Cursor 1.7

Original Article ↗ Hacker News Discussion ↗

Perceived Value & Pricing

  • Many feel Cursor was uniquely compelling a year ago but is less so now as VS Code, Claude Code, and Codex improved.
  • Several users complain that the Pro plan’s included credits run out in 1–2 weeks with heavy use; some report bills in the hundreds per month.
  • Confusion and frustration around usage visibility: itemized token spend is shown, but included-plan vs on-demand usage is unclear for some.
  • Some still consider it good value if you want multiple frontier models under a single subscription; others say you can get similar or better results cheaper elsewhere.

Autocomplete & Prompt UX

  • Strong consensus that Cursor’s tab-complete is its standout feature; multiple users say it’s dramatically better and faster than VS Code/Copilot, to the point of being the main reason they stick with Cursor.
  • A minority actively dislike the aggressive autocomplete, finding it distracting, “doing too much,” or subtly steering their code; some disable it or change keybindings.
  • New autocomplete in the prompt box sparked debate: some see it as helpful filename/context completion, others worry it encourages “vibe prompts” and degrades rigorous prompting.

Agents, Workflows & IDE vs CLI

  • Some praise Cursor for parallel agents, easy model switching, and zero-config IDE integration, especially for incremental refactors and guided changes.
  • Others prefer CLI tools (Claude Code, Codex, Aider, Kilo/Opencode) where AI feels like a discrete tool: run a job, then review diffs via git.
  • Several users say Cursor’s agentic behavior can become chaotic on larger tasks, misapply edits, or require heavy babysitting; they value fine-grained control and small, reviewable changes.
  • Concerns raised about agents introducing security issues (e.g., weak passwords) or massive unreviewed diffs, and about humans’ limited capacity to oversee many concurrent agents.

Comparisons to Alternatives

  • Claude Code and GPT-5 Codex are frequently cited as equal or better for reliability and large edits, especially via CLI or new VS Code extensions.
  • Some now favor VS Code + official extensions + git for rollback, saying Cursor’s earlier differentiators (state management, diffs) have been matched.
  • Autocomplete competitors: Supermaven is considered decent but still behind Cursor; Copilot’s autocomplete is often described as slow or poor.

Reliability, Bugs & Product Direction

  • Users report intermittent latency, terminal integration issues (zsh themes, state), keyboard shortcut breakage, and regressions between versions.
  • Terminal sandboxing is noted (implemented via sandbox-exec), but details and UI affordances remain unclear to some.
  • Skepticism about Cursor’s large valuation and long-term moat: many see it as a polished wrapper dependent on underlying model providers, vulnerable as official tools catch up.