When developers give AI agents coding tasks, the agents plan, generate code, run tests, and repeat to correct errors. A single task can consume 10 to 50 times more tokens than a basic chat prompt, driving up costs significantly. With per-token pricing, the workflows creating the most value can also become the most expensive.
Uber reportedly exhausted its entire 2026 AI budget by April after expanding its use of Claude Code. GitHub Copilot moved from flat-rate pricing to metered credits. Gartner predicts AI coding token costs will surpass average developer salaries by 2028.
The natural response is rationing usage caps, seat limits, and approvals before developers launch agents. However, blanket limits restrict value-creating workflows and optimize the wrong metric. Cost per token measures consumption, not results.
Some spending reflects waste, including retries and runaway loops, while some represents the work of shipping software. A meter cannot distinguish between them. A better measure is cost per merged pull request: total AI-assisted development cost divided by the number of changes that pass review and are merged.
Tracking rework and defects alongside it ensures cheaper changes do not conceal lower quality. If that cost falls as AI usage rises without compromising quality, developers are converting more consumption into useful software. The question is not how many tokens are used, but how much finished work the investment produces.
Improving that equation requires two changes: reserve expensive inference for tasks that need it, and lower the cost of routine work. Agentic coding does not require the most powerful model at every step. Use a frontier model or on-premises flagship for difficult planning and novel reasoning.
Route suitable work, including code generation and test creation, to open-weight models operated in-house. A gateway can route requests according to task complexity, data sensitivity, cost, and latency, reserving premium inference for moments that justify it. Meter the exception, not the workload.
Open-weight models make this division increasingly practical, with benchmarks showing leading models can approach frontier performance on some tasks. Evaluating candidates against own code, tests, and acceptance criteria before assigning production work is essential. At HPE, a hybrid AI platform routes routine workloads from over 150 internal applications to open-weight models running on premises, resulting in token costs down 30 to 50 percent.
That is the price of shipping software intelligently.
Source: MIT Technology Review AI · Summarized by HeadlinesBriefing