TL;DR by CuriousCats.ai
Microsoft imposes internal token budgets
- Microsoft moves away from tokenmaxxing as it shifts to limit AI use by employees.
- Parikh emails about mindful token consumption to employees as Copilot ramp continues.
- Token pools allocated to each department, with usage adjusted up or down as needed.
- Default to GPT-5.6 Sol in GitHub Copilot, per Parikh's instruction.
- Spokesperson confirms default GPT-5.6 Sol as default for internal Copilot use, with other models available.
Context: AI spending trends and provider ties
- Token budgeting reflects a broader cost-control trend as AI costs rise and tokenmaxxing fades across corporate America.
- OpenAI and Anthropic ties show Microsoft aligning with major AI providers.
- Copilot user base has reached 50 million users, illustrating scale.
- CoreAI governance signals model-defaults will be updated as models evolve.
Other critical updates on internal AI token policy
- CoreAI governance will continue to adjust model defaults as products evolve.
CuriousCats Full Story
Key Insight
“The move marks a reversal from the era of 'tokenmaxxing,' when developers were encouraged to run up large token bills. Microsoft's cash generation fell 23% from a year earlier, and Wall Street is pressuring hyperscalers to justify massive AI spending, with capex expected to top $700 billion collectively this year.”
CuriousCats studied:
1
computerworld.com
computerworld.com“Until recently, it was common for companies and organizations to engage in “tokenmaxxing” — that is, . But with AI costs going up, companies are now looking to save money, a trend underscored by a recent Microsoft decision to limit AI use by its employees.”
computerworld.com →2
CNBC
CNBC“"Internally, shifting more workloads to OpenAI models helps us get greater value from our token investment," Jay Parikh, executive vice president of Microsoft's CoreAI engineering group, wrote this week in a memo to employees that was viewed by CNBC.”
CNBC →Ask CuriousCats
What is token budgeting at Microsoft?
Why default to GPT-5.6 Sol in Copilot?
What triggered the efficiency push?
Are AI budgets rising at hyperscalers?
Which tools beat tokenmaxxing strategies?
