AI Coding Assistants: Cursor vs Claude Code vs Copilot

I’ve spent the last year bouncing between AI coding assistants. Here’s what I learned about each one, without the hype.

Quick Comparison

Cursor

What it is: A fork of VS Code with AI baked in at every level. Not a plugin, but a full editor replacement.

Strengths:

• Composer creates entire features from descriptions
• Tab completion predicts multi-line edits
• Agent mode runs terminal commands automatically
• Project-wide context understanding

Weaknesses:

• Another editor to maintain
• $20/mo adds up if you’re also paying for other AI tools
• Can feel overwhelming with all the features

Pricing:

• Free: Limited features
• Pro: $20/month
• Business: $40/user/month

Claude Code

What it is: A terminal-native coding agent from Anthropic. You run claude in your project directory and talk to it.

Strengths:

• Works in your existing terminal workflow
• Unix philosophy: composable, scriptable
• Can run commands and make commits
• MCP integrations for external tools (Jira, Google Drive)
• No IDE lock-in

Weaknesses:

• Requires Claude subscription
• Terminal-only can be limiting for visual work
• Newer tool, still evolving

Pricing:

• Needs Claude Pro ($20/mo), Max ($100/mo), or API access
• Enterprise plans available

If you’re curious about getting started, see my Claude Code setup guide .

GitHub Copilot

What it is: The original AI coding assistant. A plugin that works in VS Code, JetBrains, Neovim, and more.

Strengths:

• Works in your existing IDE
• Solid inline completions
• Free tier available
• Clean commit messages
• Best ecosystem integration

Weaknesses:

• Multi-file edits can be slow or stuck
• Less project awareness than competitors
• Agent features behind Cursor

Pricing:

• Free: 2,000 completions/month
• Pro: $10/month
• Business: $19/user/month
• Enterprise: $39/user/month

Codex CLI

What it is: OpenAI’s open-source terminal agent. Similar philosophy to Claude Code but uses OpenAI models.

Strengths:

• Open source (Apache 2.0)
• Lightweight
• Scriptable and composable
• Works with any OpenAI-compatible API

Weaknesses:

• Requires API key and costs
• Less polished than commercial options
• Community-driven development

Pricing:

• Free to install
• Pay for API usage (varies by model)

When to Use Each

Choose Cursor if:

• You want everything in one place
• You’re comfortable leaving VS Code
• You work on complex multi-file features often

Choose Claude Code if:

• You live in the terminal
• You want to automate coding workflows
• You already pay for Claude Pro

This fits naturally with terminal-native AI coding approaches.

Choose GitHub Copilot if:

• You want the safest, most proven option
• Your company already pays for it
• You use JetBrains or other non-VS Code editors

Choose Codex CLI if:

• You want open source
• You’re building custom automation
• You prefer OpenAI models

My Setup

I use Claude Code as my primary tool because I spend most of my time in the terminal anyway. When I need visual diffs or complex refactoring, I open Cursor.

Copilot stays installed for quick completions in situations where I don’t need a full agent.

The honest truth: they’re all good enough. Pick the one that fits your workflow and stop switching every week.

What You Can Steal

1. Try the free tiers first. Copilot’s free tier and Cursor’s limited plan let you test before committing.

2. One tool per workflow. Terminal work → Claude Code or Codex. IDE work → Cursor or Copilot. Don’t run both simultaneously.

3. Custom instructions matter. All these tools support project-specific rules (.cursorrules, CLAUDE.md, copilot-instructions.md). Use them.

4. Agent mode is the future. The tools that can plan, execute, and iterate will win. Get comfortable delegating multi-step tasks.

ecosystem tools

tweakcc

tweakcc (1,025 ★) — customize Claude Code system prompts and add AGENTS.md support. no config files, no YAML. just drop markdown files in your repo and Claude reads them.

this is the missing piece for Claude Code. vanilla Claude Code doesn’t support per-project instructions out of the box. tweakcc fixes that. you write an AGENTS.md in your repo, tweakcc makes Claude respect it.

the pattern: configuration as markdown. same philosophy as Backlog.md and the rest of the .md protocol layer. human-readable, agent-parseable, version-controlled.

install:

# check repo for current instructions
# https://github.com/Piebald-AI/tweakcc

first spotted in signals — 2026-02-09 .

→ tool page: tweakcc (coming soon)

awesome-claude-skills

awesome-claude-skills (32,999 ★) — curated directory of Claude Skills, resources, and tools. the ecosystem catalog.

this is where you find what’s possible with Claude. skills for research, coding, automation, analysis. community-maintained, constantly updated.

the value: discovery. Claude can do a lot, but knowing what to ask for is half the battle. this repo is the answer to “what should I build?” and “has someone already solved this?”

browse it when you’re setting up a new project or looking for workflow inspiration.

openai/skills

openai/skills (7,325 ★) — official Skills Catalog for OpenAI Codex. every skill is markdown.

this is OpenAI’s answer to the same problem: how do you make AI coding assistants more capable? modular skills, all documented in .md files.

the format: each skill is a markdown file describing what it does, how to invoke it, and what it returns. agents read these files to understand what’s available.

the pattern: .md files as the skill layer. not JSON schemas, not API docs. markdown.

same philosophy as AGENTS.md, Backlog.md, and the rest of the emerging protocol stack. the format that works for humans also works for agents.

related

• Claude Code Setup Guide — get started with terminal-first AI coding
• AI Coding Tools Comparison 2025 — detailed pricing and features breakdown
• Terminal-Native AI Coding — why CLI approaches win
• AGENTS.md Guide — configure AI instructions per project
• Vibe Coding — the emergent workflow pattern
• Backlog.md — git-native project management for humans and agents