Claude Code vs Codex CLI vs Cursor: An Honest Field Guide for Working Developers

Claude Code, Codex CLI, and Cursor are three AI coding assistants that work differently and cater to distinct workflows. Claude Code is a terminal assistant with memory capabilities, while Codex CLI separates what the AI can touch from when it must seek approval. Cursor, on the other hand, integrates AI into the IDE, offering context-aware editing shortcuts.

Claude Code's defining idea is that the AI should remember your project. It keeps project memory in CLAUDE.md, personal preferences in ~/.claude/CLAUDE.md, and local-only notes in CLAUDE.local.md. AGENTS.md is another file it reads, allowing multiple conventions to coexist. The two key habits for Claude Code are writing verifiable instructions in memory files and starting big tasks in plan mode. This allows the AI to suggest changes before execution, reducing the time spent on costly mistakes.

Codex CLI separates the questions of what the agent can touch and when it needs approval. It has three sandboxes - read-only, workspace-write, and danger-full-access - and an approval policy that defaults to on-request. Its superpower lies in automation, with commands like codex exec fix fixing failing tests non-interactively. However, beware of the dangerously flagged flags, as they should only be used in isolated environments. Codex CLI is best for scripted and automated agent runs in CI and script environments.

Cursor's core pitch involves three main shortcuts: editing selected code inline, opening a chat with selection context, and handing the wheel to the Agent for multi-file rewrites. Context control is explicit, allowing AI to operate on files, codebase, docs, web, Git, or individual files. Project rules live in .cursor/rules/ as one .mdc file per rule, with attachment modes like Always, Auto Attached, Agent Requested, and Manual.

Three postures define the Agent - Ask for read-only exploration, Agent for editing with confirmation, and Plan for proposal-based actions. Cursor excels in day-to-day editing within the IDE environment.

When working with these tools, remember to keep your rules, secrets, and workflows consistent. Use plan mode for anything beyond minor fixes, write machine-readable rules, and avoid pasting sensitive information into the chat. These habits will make your interactions with Claude Code, Codex CLI, and Cursor more efficient and effective.