Terminal AI Coding Agents Surge as JetBrains AI Costs Jump 10x

The shift away from the IDE

In 2026, the center of gravity in AI-assisted programming moved toward the command line. Market-share estimates from First Page Sage put Anthropic's Claude Code, a terminal-based agent, at 29.4% of developers' primary AI coding tools as of October 2026. That is up nearly 11 points since January, and it overtook GitHub Copilot in mid-year 3. Copilot, an IDE extension, fell almost 8 points to 22.7%. Cursor, an AI-native IDE, dropped to 13.1% 3. OpenAI Codex, which runs in the terminal and the cloud, nearly tripled its share to 11.8%. First Page Sage attributes that growth to its cloud task mode 3.

Taken together, the two terminal-first tools in that table account for roughly 41% of the market. Exact figures depend on how a given tracker classifies tools like Codex, which spans both terminal and cloud. The direction is consistent, though. Editor plug-ins are losing ground, and agents that run from a shell are gaining it. First Page Sage also reports consolidation, with the four largest tools now holding 77% of the market, up from 71.2% in January 3.

JetBrains' uncomfortable position

JetBrains shows the shift from both sides. Its own AI product, Junie, holds just 4.3% of the market and slipped 0.6 points over the year 3. At the same time, the company says its internal AI development spending rose about tenfold in six months 4. Commentary on the disclosure highlighted JetBrains' admission that it had not known how to control those costs systematically before it introduced new measures 5.

The more revealing detail is how JetBrains' engineers actually work. The company says most of its developers use three to five AI tools in a given month 4. They still use JetBrains IDEs as much as before. On top of the IDE, they now layer CLI agents, agentic development environments that run several agents in parallel, and IDE-integrated assistants 4. JetBrains describes the result as a "tooling zoo." It says the challenge is balancing developer freedom against cost, compliance, security, and budgeting, rather than simply limiting staff to one or two tools as some other companies do 4.

In practice, a company that sells developer environments is paying for competitors' agents because its own developers want them. That does not mean the IDE is dead, since JetBrains' usage data suggests the editor remains the home base. But much of the AI value, and much of the AI spending, now flows to tools that run alongside the IDE rather than inside it.

Warp bets on open source and agents

Warp offers a different response to the same trend. The company has open-sourced its client, which it now calls an agentic development environment 12. OpenAI is the founding sponsor of the new repository, and GPT models, including GPT-5.5, power the workflows that manage contributions 12. Under the new model, users submit feature requests and bug reports from inside the app or through GitHub. Public issues serve as the system of record, and agents orchestrated by Warp's Oz cloud platform implement and ship changes in the open 2. Warp says even non-technical users can shape the product this way 2.

Alongside the open-source move, Warp expanded support for open-weight models, including Kimi, MiniMax, and Qwen. It also added an "auto (open)" router that picks the best open model for a given task 1. Warp now claims more than a million active developers 5. That suggests the terminal-agent boom extends beyond the big labs' own tools to independent players 5.

The arrangement has a tension built into it. Warp is widening access to open models for end users while relying exclusively on one large lab's models to run its own development pipeline 12. OpenAI, for its part, gains a showcase for agent-managed software development in a category where Codex is already growing fast 3.

What it adds up to

Three developments point the same way. Terminal agents are winning share, incumbents are paying heavily to keep up, and newer entrants are rebuilding their development process around agents.

The market is not simply "terminal versus IDE." JetBrains' internal data shows developers stacking tools rather than switching between them 4. The real competition may be over which product becomes the orchestration layer: the place where agents are launched, monitored, and paid for. JetBrains is approaching that through cost governance 4. Warp is approaching it through an open, agent-run repository 12. Anthropic and OpenAI are approaching it through agents that developers increasingly choose first 3.

The tenfold cost jump is the warning sign. If one well-resourced toolmaker struggled to keep its AI bills under control, smaller engineering organizations should expect cost management to become as important as model quality when they choose tools.