Cursor 3

Original Article ↗ Hacker News Discussion ↗

Cursor’s new “agent‑first” Cursor 3 release is seen as a strategic shift away from its original role as an IDE that tightly augments a human developer, toward a Codex/Claude Code–style workspace where swarms of agents own more of the workflow. Commenters are split: some praise multi‑agent orchestration, cloud “computer use,” multi‑model support, and deep integration with tools like Linear and GitHub, while many power users worry about losing code‑centric workflows, rising token‑based costs versus first‑party offerings, and growing confusion over what Cursor actually is. A recurring theme is whether the future of software work should prioritize human‑driven coding with strong autocomplete and inline assistance, or higher‑level task planning where agents generate most of the code and developers mainly review and steer.

Overall reception

  • Very mixed response. Some praise Cursor 3 as a strong agent workspace; many existing users are uneasy or considering switching tools.
  • Several note Cursor has “fallen out of the cool kids club” but remains very productive for some heavy users.

What Cursor 3 adds / does well

  • Multi-model support (Anthropic, OpenAI, Gemini, Composer, some OSS models) remains a major selling point, including mixing expensive and cheap models in workflows.
  • Strong inline autocomplete in the IDE is repeatedly cited as best-in-class and the main reason some people still use Cursor.
  • Cloud platform + “computer use in the cloud” (remote dev envs, SSH, browser, agents that can test what they write) is seen as Cursor’s most differentiated feature.
  • New UI for managing many agents and cloud worktrees in parallel is welcomed by users who already work in “agent swarm” style.

Concerns about direction (agent-first vs code-first)

  • Many long-time users liked Cursor as “IDE-first with an assistant,” i.e., human driving, AI assisting within a traditional editor.
  • Cursor 3’s Codex/Claude Code–like agent workspace is perceived by some as moving toward “vibe coding” and away from code visibility, LSP tools, and fine-grained control.
  • Multiple commenters fear that autocomplete and IDE-centric features will be de-prioritized or sunset over time.
  • Others argue fully agentic, chat-first workflows are the future and IDEs will matter less.

Value vs Claude Code / Codex / VS Code

  • Recurrent question: why pay Cursor when Claude Code, Codex, Copilot + VS Code (or Zed, Sublime, Neovim) can provide similar or better models and workflows?
  • Several report massive cost savings moving from Cursor enterprise/API-based usage to Claude Max / ChatGPT-style fixed plans, with comparable or better productivity.
  • Others say Cursor’s harness, worktree management, and cloud agents still feel smoother and more capable than raw Claude Code or VS Code extensions.

Pricing, business model, and lock‑in

  • Complaints about hitting limits, expensive overages, and enterprise renewals shifting to heavy usage-based pricing.
  • Some see Cursor’s attempts to center its own agent UI and reduce room for third‑party extensions as anti‑user but rational for monetization.
  • Skepticism that a non–model-owning IDE company can sustain margins against Anthropic/OpenAI’s subsidized subscriptions.

Stability and platform issues

  • Reports of Cursor 3 crashing on Linux (Wayland/Fedora), with mixed experiences depending on install method.
  • Some revert to plain VS Code or alternative editors due to regressions, plugin breakage, or UI churn in Cursor.