--- title: Microsoft Puts Its Engineers on an AI Token Budget description: An internal Microsoft memo sets AI token budget targets by division, makes OpenAI's GPT-5.6 Sol the default model and tells staff tokenmaxxing is not the goal. author: Darie Nani (Editor-in-Chief) date: 2026-08-05T15:30:57.532Z updated: 2026-08-05T15:30:57.546Z canonical: https://www.sovereignmagazine.com/article/microsoft-ai-token-budget-copilot-engineers image: https://cdn.nanimediahouse.com/microsoft-ai-token-budget-104377.webp categories: Artificial Intelligence content_type: News region: Global publication: Sovereign Magazine schema_type: Article --- Microsoft has begun setting AI spending targets for its engineering divisions and has told staff that using as many tokens as possible is not what the company wants from them, according to an [internal memo reported on August 4 by 404 Media](https://www.404media.co/microsoft-tells-engineers-tokenmaxxing-is-not-what-we-are-optimizing-for/). The memo came from Jay Parikh, an executive vice president at Microsoft, and concerns how engineers use GitHub Copilot. Alex Heath, who writes the newsletter Sources, [reported on X](https://x.com/alexeheath/status/2084850831636234691) the following day that it also makes OpenAI's GPT-5.6 Sol the default model for internal use. ## Parikh Asked for More Impact per Token "As we accelerate our use of GitHub Copilot to deliver on our goals, we all need to be aware of how we consume tokens," Parikh wrote, in memo text quoted in reporting on it. He told staff that burning through as many tokens as possible, known informally as tokenmaxxing, is not the goal, and that he wants people focused on outcomes that move the needle for customers and the business. > "Tokenmaxxing is not what we are optimizing for." > — Jay Parikh, executive vice president, Microsoft The aim is not fewer tokens overall, Parikh said, and the changes are not meant to slow the company's push to become what he called AI-first. ## The Default Model Is Now OpenAI's Cheaper GPT-5.6 Sol The memo says Microsoft switched to GPT-5.6 Sol to get greater value from its token investment. The model is reported to cost less to run than Anthropic's Claude family, which Microsoft's Copilot tools have also drawn on. Microsoft sells more than 70 AI tools under the Copilot name and has developed its own MAI family of models. Reports say it may also route some Microsoft 365 prompts to MAI rather than to models from OpenAI or Anthropic. ## Divisions Have a Budget Target, and No Personal Cap Yet Updated internal Copilot guidelines linked from the memo state that, as of July 2026, Microsoft divisions carry an "AI token budget target" and that employees can see what they individually spend. No target has been set yet, and no personal spending limits have been enforced. The guidelines put the current range plainly: "The data shows that many engineers spend in the range of hundreds of dollars a month to a few thousand dollars in tokens." That is per engineer, for model usage alone. ## Amazon, Adobe, Citi and Uber Tightened AI Spending First 404 Media describes Microsoft as one of the last major companies to rein in employee AI use, part of a pattern the outlet has followed as the price of using AI has risen without always delivering matching productivity gains. Amazon, Adobe, Atlassian, Citi and Uber have all throttled staff AI use. Meta set token budgets and then shut down Claudeonomics, an internal leaderboard on which employees competed to burn tokens. The change does not follow a bad quarter. Microsoft's most recent results showed revenue, operating income and net income all up and ahead of Wall Street expectations. ## FAQ **Q: What is tokenmaxxing?** Tokenmaxxing is the informal term for using as many AI tokens as possible rather than using them efficiently. Jay Parikh's memo tells Microsoft staff it is not what the company is optimizing for. **Q: What is an AI token and why does it cost money?** Tokens are the units in which AI model usage is measured and billed. Longer and more frequent requests consume more of them, so the cost of a coding assistant rises with how heavily it is used. **Q: How much are engineers spending on AI coding tools?** Microsoft's own internal guidelines say many of its engineers spend between a few hundred dollars and a few thousand dollars a month on tokens. **Q: What is GPT-5.6 Sol?** An OpenAI model that Microsoft has made the default for internal use, chosen for better value on its token spending and reported to cost less to run than Anthropic's Claude models.