CursorってClaude Codeより何が強いの？中の人に聞いてみたA Claude Code power user tried Cursor for the first time and built a…

A developer who runs nearly all of their day-to-day coding through Anthropic's Claude Code recently sat down with Cursor for the first time, and the resulting write-up is worth reading precisely because it comes from a committed user of a rival tool. Published on the Japanese engineering platform Zenn, the post pairs a hands-on first impression with an interview with someone on the inside at Cursor, aiming to pin down where the editor has an edge.

The standout moment is the demo itself. According to the author, three rounds of prompting were enough to produce a browser-based 3D shooter in the mold of Star Fox, the on-rails classic in which the player flies into the screen while firing at incoming enemies. A linked video shows the game running in a browser. The author's candid takeaway was disbelief that a first session, with no prior familiarity with the tool, could reach a playable result that fast.

For readers unfamiliar with the two products, some context helps. Cursor, built by Anysphere, is an AI-first code editor forked from Visual Studio Code. It folds chat, inline edits, tab-style completions, and an autonomous agent mode into a familiar graphical editor, and it can route requests to a range of underlying large language models, including those from Anthropic and OpenAI. Claude Code, by contrast, is Anthropic's terminal-based coding agent. It lives in the command line, operates directly against a project's files, and leans on Claude models to plan and execute multi-step changes.

That difference in form factor is part of what the comparison is really about. Claude Code appeals to developers who are comfortable in a terminal and want an agent that reads, edits, and runs code with minimal ceremony. Cursor keeps the developer inside a full editor, where generated code, the file tree, and a running preview can sit side by side. For a task like building a small game and immediately seeing it move in a browser, that visual, tightly integrated loop appears to be a meaningful advantage, though a single anecdote is not a benchmark.

The author frames the piece as an attempt to understand what specifically makes Cursor strong, and the inclusion of an interview with a Cursor insider is meant to move beyond first impressions. The write-up is a personal account rather than a controlled evaluation, so its conclusions are best read as one experienced practitioner's perspective rather than a definitive verdict.

The broader backdrop is a fast-moving market for AI coding tools. Over the past couple of years, agentic coding, where the model does not just autocomplete but plans, edits multiple files, runs commands, and iterates, has become the main battleground. GitHub Copilot added agent-style features, Anthropic shipped Claude Code, OpenAI has pushed its own coding agents, and a wave of editors and command-line tools compete on how well they orchestrate models against real repositories. Cursor has been among the most visible in this space, and its rapid adoption has