Microsoft is putting guardrails on how much its own engineers spend querying AI models. Jay Parikh, an executive vice president at the company, emailed staff this week introducing division-level “AI token budget targets,” telling employees that some engineers currently burn hundreds to thousands of dollars a month in tokens through GitHub Copilot — spending Microsoft now wants managed with, in Parikh’s words, “the same discipline we apply to every other critical resource.”
Photo by Mizuno K on Pexels
Not About Using AI Less
Parikh was careful to frame the move as a cost-efficiency push rather than a retreat from AI-assisted coding. “We are not optimizing for fewer tokens,” he wrote. “We are optimizing for more impact per token.” The phrase “tokenmaxxing is not what we are optimizing for” quickly circulated among engineers as shorthand for the new policy, capturing Microsoft’s concern that unconstrained token usage was becoming a proxy for productivity rather than a means to it.
Defaulting to a Cheaper Model
As part of the shift, Microsoft is making OpenAI’s GPT-5.6 — a model that costs less to run than some alternatives — the default internal option for Copilot. Individual employees will also be able to track their own AI spending, giving teams visibility into who is consuming the most tokens and whether that consumption is translating into shipped work.
A Signal for the Wider Industry
Microsoft is one of the largest single deployers of AI coding tools in the world, given its ownership of GitHub Copilot and its enormous internal engineering headcount. A policy shift of this kind, from a company that has spent years pushing AI adoption internally and externally, suggests the honeymoon phase of unlimited AI usage may be giving way to a more cost-conscious era across the industry — particularly as enterprises everywhere grapple with ballooning inference bills.
Balancing Adoption With Discipline
The policy walks a fine line: Microsoft still wants engineers reaching for Copilot constantly, but it wants that usage justified by outcomes rather than treated as a free-for-all. Whether token budgets end up throttling genuine productivity gains or simply trimming waste will likely become clearer as other large tech employers watch how Microsoft’s own engineers respond to being measured on efficiency rather than raw usage.
For now, the message from Redmond is clear: AI tools are here to stay, but so is the accounting.
