How JetBrains Is Bringing Surging AI Development Costs Under Control

Introduction

As generative AI becomes part of everyday software development, enterprises are facing a problem beyond tool selection: they need to understand usage, allocate budgets, and intervene when consumption rises. JetBrains’ account suggests that AI cost management is increasingly becoming an infrastructure challenge rather than a finance-only exercise.

Key takeaways

• Spending growth is happening alongside tool fragmentation. JetBrains says its development-related AI spending increased by roughly ten times over six months. Most developers use three to five AI tools per month. From January 2026, token consumption began rising sharply as stronger models appeared, including Claude Opus 4.5 and 4.6.
• The first step was visibility. The company initially spent four days manually consolidating usage and billing data from different vendors in a spreadsheet. It later automated the process with vendor APIs and an internal dashboard, providing views of current and projected spending.
• Visibility alone could not change behavior. The dashboard showed where money was being spent, but requests still traveled directly from each tool to its supplier. JetBrains therefore expanded an internal wrapper, originally built by one developer, into Central CLI, a common way to invoke both internal and third-party AI tools.
• The control point moved into the request path. Requests made through the CLI are routed through JetBrains’ existing AI platform. This allows the company to apply its AI-credit system to third-party tools and configure limits for individual developers, teams, or larger organizational units.

Why it matters

JetBrains deliberately avoided solving the problem by approving only one or two products. The company believes the AI market is changing too quickly for a fixed tool standard to remain optimal for long. A centralized access layer separates tool choice from governance: developers retain options, while administrators gain a common foundation for accounting, access management, and spending controls.

The approach is not complete. JetBrains says more than 1,000 developers adopted Central CLI within several weeks, but some terminal-based agents and personal subscriptions remain outside the managed path. The company is also working on a fair way to distribute AI budgets across users and teams.

The broader lesson is that AI cost control may require more than asking employees to use fewer tokens. A more durable approach is to connect identity, routing, metering, and limits through one control plane, then expand coverage over time. The objective is not necessarily to eliminate tool diversity, but to make that diversity measurable, governable, and adjustable.

Source: InfoQ 中文