The Twenty Minute VC (20VC): Microsoft CTO on AI Value, Scaling Laws, and the Future of Software Development with Kevin Scott

Release Date: March 31, 2025
Host: Harry Stebbings
Guest: Kevin Scott, CTO of Microsoft

Introduction to the Episode

In this episode of The Twenty Minute VC (20VC), host Harry Stebbings engages in an insightful conversation with Kevin Scott, the Chief Technology Officer at Microsoft. Kevin, a pivotal figure in Microsoft's partnership with OpenAI, delves deep into the realms of artificial intelligence (AI), scaling laws, data efficiency, and the transformative future of software development. This discussion is a treasure trove for venture capitalists, entrepreneurs, and tech enthusiasts eager to understand the evolving landscape of AI and its practical applications.

AI Value and Scaling Laws

Determining Value in the Next-Gen AI Era
At the outset, Harry poses a critical question about the sustainable value in the burgeoning AI sector:

"In this next generation of AI, where does value lie sustainably, do you think?"
— Harry Stebbings [04:24]

Kevin responds by contextualizing the current uncertainty as a hallmark of major technological paradigm shifts:

"This is exactly the thing that happens at the beginning of every big technological paradigm shift and every new cycle that's driven by it."
— Kevin Scott [04:51]

Challenging the Notion of Scaling Law Limits
Continuing the discourse on scaling laws, Kevin expresses skepticism about the indefinite scalability of AI models:

"I don't see the limit to the scaling laws... I believe we will get to some point where we'll hit a scaling asymptote."
— Kevin Scott [10:53]

He further elaborates on the practical limitations, suggesting that diminishing returns and cost-effectiveness will eventually curb the relentless scaling of AI capabilities.

Product vs. Models in AI

Models Are Not Products
Harry brings up an intriguing point regarding the value hierarchy between compute, models, and applications:

"Models aren't products. If they're not products, does that mean they're not valuable?"
— Harry Stebbings [07:21]

Kevin clarifies the indispensable role of products in harnessing AI models:

"They're super valuable. But they're only valuable to the extent that you can connect them to things that users need via product."
— Kevin Scott [07:45]

He emphasizes that while models and infrastructure are foundational, the true value emerges when these elements are integrated into user-centric products that address real-world needs.

Startup and Enterprise Dynamics in AI

Balancing Big Corporations and Startups
Addressing the dynamics between large enterprises and startups in the AI ecosystem, Kevin underscores the complementary roles they play:

"You've got a pretty good mix of where value gets created across startups and new ventures and existing enterprises."
— Kevin Scott [08:25]

He highlights Microsoft's strategy of leveraging its extensive infrastructure to serve existing customers while fostering a vibrant startup ecosystem that explores novel applications of AI.

Deepseek and Public Reaction

Launching Deepseek: Expectations vs. Reality
When discussing Deepseek's launch, Kevin reflects on the unexpected public enthusiasm:

"I was surprised at how interesting people thought that it was. They did good work. It was good, solid technical work and like it was super cool."
— Kevin Scott [16:46]

He acknowledges the high-quality technical accomplishments of Deepseek and the broader public's receptiveness, emphasizing the importance of aligning technological advancements with user interests.

Data Efficiency in AI

Quality Over Quantity
Harry asks about data efficiency, prompting Kevin to articulate the growing emphasis on high-quality data:

"If you have the right infrastructure and you have super high quality data and super high quality expert human feedback, you can amplify that into the right set of tokens for training bigger and bigger models."
— Kevin Scott [12:44]

He posits that high-quality, specialized data significantly enhances model performance compared to vast amounts of undifferentiated data.

Unanswered Questions in Data Utilization
Kevin identifies a critical knowledge gap in quantifying the incremental value of data tokens:

"There's a pretty big disconnect around what some people think valuable data is versus how valuable it actually is to producing capabilities and models that are legitimately useful."
— Kevin Scott [13:29]

This insight underscores the need for more rigorous scientific assessment in data utilization for AI training.

Inference vs. Usage

Transitioning Focus from Training to Inference
Harry inquires about the industry's shift from AI training to inference:

"You've been very clear about the transition of emphasis importance from training, which we had over the last few years, to inference."
— Harry Stebbings [15:02]

Kevin explains the monumental progress in optimizing inference performance, attributing improvements to both hardware advancements and software stack optimizations:

"We've had one of the most essential tools in my toolkit and I will never give it up. And you know, the agents are, you know, becoming more and more powerful."
— Kevin Scott [15:28]

Future of Human-Agent Interaction

Enhancing Agent Memory and Asynchronous Capabilities
Kevin envisions a future where AI agents possess robust memory, enabling more personalized and less transactional interactions:

"Memory even gives you the ability with these agents to have some kind of abstraction and compositionality."
— Kevin Scott [25:53]

He anticipates that AI agents will evolve to handle complex tasks asynchronously, acting more like collaborative coworkers than real-time assistants.

AI in Software Development

AI-Generated Code Dominance
Exploring the impact of AI on software development, Kevin predicts a significant shift towards AI-generated code:

"Five years, 95% is going to be AI generated. I think very little is going to be line by line is going to be human-written code."
— Kevin Scott [28:48]

Despite the rise of AI-generated code, he asserts that human authorship remains crucial for high-level decision-making and problem-solving in software engineering.

Redefining Authorship in AI-Driven Development
Kevin discusses the elevated abstraction levels in programming with AI, drawing parallels to modern development environments:

"It's like raising the level of abstraction. Like we are changing the interface that the programmers use to communicate to the machine."
— Kevin Scott [29:17]

This evolution allows developers to focus more on design and functionality rather than boilerplate code generation.

Engineering Teams and Technical Debt

AI to Mitigate Technical Debt
Addressing the perennial challenge of technical debt, Kevin shares Microsoft's initiative to leverage AI for its reduction:

"I think we can turn this very zero-sum problem of tech debt accumulation into something non-zero sum."
— Kevin Scott [32:11]

He elaborates on a research project aimed at eliminating technical debt at scale, highlighting AI's potential to streamline engineering workflows and enhance productivity.

Personal Reflections on Leadership and Technical Debt
Kevin candidly reflects on his struggle with bureaucratic tasks, admitting:

"I'm super impatient with bureaucratic things. Like, I hate budgets and facilities and like all of the mechanical parts."
— Kevin Scott [36:36]

This honesty underscores the human aspect of tech leadership and the challenges of balancing innovation with administrative responsibilities.

Quick-Fire Insights

Respecting Competitors
When asked about competitors, Kevin mentions his respect for Anthropic and their leader Dario:

"If I got to pick one, maybe Anthropic. Dario's doing a good job."
— Kevin Scott [35:10]

Best Advice Received
Kevin shares impactful career advice about focusing on strengths:

"The best you're ever going to do if you're an idiot at something is to get mediocre at it. And all of the time that you spend trying to get to mediocre, you are not spending doing the things that you're a genius or very good at."
— Kevin Scott [35:16]

Lessons from Satya Nadella
Highlighting leadership principles, Kevin praises Satya Nadella for balancing energy and clarity:

"His core leadership principle is that you have to simultaneously for people, create energy and you have to produce clarity."
— Kevin Scott [37:06]

China's AI Capabilities
Addressing geopolitical concerns, Kevin affirms the prowess of Chinese AI talent:

"We should really, really, really respect the capability of Chinese entrepreneurs, scientists and engineers."
— Kevin Scott [38:05]

AI in Health Diagnostics
Kevin makes a groundbreaking prediction about AI's role in healthcare:

"It is already the case that I think the frontier models are probably better health diagnosticians than your average GP is."
— Kevin Scott [38:39]

Speed of AI Development
On the pace of AI advancements, Kevin admits:

"No." (in response to whether we're going fast enough)
— Kevin Scott [39:43]

He advocates for accelerated investment in education and the widespread deployment of AI tools to harness their full potential.

Conclusion

The episode concludes with Harry expressing his appreciation for Kevin's insightful contributions and highlighting the transformative potential of AI in various sectors. Listeners are encouraged to watch the full episode on YouTube and stay tuned for upcoming segments.

Notable Quotes

• Kevin Scott [04:51]: "This is exactly the thing that happens at the beginning of every big technological paradigm shift and every new cycle that's driven by it."

• Kevin Scott [07:45]: "They're super valuable. But they're only valuable to the extent that you can connect them to things that users need via product."

• Kevin Scott [10:53]: "I believe we will get to some point where we'll hit a scaling asymptote."

• Kevin Scott [28:48]: "Five years, 95% is going to be AI generated. I think very little is going to be line by line is going to be human-written code."

• Kevin Scott [38:39]: "It is already the case that I think the frontier models are probably better health diagnosticians than your average GP is."

Final Thoughts

Kevin Scott's conversation on this episode of 20VC offers a profound exploration of current and future AI trends, emphasizing the critical balance between technological advancement and practical application. His perspectives provide valuable guidance for investors, entrepreneurs, and technologists aiming to navigate and leverage the rapidly evolving AI landscape.

For more episodes and resources, visit www.20vc.com.