Microsoft Conference: Kevin Scott Says AI Outpacing Use Cases as Code Review, Compute Constraints Bite

Key Points

• AI capabilities outpacing use cases: Kevin Scott says models are “way more powerful than what people are using them for,” creating a capabilities overhang where the near-term challenge is applying existing AI with judgment and purpose.
• Code review, not generation, is the bottleneck: AI has sparked an “absolute frenzy” in software development—code production is easy but review, domain understanding, and product judgment lag, shifting engineering toward problem selection and quality over mere output.
• Infrastructure and access limits persist: Inference demand and costs remain high (Scott cites ~\$150,000/year for some ambitious coding-agent use), so hyperscaler constraints and the need for silicon diversity will keep access limited despite Microsoft’s platform and OpenAI partnerships.

In a wide-ranging conversation about artificial intelligence, software development, and infrastructure constraints, Microsoft NASDAQ: MSFT executive Kevin Scott said AI progress has accelerated faster than he expected, even if the direction of change has been foreseeable for years. Scott described today’s AI models as already “way more powerful than what people are using them for,” and argued that the biggest near-term challenge is not just building more capable systems, but learning how to apply existing capabilities with judgment and purpose.

AI progress is accelerating, with “capabilities overhang”

Scott said that even seven years ago it appeared likely that “scaling laws were gonna work” and that increasingly capable models would behave like a platform others could build on. What has surprised him is the speed of the acceleration, not the underlying trajectory.

He also said the industry is “not yet at the point of diminishing marginal return” on capability improvements in the infrastructure and platforms driving AI. At the same time, he sees a “capabilities overhang,” with real-world usage lagging what the technology can do.

Software development: more code, but review becomes the bottleneck

Scott pointed to coding as the clearest example of advanced utilization of AI model capability. He described an “absolute frenzy” in software development and said even highly experienced developers are struggling to keep up with the pace of change.

He argued that AI shifts software engineering away from the mechanics of typing code and toward the fundamentals good engineers have always focused on: understanding what to build, why it matters, and how value is created. Scott emphasized that producing code is not the same as producing good code, warning that teams can “spray a bunch of stuff” using coding agents without improving outcomes.

In his view, “review has become the bottleneck,” and teams need to avoid confusing “activity and progress.” He added that defining true engineering productivity has long been difficult, and AI increases the importance of “choice and taste,” domain understanding, and clarity about customer needs.

Scott also said he expects vocational aspects of the software engineering job to change “so radically over the next handful of years” that the work may become “unrecognizable,” while the more analytical and conceptual skills become more important.

Education and the return to “computer scientists”

Offering a personal view he noted was not an official Microsoft position, Scott said he hopes computer science education shifts away from vocational training toward broader scientific thinking. He said students should focus on the ability to think algorithmically, decompose problems, choose the right problems to work on, and understand how computing fits into society and science more broadly.

He also highlighted the importance of curiosity and the ability to “punch down through the layers of abstraction” to debug when systems do not behave as expected.

An optimistic AI future tied to demographics—and a pessimistic one tied to distraction

Scott’s optimistic scenario for AI is rooted in demographic trends. He recounted a conversation with a leader of an educational institution in Japan who said the current year marks “peak high school graduation” in Japan, implying fewer graduates in future years due to birth rates and population decline. Scott extended the point to other countries, including China, Korea, parts of Western Europe, and the United States “sans immigration,” arguing that aging populations and shrinking workforces will create large productivity challenges.

In that context, Scott said technological interventions are necessary to maintain what society considers normal quality of life. He characterized AI as arriving at a pivotal moment, providing at least a partial answer to how work gets done when labor dynamics change over the coming decades. However, he cautioned that AI “can’t do everything” and is not a substitute for all human work.

His pessimistic scenario was less about science fiction and more about misallocation of attention: using AI in “superficial” ways that distract rather than address important problems. He offered an example from his own household, describing how his kids sometimes use AI for ambitious technical projects, but other times for frivolous image generation, and said he hopes society resists making AI’s narrative about sensationalism rather than what society “really need[s] from this technology.”

Microsoft’s platform mindset, OpenAI partnership, and ongoing infrastructure constraints

Asked what is most misunderstood about Microsoft, Scott said the company is fundamentally a platform company that focuses on building tools others can build on. He said decades of platform experience have given the company “infinite patience” for the messy reality of technological transformations, including a willingness to start early, get some things wrong, and iterate while serving enterprise needs.

Scott said he is especially proud of helping recognize the shift from narrow, domain-specific machine learning to more general AI, and of helping bring those capabilities “out into the open” through Microsoft’s partnership with OpenAI and associated work. He said he values a world where powerful AI capabilities are accessible broadly—such as via an API—rather than controlled solely by a single company’s internal decisions.

On hyperscaler infrastructure, Scott said he expects constraints to persist “for a while,” noting that demand keeps “exploding.” He said he does not see demand for inference declining and cited an example that some of the most ambitious teams using coding agents can face inference costs of roughly $150,000 per year, limiting access today to a small segment of developers even though many could benefit.

He also addressed silicon strategy, emphasizing Microsoft’s reliance on partnerships and “silicon diversity.” Scott said the company has its own chips, but also operates “gigantic fleets” of NVIDIA and AMD hardware, and will deploy what is most cost efficient at scale while managing the complexity across varied infrastructure.

Closing the discussion, Scott said he wants people to remember technology is “a tool,” and that outcomes depend on choices about how it is used and prioritized. He said society should aim to use technology to convert “zero-sum” problems—defined by scarcity and constraints—into “non-zero-sum” outcomes that better serve others.

About Microsoft NASDAQ: MSFT

Microsoft Corporation is a global technology company headquartered in Redmond, Washington. Founded in 1975 by Bill Gates and Paul Allen, Microsoft develops, licenses and supports a broad range of software products, services and devices for consumers, enterprises and governments worldwide. Its operations span personal computing, productivity software, cloud infrastructure, enterprise applications, developer tools and gaming.

Microsoft's product portfolio includes the Windows operating system and the Microsoft 365 suite of productivity and collaboration tools (Office apps, Outlook, Teams).

This instant news alert was generated by narrative science technology and financial data from MarketBeat in order to provide readers with the fastest reporting and unbiased coverage. Please send any questions or comments about this story to contact@marketbeat.com.