由于 Token 成本飙升，微软限制工程师 AI 支出
微软为工程师设定了严格的代币预算，并将内部默认模型切换为更便宜的版本，标志着从无节制的 AI 采用向成本效益型的转变。
microsoft is restricting its engineers' use of ai tools like github copilot due to soaring token costs, with some employees spending thousands monthly. executive vice president jay parikh stated 'tokenmaxxing is not what we are optimizing for' and mandated a shift toward cheaper models like gpt-5.6 to get 'more impact per token.' starting july 2026, divisions will face specific 'ai token budget targets,' forcing a re-evaluation of how ai drives productivity. this shift underscores the growing tension between the hype of widespread ai adoption and the economic reality of running inference at scale.