Two Hacker News threads dominated developer Slack channels this week. Cursor Camp dropped — their structured programme for getting teams productive with AI coding workflows. Within 48 hours, a separate thread about Claude Code's billing behaviour hit 1,300 points and 400+ comments, full of developers discovering charges they hadn't anticipated. Both tools are genuinely capable. Both have real frustrations. Here's the task-focused breakdown that actually matters for frontend work.
Codebase Awareness: Claude's Context Window Changes Everything
Cursor works inside VS Code. Its codebase indexing uses embeddings to pull relevant files into context when you make a request. For most day-to-day tasks this is perfectly fine. Ask it to refactor a component and it finds the file, understands the imports, makes sensible edits.
Where it breaks down is cross-cutting changes. Renaming a shared hook used across 40 files, migrating from one state management library to another, auditing every API call for a new auth pattern — Cursor's context retrieval becomes the bottleneck. It retrieves the most semantically similar files, not necessarily all the relevant ones. You end up babysitting the navigation. This took me longer to internalise than it should have, because the tool is so smooth for single-file work that the multi-file limitation only surfaces when you're deep into a large task.
Claude Code takes a different approach entirely. It's a CLI agent — not an IDE — that uses tool calls to read files on demand. When you give it a task, it plans, reads what it needs, then executes. There's no embedding retrieval guessing which files matter; it navigates the filesystem directly.
# Claude Code on a real migration task
claude "migrate all React class components in src/components to functional components with hooks. Preserve all existing prop types and add missing TypeScript annotations."It recursively finds class components, understands their lifecycle methods, and rewrites each one with appropriate hooks. Coverage is better than anything Cursor manages on the same task without manual file-by-file direction. For single-file work, the gap is negligible. For anything that spans the repo, Claude Code wins clearly.
Multi-File Edits: Where Cursor Still Has the UX Edge
That said, Cursor's VS Code integration is genuinely valuable. You see diffs inline, accept or reject changes per-file, and stay in the flow you already know. There's no context switching between a terminal and your editor.
Claude Code's terminal-first model is more powerful but more opaque whilst it's working. You submit a task, it runs autonomously, and then you review the result. For large feature-scale tasks this is exactly what you want. For small iterative edits — tweaking a component's prop types, adding an aria-label, adjusting a Tailwind class — the round-trip friction accumulates quickly.
Cursor's Cmd+K for inline edits with immediate visual diff feedback is still the best experience for micro-edits. If your day is mostly focused, small changes, that IDE integration is worth a lot. Claude Code rewards patience and batch thinking. Submit a well-described task, review the output, iterate once or twice. If you approach it like pair programming with instant feedback, you'll be disappointed.
Pricing: The Conversation Nobody's Having Honestly
This is what that HN thread was actually about, and it's worth being direct.
Cursor is $20/month on Pro, $40 for Business with more fast model requests. Predictable. You know exactly what you're spending before the month starts.
Claude Code uses Anthropic's API with usage-based billing. A light week might cost $12–15. A heavy week running agentic tasks — migrating a codebase, writing comprehensive test suites, multi-file refactors across a large project — can land anywhere from $40 to over $100. The developers in that HN thread weren't being careless; they were using Claude Code as intended, for autonomous multi-step tasks, and their bills reflected it. The docs don't make this obvious enough until the first invoice arrives.
Anthropic has since added better usage dashboards and configurable spend limits. Use them immediately. Set a monthly cap on day one. Without that guardrail, an overnight autonomous run on a complex task can consume more than a month's Cursor subscription in a single session.
For solo developers on a budget: Cursor's flat rate is easier to manage. For a team where Claude Code's agentic capabilities save multiple hours of developer time per week, the ROI calculation shifts quickly — but you need to track it honestly rather than assuming the savings are there.
Which One Ships Frontend Features Faster?
I ran the same feature set through both tools on a mid-size React + TypeScript project: a notification system with a backend API integration, toast components, user preference storage, and Playwright E2E tests.
Cursor (with Claude 3.7 Sonnet backend):
The component work was fast and the inline diff experience for the toast UI was genuinely pleasant. Struggled with the end-to-end flow — I had to manually direct it to each file in sequence: API client, then hook, then component, then tests. Test generation caught the happy path but missed error states I had to add manually. Total time including review: around 3.5 hours.
Claude Code:
One well-structured prompt and it planned the entire feature. It read the existing API client, matched the patterns already established in the codebase, wrote the integration without being pointed at it. Tests were more thorough because it read the existing test suite first and matched conventions. The code wasn't flawless — it made one incorrect assumption about an auth header format — but the coverage was significantly better. Total time: roughly 2 hours, of which maybe 20 minutes was reviewing and correcting generated code.
The caveat matters: Claude Code rewards prompt quality. A vague instruction gets vague results. Cursor is more forgiving because you're collaborating in real time and can steer it mid-task. If you're not used to writing precise task descriptions upfront, there's a genuine learning curve with Claude Code before you see the speed benefit.
If your team already runs a fast local dev setup — Vite or one of the newer bundlers with sub-second HMR — Claude Code's terminal-first workflow pairs naturally with it. You're already living in the terminal for your dev server, so the context switch is minimal. Cursor makes more sense if you're firmly IDE-native and that's not changing.
The Verdict
Use Claude Code if you do feature-scale agentic work, your tasks span multiple files, you're comfortable with CLI tools, and you're prepared to manage usage-based billing with a spend limit configured. It's the more capable tool for complex frontend work and genuinely saves hours on large tasks.
Use Cursor if you want IDE integration, predictable monthly costs, and a lower floor for useful output from short prompts. It's the better choice for developers who work in small focused increments throughout the day and want AI assistance that stays out of the way.
Choosing just one: if you work on anything beyond simple component edits, Claude Code is the better long-term investment. Set the spend limit, spend a week learning to write detailed task descriptions, and the capability gap over Cursor becomes obvious. If budget predictability matters more than raw capability, Cursor at $20/month is a reasonable trade.
One adjacent consideration: if the VS Code dependency is your main reason for leaning towards Cursor, the Zed 1.0 review is worth reading. Zed's baked-in AI features are improving fast, and for some workflows it reduces the need for a dedicated AI coding agent layer at all.
Whatever tool you ship with, get your preview deployments sorted. Vercel is still the fastest way to get per-branch preview URLs — when you're reviewing AI-generated features before they hit main, that feedback loop matters more than ever.