GitHub Copilot has moved to a usage-based billing system where subscriptions include a monthly allowance of AI Credits. Meanwhile, agent mode has become the default way we work. Agentic workflows can gather context, read files, run tools, and iterate repeatedly, so one task may consume far more tokens than a simple question. And most organisations operate within fixed budgets, with little appetite for overages - it means you can run out of credits mid-month and lose access to the tool your workflow now depends on.

It doesn't have to be like that.

In this talk, we'll pull back the curtain on where your credits actually go: input tokens, output tokens, and cached context. Then we'll turn credit consumption into something developers can actively understand and control.

This talk is for developers who reach for GitHub Copilot every day, though the ideas apply to any coding assistant. You’ll leave with concrete techniques for getting more useful work from a fixed AI budget - without giving up the way you work.