Microsoft CEO Satya Nadella says the company should stop reaching for the largest AI model every time it faces a problem. Instead, he wants employees to think more carefully about which tool fits the task at hand.
Speaking at a live recording of The New York Times' Hard Fork podcast, Nadella acknowledged that heavy AI use has become common inside Microsoft. When asked about so-called tokenmaxxing, a term used for pushing as much work as possible through AI systems, he replied that there is "a lot" of it happening at the company.
Nadella described the behavior as tempting because of the novelty of AI tools, but said workers eventually need to ask what they are actually trying to build. His message was that AI should be used intentionally, not simply maximized for its own sake.
The Microsoft chief said teams should avoid using frontier AI systems for routine problems. He pointed to Microsoft Copilot's auto mode, which is designed to route tasks to the most suitable model. In his view, the goal is to get useful output at a reasonable cost, rather than treating every request as a chance to use the most expensive or powerful system available.
That remark reflects a broader shift in the industry. Over the past year, many technology companies have encouraged workers to use AI extensively, sometimes through internal systems that measure usage in tokens, the units of data processed by AI models. As those costs become harder to ignore, companies are starting to focus more on efficiency.
Nadella did not say Microsoft is imposing new limits on employee AI use. But his comments suggested that the company wants a more practical approach, one that aligns model choice with business needs.
Nadella also said he recently built a tool himself using vibe coding, the practice of directing AI to help create software with minimal manual coding. The tool, he said, follows workplace conversations tied to a project and keeps the code updated when people discuss a related change.
According to Nadella, the system can generate a plan, make the update and preserve the software's functionality without him having to attend every meeting or read every thread. He presented the example as a sign of how AI can support day-to-day engineering work when used well.
The comments fit into Nadella's wider effort to reshape Microsoft for the AI era. The company has been trying to adapt its structure and culture so it can compete with smaller and faster rivals that are moving quickly in the AI market.
Microsoft has already made leadership changes that appear tied to that push. In October, Nadella named a new chief executive for Microsoft's commercial business, a move that allowed him to devote more attention to technical issues. In November, he brought in a new AI adviser to help rethink the company's business model.
During the event, cohost Kevin Roose also gave Nadella a T-shirt reading "Microsoft Advanced AI Research." Roose said the shirt came from an OpenAI employee and dated back to 2023, when Microsoft had been preparing a new AI lab for OpenAI employees after a brief leadership crisis at the startup. That lab was never created after Sam Altman was reinstated.
Nadella accepted the shirt with a laugh.
Taken together, his remarks suggest Microsoft is now emphasizing not just more AI, but better AI. The message from the company's chief executive was clear. Use the right model, spend wisely and avoid turning AI usage itself into the goal.