 |
| Zed | Speed (Rust), multiplayer editing, native agent panel with external agents via ACP | Leaner extension ecosystem; newer AI layer | Open source; AI free with BYO key + hosted plan 🔄 |
Also: JetBrains AI Assistant / Junie, Kiro (AWS spec-driven IDE), Trae (ByteDance).
CLI / terminal agents
These live in your shell, see the whole repo, and can run commands. Increasingly the primary interface for serious agentic coding.
| Tool | Best for | Watch out | Price / open |
|---|---|---|---|
| Claude Code | Deep multi-step work across big codebases (plan, edit, test, git); subagents, hooks, skills, MCP. A de-facto reference harness | Will run commands and burn tokens; use permission modes and review | Claude Pro/Max or API; CLI free to install |
| OpenAI Codex CLI | Local agentic coding on OpenAI models, with local-to-cloud handoff | "Codex" spans a CLI, IDE extension, and cloud agent, so know which you're using | Open source (Apache-2.0); ChatGPT plan or API |
| Aider | Tight, git-native editing loops (auto-commit each change); model-agnostic, scriptable | A precise pair-programmer more than an autonomous agent; bare-bones UI | Free (MIT); pay only model API |
| Gemini CLI | Generous free Gemini access, large context, web/Google tools; MCP + ACP | Newer than Claude Code/Aider; tracks whichever Gemini you're allotted | Free tier via Google account; open source |
| Amp | High-autonomy "let the agent cook" runs, subagents/threads, team sharing (Sourcegraph) | Token-metered; can get pricey on big autonomous runs | Usage-based credits, free to start 🔄 |
Also: OpenCode, Cline, Charm Crush (all open source).
PR / review & background bots
| Tool | Best for | Link |
|---|---|---|
| GitHub Copilot code review | Assign an issue and it opens a PR, or reviews yours inline, inside GitHub's flow | copilot |
| Cursor Bugbot | Automated PR bug and security review with fast turnaround; can auto-review agent-written PRs | cursor.com/bugbot |
| CodeRabbit | Line-by-line PR feedback and summaries; free for open source | coderabbit.ai |
| Greptile | PR review that reasons over your whole repo graph | greptile.com |
| Devin (Cognition) | An autonomous engineer you assign well-scoped tickets; runs in its own cloud env (supervise) | devin.ai |
| Graphite Diamond / Qodo | Review + test generation | graphite · qodo |
Rule of thumb: IDE assistant for flow (you're editing, it helps), CLI agent for delegation (you describe, it executes across the repo), PR bot for gatekeeping (it reviews what humans and agents produced).