A Microsoft executive reportedly emailed engineering staff introducing division-level "AI token budget targets," pushing back on what the memo described as "tokenmaxxing" — excessive, cost-inefficient use of AI coding tools.
What Prompted It
Internal guidance reportedly noted that some engineers were running up costs of hundreds to several thousand dollars a month in token usage through Microsoft's Copilot tooling.
The Fix
As part of the cost-conscious shift, OpenAI's GPT-5.6 has reportedly become the default internal model for many workflows specifically because it's cheaper to run than the alternatives.
Why It's Notable
Even as AI companies tout falling model prices, this suggests that unmanaged usage at scale inside large organizations can still add up to a meaningful cost — a lesson likely relevant well beyond Microsoft.
