Breaking down JetBrains' complex AI agent strategy (leaddev.com)

JetBrains has started weaving third‑party AI agents into its IDEs while also pushing its own model strategy: it offers Junie (running internally on ChatGPT‑5 by default, with Claude variants available) and has integrated Anthropic’s Claude Agent via an AI chat interface that uses JetBrains’ Model Context Protocol (MCP) to give agents full IDE access. The company also adopted the emerging Agent Client Protocol (ACP) alongside Google and Zed Industries, promising users the ability to switch between multiple agents under a single subscription. At the same time JetBrains says it will train proprietary LLMs and is collecting rich telemetry—code snippets, prompts, responses, edit history, and terminal usage—to improve models. The rollout’s technical ambition is tempered by friction: Junie’s initial popularity has been undercut by new “AI Ultimate” pricing ($30/mo for 35 credits) and tighter credit economics that many power users say make the product unusable mid‑subscription, prompting migrations to pay‑per‑use alternatives like Claude Code. The episode highlights a core industry tension: build-first (own models + data collection) versus integrate‑first (best‑of‑breed agents). For the AI/ML community it matters because MCP/ACP adoption could standardize deep editor‑agent interactions, but developer trust—around pricing transparency and data usage—will determine whether teams share the telemetry needed to improve models or move to competitors offering clearer economics and privacy guarantees.