Microsoft urges engineers to limit AI use due to high costs

Microsoft has internally opened access to OpenAI's GPT-5.6 model to derive value from its token investments, but will soon introduce a 'target AI token budget' for employees to track individually. The company's executive vice president Jay Parikh announced in an internal email that they will update internal guidelines and manage token costs with the same discipline as other critical resources. Token consumption is proportional to the volume of AI requests and model responses. Previously, tech workers were encouraged to use AI more actively, but the shift to usage-based billing has led to soaring costs. In late May, Microsoft moved GitHub Copilot to a per-token pricing model, and some users quickly exhausted their limits. The company is also trying to reduce AI spending by connecting Microsoft 365 apps to its own MAI models instead of more expensive Anthropic and OpenAI models. Amazon, Adobe, Atlassian, and Citi have also tightened control over employee AI token spending.