JetBrains AI Costs Surge 10x as Terminal Agents Gain Ground

What happened

JetBrains, the Prague-based maker of IntelliJ IDEA, PyCharm and a long list of developer tools, has acknowledged that its own spending on AI development rose roughly tenfold over six months. In a post titled "Our First Moves to Get AI Spend Under Control," the company said it noticed the costs climbing but realized it had no systematic way to rein them in 1. The admission comes alongside financial data showing JetBrains reporting a net loss for the first time in its tracked history, based on revenue and net income figures for JetBrains s.r.o. stretching from 2005 to 2025 1.

One analysis puts that loss at about $14 million and ties it directly to the company's AI push 4. That push includes Junie, a coding agent now built into all JetBrains IDEs and reportedly used by 343,000 people; JetBrains Air, a standalone agentic environment; and Mellum, an open-sourced coding model 4.

Why the cost story matters

The 10x figure is notable because it comes from a company that understands developer workflows as well as anyone. If JetBrains struggled to predict and govern AI spending, smaller teams adopting agents should expect the same surprise. Agentic tools don't just answer a prompt. They read repositories, plan multi-step changes, run commands, execute tests and loop on failures, and every iteration consumes compute.

The broader analysis frames JetBrains' loss as part of a power shift. It contrasts the company's red ink with Cursor's reported $4 billion in annual recurring revenue and argues that the IDE era is fading while a new fight over the "agentic terminal" begins 4. That is a provocative reading. Calling it a collapse probably overstates what a first loss after two decades of profitability means. Still, the direction of travel is hard to dismiss.

The terminal as the agent's home

The argument rests on where coding agents now do their work. Claude Code runs in the terminal and can handle a full edit-test-iterate cycle without an IDE open. OpenAI has rebuilt Codex CLI into a full-screen text interface with parallel agent management, voice input and remote control from a phone. Gemini CLI runs locally with a generous free tier 4. For agents, the command line is a natural fit. It is scriptable, composable and already the place where builds, tests and deployments happen.

Warp shows how much attention the terminal itself is getting. The Rust-based terminal emulator, available on macOS, Windows and Linux, offers AI command suggestions, code generation, shared runbooks through Warp Drive and an IDE-like editor 2. Its block-based interface groups terminal history into discrete input-output units that users can navigate and share 3. In April 2026, Warp said it would open-source both its terminal and its Oz orchestration platform for cloud agents under MIT and AGPL licenses, with financial support from OpenAI. Cloud products like Warp Drive will stay closed 2.

The funding history reflects investor conviction, though accounts of it differ. Both describe a $50 million Series B led by Sequoia in June 2023 23. One source describes a $17 million Series A led by Figma CEO Dylan Field plus a separate $6 million GV-led seed round 2. Another describes a $23 million Series A led by GV and Field, and puts total funding at $73 million 3. The gap looks like a question of how the early rounds are grouped rather than a real disagreement over the money raised.

JetBrains' counter-move

JetBrains is not simply conceding. In March 2026 it launched Air, which it calls an "Agentic Development Environment," where developers guide agents and review their output instead of writing code 4. Air supports several outside agents, including Codex, Claude, Gemini CLI and JetBrains' own Junie, through the Agent Client Protocol. It comes in three parts: orchestration inside JetBrains IDEs, Air Teams for coordinating humans and agents, and Air Governance for policy, auditing and cost management 4.

That last part is telling. A company that just watched its own AI bill jump tenfold is now selling cost controls to other teams.

The takeaway

The clearest lesson here is less "terminal beats IDE" than "agents change the economics of tooling." Agents are moving toward the command line, and vendors are positioning around orchestration rather than editing. JetBrains' bet is that it can become a neutral control plane that sits above whichever CLI agents developers prefer, rather than competing with them head-on. Whether that works depends on something its own budget suggests nobody has solved yet: making agentic development predictable enough to pay for.