From Flat Fees to a Running Meter
For years, budgeting for developer tools was simple: multiply the number of engineers by a fixed monthly seat price. But that model is breaking down. Major players like GitHub Copilot have shifted from predictable subscriptions to usage-based billing.
Now, on top of a base seat fee, your bill is determined by 'tokens'—the basic units of data processed by the AI model. This turns your coding assistant from a fixed asset into a variable utility, like electricity. The more your team uses it, and the more complex their requests are, the higher the bill. One developer running an 'agentic' task to refactor an entire feature can consume more tokens than another developer's entire week of simple code completions.
What Exactly Is a 'Token'?
Think of tokens as the words of an AI model. A token isn't a single character or a whole word; it's a chunk of text that could be a word like 'function' or just a piece of one, like 'func'. Everything sent to the AI—the code it reads, the questions you ask, the instructions you give—is converted into tokens. The model's response is also measured in tokens. Providers bill for both 'input' (what you send) and 'output' (what you get back), often at different rates. This means a long, context-heavy chat session or a request that involves reading multiple large files will consume a significant number of tokens before the AI even generates a single line of code.
The Predictability Problem
This new model creates a major headache for finance and tech leaders: forecasting costs is incredibly difficult. Usage can spike unpredictably. A team on a tight deadline might lean heavily on AI agents, driving up token consumption. A single engineer experimenting with a powerful but expensive 'frontier' model for a complex task can generate a bill far exceeding their monthly seat price. Some reports show top-tier power users costing companies anywhere from $200 to over $2,000 per month in token fees alone. Without clear visibility, these costs can spiral, leading to budget overruns that undermine the very efficiency gains the tools were meant to provide.
Strategies for Taming the Token Bill
Managing this new expense requires a shift in mindset from simple procurement to active governance. The first step is visibility: tracking usage to see which teams, developers, and workflows are driving the most cost. Many companies are now implementing 'guardrails', such as setting budgets or spending alerts for individual users or teams. Another key strategy is matching the AI model to the task. Using a powerful, expensive model for a simple task like generating documentation is wasteful. Instead, teams can be trained to use more cost-effective models for routine work, reserving the 'premium' models for complex problem-solving where the higher cost is justified. This requires educating developers on the cost implications of their choices and providing them with the tools and policies to work efficiently.
