Cursor 1.0

Original Article ↗ Hacker News Discussion ↗

Tool landscape and comparisons

  • Commenters note the ecosystem is crowded: Cursor, Claude Code, Cline, Roo Code, Aider, Copilot, Zed agent, JetBrains Junie, Windsurf, Emacs+gptel, AugmentCode, Ampcode, etc.
  • Benchmarks like liveswebench exist but are seen as incomplete; differences in workflow, model choice, context size, MCP support, and UI matter more than a single score.

Cursor vs other AI coding tools

  • Strong praise for Cursor’s tab completion/“next edit”; many say it’s the best autocomplete they’ve used and a major reason to stay, even if they rarely use agents.
  • Others find Cursor’s agent weaker than Claude Code, Cline, Roo, or Aider: reports of wrong tool calls, premature stopping, messy diffs, and issues on large codebases.
  • Claude Code is widely praised as a smarter, more capable agent (good at using CLI/SSH, grokking big repos), but burns tokens quickly; many end up on expensive Max plans.
  • Aider is liked for tight git integration and control (micro‑commits, undo per prompt, custom rules). Cline/Roo are praised for pure agent workflows but can be very costly with reasoning models.
  • JetBrains Junie and Zed’s agent are seen as “good and improving”, appealing to those who dislike VSCode forks.

Pricing, value, and economics

  • Strong debate over paying $100–$200/month personally: some see it as trivial vs developer time; others outside high‑pay markets say it’s unaffordable.
  • Cursor’s old opacity around “Max” pricing is criticized; current “API cost + ~20%” model is viewed as more transparent.
  • Several note it’s easy to burn $10–$70/day on API‑metered agents; Cursor’s flat fee is valued as cost control.
  • Many assume all players are subsidizing usage and not yet profitable; some question the sustainability of current pricing.

Workflow, UX, and agents vs autocomplete

  • Split preferences: some want “agent as OS” (Claude Code, Codex) orchestrating across filesystem, terminals, and git; others prefer staying in the editor with strong autocomplete and light chat.
  • Concerns that agents require constant command approvals and can create sprawling, hard‑understand diffs; requests for better mapping from each change back to the agent’s reasoning.
  • Heavy users often run multiple tools/IDEs in parallel (e.g., Zed or JetBrains for editing, Cursor/Claude Code for agents).

Technical and product concerns with Cursor

  • Complaints: frequent breaking updates, sparse or late docs, opaque context selection, Python regressions, Windows “q” command bug, memory leaks, lagging behind upstream VSCode and its extensions.
  • Multi‑root and large‑repo behavior can be flaky; users resort to custom rules files and git workflows to compensate.
  • Some dislike Cursor’s divergence from VSCode (marketplace access, dev containers), and the closed‑source nature of a VSCode fork.

Business model, strategy, and trust

  • Skepticism that a proprietary VSCode fork is a durable strategy given Microsoft’s incentives and Copilot’s deep integration.
  • Others argue Cursor’s fast growth and polish justify sticking with it rather than chasing every new agent.
  • Multiple commenters express unease about possible astroturfing and bot‑written “glowing” reviews in AI‑tool threads generally, making online feedback feel less trustworthy.