Google's Antigravity bait and switch

Original Article ↗ Hacker News Discussion ↗

Google’s abrupt overhaul of its Antigravity coding IDE into a chat-style AI agent tool — overwriting the existing app, breaking workflows, and changing limits and pricing — has reignited long‑standing frustration with the company’s product instability. Commenters describe lost settings and history, weaker functionality compared to the previous IDE and Gemini CLI, confusing migration paths, and aggressive upsell behavior, framing it as part of a broader pattern of Google “rug pulls” and poor portfolio management. Many advocate shifting to open tools, local or model‑agnostic agents, or rival IDEs to avoid lock‑in and future surprises.

Antigravity 2.0 Changes & Immediate Fallout

  • Update replaced the VS Code–style Antigravity IDE with a standalone “agent chat” app, often without clear warning or migration path.
  • Many users lost their IDE setup, history, and extensions; on some platforms the old IDE could be reinstalled, on others it required hacks or fresh installs with auto-update disabled.
  • New Antigravity CLI is promoted while the older, open-source Gemini CLI is being sunset; documentation around headless use and auth flows is described as confusing and sometimes buggy.

Reactions to Google’s Product & UX Strategy

  • Strong theme of “rug pull” and comparisons to past Google shutdowns (Reader, chat apps, various AI tools).
  • Several argue this confirms Google’s low trustworthiness as an enterprise provider and reinforces its reputation for poor product/portfolio management and internal fiefdoms.
  • Some see this as part of a broader pattern: aggressive upsell prompts in Workspace, shifting AI plans/quotas, and frequent strategic resets.

AI Coding Workflows: IDE vs Agent

  • Many preferred the old Antigravity IDE for tab completions and integrated editing; they dislike being pushed toward a single-prompt, agentic workflow.
  • Others argue the “prompt → plan → implement” agent model is the future and that separate CLIs plus an editor (VS Code, JetBrains, Vim/Neovim) are more flexible and less risky.
  • Cursor is repeatedly cited as an example of doing the IDE + agents transition well by supporting both in one environment.

Model Quality, Pricing, and Limits

  • Mixed views on Gemini/Gemma vs OpenAI/Anthropic/other labs; several say Gemini lags for coding, though it’s praised for images and on-device models.
  • Complaints about shrinking quotas, new compute-based limits, removal of bundled AI credits, and the need to buy “AI credits,” though some mention a later 3× limit bump after backlash.

Trust, Lock-In & Open Alternatives

  • Many see this as a cautionary tale against tying core workflows to proprietary, auto-updating tools from large vendors.
  • Strong advocacy for open-source or agent-agnostic harnesses and local/open-weight models to avoid lock-in and sudden product changes.