Microsoft tracks GitHub Copilot token spend, pushes GPT-5.6 for internal AI

Microsoft's executive vice president Jay Parikh sent an internal email to the CoreAI organization this week, making clear that engineers' GitHub Copilot usage is now monitored. The memo, obtained by 404 Media, says 'tokenmaxxing is not what we are optimizing for' and establishes concrete guardrails: OpenAI's cheaper GPT-5.6 is the default model for internal use, every division has an AI token budget target, and employees can access a dashboard showing their personal consumption costs. Microsoft reports some engineers spend anywhere from a few hundred to a few thousand dollars per month on tokens, and warns that further restrictions may follow as spending is watched.

Parikh framed the move as 'more impact per token' rather than a retreat from AI-first strategy. Microsoft's earnings remain strong, but the financial reality of agentic coding is stark: per-token prices have dropped roughly 98% since late 2022, yet enterprise AI bills have tripled because autonomous agents burning through codebases consume vastly more than simple autocomplete. Microsoft has already canceled most Claude Code licenses and pushed employees toward GitHub Copilot CLI, and CEO Satya Nadella admitted in June to being a self-described tokenmaxxer who says 'novelty wears off' and users should match model to task. Microsoft follows Amazon, Adobe, and Citi in formalizing AI throttling and spend visibility, while Meta shut down its 'Claudeonomics' leaderboard. The leaked memo signals that even AI infrastructure giants see token discipline as a core operational concern.