Microsoft CEO and head AI peddler Satya Nadella wants you to know that it’s time for the next phase of AI acceptance, where we focus on how humans are empowered by tools and agents and how we deploy resources to support this growth.
Amid doubts that revenue from Microsoft Copilot subscriptions and cloud AI services will compensate for data center capital expenditures any time soon, Satya has some incentive to convince customers and investors that AI is a financially intelligent long-term bet.
Thus he has taken up the pen in an act of “thought leadership,” to use the marketing jargon of his company’s LinkedIn where he announced his post on a new blog entitled “sn scratchpad.”
Early in his post Nadella writes, “We have moved past the initial phase of discovery and are entering a phase of widespread diffusion.”
That’s fair enough, given that widespread can mean anything. According to Pew Research, 62 percent of US adults say they interact with AI at least several times a week. The more relevant stat for Microsoft, however, is the percentage of customers who pay for Copilot and other AI services. That remains a work in progress.
He also offers a new variant on the filler phrase “it’s early days,” so often heard from tech execs deflecting inquiries about lack of results. “We are still in the opening miles of a marathon,” he opined. “Much remains unpredictable.”
Getting down to business, the Microsoft CEO says that there are three things the tech industry and society need to “get right” in 2026 to make AI return real value for everyone. First, we must develop a “theory of mind” that treats AI as a tool that amplifies humans and products should be designed around this belief. He uses Steve Jobs’ famous quote that computers are “bicycles for the mind” as a jumping off point.
When Microsoft is having to reassure people that its own research on the labor impact of AI shouldn’t be interpreted to mean that jobs will be eliminated, you can see why Nadella might be keen to avoid the impression that AI is intended as a substitute for human labor.
“What matters is not the power of any given model, but how people choose to apply it to achieve their goals,” Nadella wrote.
Second, Nadella posits that we move from “models” to “systems” for AI such that multiple models and agents working together can provide value.
“We are now entering a phase where we build rich scaffolds that orchestrate multiple models and agents; account for memory and entitlements; enable rich and safe ‘tools use,'” said Nadella.
So perhaps we can look forward to agents failing less than 70 percent of the time.
Finally, he argues, that society needs to make tough decisions about how and where to deploy AI so that it has “real world impact.”
“The choices we make about where we apply our scarce energy, compute, and talent resources will matter,” he wrote. “This is the socio-technical issue we need to build consensus around.”
Building consensus in this case apparently means forcing AI on everyone despite protestations, expressing bafflement that anyone would reject the technology, and spending millions on lobbying. ®