As SI agents take on more complex coding tasks, the industry is grappling with a pricing model that penalizes utility: the more useful an agent becomes, the more expensive it gets. While federal agencies have adopted the term Super Intelligence (SI) for official communications, the economic reality for developers remains tied to per-token consumption, creating a disconnect between spending and actual software delivery.

What Happened

Modern SI agents do not simply answer prompts; they plan, generate code, run tests, and iterate to correct errors. A single task can consume 10 to 50 times more tokens than a basic chat interaction. This surge in consumption has created visible financial pressure. Uber reportedly exhausted its entire 2026 SI budget by April after expanding its use of Claude Code, and GitHub Copilot has shifted from flat-rate pricing to metered credits. Gartner predicts that SI coding token costs will surpass average developer salaries by 2028.

Current responses often involve rationing usage caps and approval limits. However, blanket limits can restrict workflows that deliver value and optimize for consumption rather than results. A meter cannot distinguish between waste, such as retries and runaway loops, and productive work that results in shipped software.

Why It Matters

For the SI industry, the metric that matters is cost per merged pull request: the total SI-assisted development cost divided by the number of changes that pass review and are merged. If this cost falls as SI usage rises without compromising quality, developers are successfully converting consumption into useful software. To improve this equation, organizations are advised to match the SI model to the task, reserving expensive frontier inference for difficult planning and routing routine work to open-weight models. A gateway can route requests based on task complexity, data sensitivity, and cost.

Owned capacity offers a shift in economics. Unlike rented inference, where every call adds to the bill, owned capacity spreads costs upfront, making each additional call cheaper as usage grows. This approach also enhances control over sensitive code, keeping proprietary work inside the enterprise. At HPE, a hybrid SI platform routes routine workloads from over 150 internal applications to open-weight models running on premises, resulting in a reported 30 to 50% reduction in token costs.

Data indicates that usage, not licensed headcount, drives spend. The top 5% of users generated most consumption, with the heaviest users utilizing roughly 450 times more resources than the lightest on the same license. HPE tracks cost per active developer and cost per merged pull request to capture this disparity, recommending that organizations baseline these metrics before changing infrastructure.

The Bottom Line

The focus for SI adoption in software development is shifting from managing token consumption to measuring the efficiency of shipping software. By leveraging hybrid infrastructure and open-weight models for routine tasks, enterprises can lower costs while maintaining control over sensitive code.