Microsoft CEO Nadella argues AI's real problem isn't capability but that people haven't learned to use it yet

Microsoft CEO Satya Nadella argues AI models are more capable than people realize; they just don't know how to use them yet. He also wants to move past the "AI slop" debate.

In a blog post, Nadella writes that AI's discovery phase is over and widespread adoption is beginning, making it easier to separate "spectacle" from "substance."

He points to what he calls a "model overhang," a scenario where trained models have capabilities that will only be unlocked later as people figure out how to use them. This could mean new prompting techniques, fine-tuning, or simply finding the right use cases.

This echoes OpenAI's view that the public massively underestimates how far AI has come. Most people still see AI as chatbots or better search engines, even though these systems can already handle complex tasks that used to take experts hours to complete, the company claims.

Nadella also wants to move past the "slop vs sophistication" debate entirely; slop being the low-quality, often pointless AI output flooding the internet. Instead, he calls for evolving Steve Jobs' "bicycle for the mind" metaphor, arguing we should see AI as scaffolding for human potential. But he goes further: we need a new "theory of mind" that accounts for how humans relate to each other when everyone has access to these cognitive amplifiers. "This is the product design question we need to debate and answer," he writes.

Systems over models in 2026

Looking ahead, Nadella predicts the focus will shift from individual models to complex systems in 2026. Instead of relying on single-model performance, developers will build systems that coordinate multiple models and agents, manage memory and access rights, and enable secure tool use.

This added complexity is necessary to deploy AI in the real world and work around the "jagged edges" of models—their unpredictable mix of surprising capabilities and unexpected failures. Nadella also stresses that smart decisions about scarce resources like energy, computing power, and talent will be essential. For AI to gain broader acceptance, it needs to tackle specific challenges facing "people and planet," though he expects a "messy process of discovery."