GitHub Shifts Copilot Chat to Token Billing

Flat-rate AI assistant packages were a loss-leader sweet spot that was unsustainable in the era of intensive agent workflows; heavy users must now audit their usage or pivot to alternative models.

* Token-based metering heavily penalizes multi-turn agent interactions and subagent loops that feed large contexts back into the LLM.

* Excluding code completions and edit predictions preserves the low-friction coding flow without sudden cost spikes for basic typing assistance.

* The transition highlights the value of Zed's open ecosystem, allowing developers to easily swap out Copilot for cheaper APIs, custom keys, or local models.

* Organizations and team managers must immediately establish spending budgets and review usage reports to avoid unexpected overages.