JetBrains Air: A System of Products for Agentic Software Development

Original Article ↗ Hacker News Discussion ↗

JetBrains’ new “Air” system for agentic software development is meeting a mixed reception from long‑time users. Many praise the company’s traditional IDEs for code navigation, refactoring, and database tooling, but express frustration with rising resource usage, confusing and fragmented AI offerings (Air, Junie, AI Assistant, etc.), and key limitations such as the inability to use personal Claude subscriptions. Some see multi‑vendor, agent‑centric tooling as a sensible direction, while others worry JetBrains is neglecting core IDE quality and determinism in favor of catching up with AI‑first competitors.

Overall reaction to JetBrains Air and AI tools

  • Many long-time JetBrains users are skeptical or disappointed with current AI offerings (Junie, Air, AI Assistant/Chat/autocomplete).
  • Common complaints: buggy, prototype quality, behind competitors like Cursor/Codex, and intrusive “AI everywhere” UX.
  • Some see Air’s multi-vendor, “AI realist” approach as sensible and aligned with JetBrains’ traditional strengths, but not yet compelling enough to change workflows.

Fragmented and confusing AI product lineup

  • Multiple overlapping names (Air, Junie, AI Assistant, AI Chat, AI autocomplete, ACP connectors) confuse users.
  • Several commenters say they cannot clearly tell what each piece does or how they relate, even after trying them.
  • Desire for a simpler, unified story and “one thing done really well.”

Model integration, Claude, and local agents

  • Major friction: Anthropic’s ToS blocks using Claude Pro/Max/Team subscriptions inside JetBrains tools. Many see this as a show‑stopper.
  • JetBrains staff say Claude API can be used via ACP/agents, but not subscription logins, and they failed to negotiate an exception.
  • Some users want OpenRouter/local models without “calling home” and report issues running fully offline. Others say ACP + third‑party agents works well for local or multi‑provider setups.
  • Auth and configuration flows (Codex/Cursor/Claude) are seen as clunky or broken by several testers.

Changing workflows: IDE vs agents

  • A visible group now spends most time in terminal/agent harnesses, using IDEs mainly for navigation, git, and debugging.
  • Others say IDEs are more valuable with LLMs, especially for navigating large, agent‑generated codebases, deterministic refactoring, and MCP‑backed tools.
  • Ongoing debate: how much humans should still read/write code vs delegating to agents and reviewing selectively.

IDE quality, performance, and product direction

  • Repeated complaints about heavy resource use, slow indexing, poor behavior with git worktrees, and UI changes (new UI vs “classic”).
  • Some feel core IDE features and determinism (autocomplete, refactoring, stability) are stagnating or regressing while AI gets priority.
  • Long‑time paying customers report canceling or considering cancelation; trust has been eroded, though some keep licenses to expose IDE tooling to agents via MCP.