Mar 31, 2025 · 47m · 20vc
Kevin Scott, CTO @ Microsoft: An Evaluation of Deepseek and How We Underestimate the Chinese · 20VC with Harry Stebbings
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this deep-dive interview, Microsoft CTO Kevin Scott joins host Harry Stebbings to discuss the future of artificial intelligence, sharing insights on scaling laws, the transition from raw models to product-focused value, the rise of specialized autonomous agents, and the democratization of software development.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Harry holds 15.5% of the talking time here. How this is scored →
Harry as informed peer 4.1 Guest teaching 3.8 Guest disagreement 1.4 Harry pushing back 3.3
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 2:22 Emphatic rejection of sitting on hands
When Harry asks if investors should sit on their hands during transitional confusion, Kevin bluntly interrupts with 'Oh god, no' and forcefully argues against passivity.
Hardest push from Harry ▶ 24:52 Challenging lack of agent lock-inHarry interrupts Kevin's enthusiasm for software agents to directly challenge their economic value, pointing out that developers admit there is zero lock-in.
Biggest teaching moment ▶ 19:21 200-year history of computing paradigm reframeKevin reframes Harry's simple question about chat interfaces by schooling him on the fundamental shift in human-computer interaction since Ada Lovelace.
Harry holds his own ▶ 30:02 Pessimists are right and optimists make moneyWhen Kevin claims there is no prize for pessimism, Harry instantly fires back with a Collison brothers quote defending the analytical accuracy of pessimists.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| Trailer Hook: The Best Time to Be Alive | 2 | 1 | 0 | 0 | Harry sets up a classic venture capital framing on identifying sustainable value creation during tech shifts. Kevin responds warmly with historical parallels regarding early internet and mobile eras. | |
| Pragmatic Action & Focus on Product over Models | 2 | 3 | 2 | 1 | Harry asks if investors should sit on their hands or act during confused transition periods. Kevin forcefully rejects the passive option and explains why models are not products. | |
| The Relationship Between Models, Infrastructure, and Products | 5 | 4 | 1 | 4 | Harry challenges Kevin's 'models aren't products' stance by citing Cerebras CEO Andrew's view that compute holds the true value. Kevin reframes how infrastructure and platform layer value connect to end-user products. | |
| Scaling Laws and the Future of AI Capabilities | 5 | 5 | 2 | 5 | Harry directly confronts Kevin on his contrarian dismissal of scaling law asymptotes. Kevin breaks down reasoning tokens versus factual recall tokens to explain why capability scaling continues. | |
| Inference Efficiency and the DeepSeek R1 Public Reaction | 4 | 4 | 2 | 3 | Harry probes Microsoft's internal reaction to the DeepSeek R1 launch. Kevin downplays the breakthrough as part of an ongoing price-performance curve and reveals Microsoft held back superior models. | |
| Open Source vs. Proprietary AI and the Evolution of Pragmatism | 3 | 4 | 0 | 1 | Kevin discusses his personal evolution from open-source zealot to pragmatic executive. He uses historical search engine market structure to illustrate how open and closed AI models will coexist. | |
| Beyond Chat: A New Paradigm of Human-Computer Interaction | 4 | 6 | 1 | 2 | Harry questions whether chat interfaces are merely a temporary default UI. Kevin provides a broad historical overview tracing computing paradigms back 200 years to Ada Lovelace. | |
| The Rise of Specialized Agents and Domain-Expert Product Managers | 6 | 4 | 3 | 7 | Harry pushes back firmly on the defensibility of AI developer agents, noting developers report no switching lock-in. Kevin counters with a search engine comparison to explain user retention. | |
| The Evolution of Agents: Memory, Asynchronous Action, and Optimism | 6 | 3 | 2 | 6 | Kevin criticizes tech pessimists by asserting there is no prize for pessimism. Harry counters immediately by quoting the Collison brothers on pessimists being right while optimists make money. | |
| AI-Generated Code and the New Abstraction of Programming | 4 | 5 | 1 | 2 | Kevin predicts 95% of future code will be AI-generated. He quizzes Harry on whether he is a programmer to explain how AI represents the next layer of abstraction in software engineering. | |
| Eliminating Technical Debt at Scale with AI | 4 | 4 | 2 | 5 | Harry presses Kevin on where Microsoft has been slow despite its massive resources. Kevin explains Microsoft Research initiatives aimed at eliminating technical debt at scale using AI. | |
| Quick-Fire Round & Final Thoughts | 4 | 3 | 1 | 3 | In the quick-fire round, Harry turns Kevin's histogram metaphor back on him to ask about his personal weaknesses. They also cover China's AI talent and medical diagnostic models. |
Statements from this episode (19)
Scott: AI agents will become less transactional and session-oriented
“The agents, they will definitely be less transactional, less session-oriented going forward.”
Scott: AI's economic value will accrue to products, not infrastructure
“But most of the value has to be in the products. Like, you know, we don't build infrastructure just for the sake of infrastructure. Like we build infrastructure so people can make product.”
Scott: AI scaling laws have not hit a limit yet
“I can very clearly see what we're doing now and like what we're doing next. And I don't see the limit to the scaling laws. Like if you're just sort of thinking about the raw capability of the models and like how well you can condition them to reason over incre…”
Scott: Claims about proprietary data's value for AI lack scientific backing
“Most of those assertions that people make are like most of the assertions that people make are just unfounded in any kind of science. Like no, no one's gotta, and like the measurements we do have show that there's a pretty big disconnect around what some peopl…”
Scott: Using AI models as factual databases is inefficient and expensive
“People who think of models as repositories of factual information, and they're treating them like the world's worst and most expensive databases. Like it's not, not super useful.”
Scott: Reasoning models require different training tokens than fact recallers
“What you want models to be able to do is to be able to reason over information. So like if you give them access to information, like how well are they able to reason over a set of information to go do something that's useful for you? And so you just need diffe…”
Scott: Software optimizations drive AI inference efficiency far more than hardware
“A little bit of that is because you get, you know, maybe like a two X benefit price performance from hardware every generation. Like if you're lucky but you get a much bigger improvement price performance wise from. All of the things that you're doing in the s…”
Scott: Microsoft chose not to launch models more interesting than DeepSeek-R1
“We've had models more interesting than Deep Seek R-One that, like, we didn't even We chose not to even launch them.”
Scott: Search economics flow entirely to large integrated infrastructure operators
“All of the economics in search go to like somebody who's stood up a gigantic infrastructure and who's sort of running like a whole. Search business like with its own feedback loop.”
Scott: AI will render traditional product engineering teams obsolete within ten years
“It basically means, and like, I don't think this is next year, but it's probably not going to be 10 years. This whole notion that the, That, that you have teams of people who are, who, whose job is to go anticipate a bunch of very granular user needs in some n…”
Scott: Future AI user interfaces will consist of many specialized agents
“The user interface that surfaces, those capabilities will probably be agents. And, you know, product managers, like I don't all don't believe in this, like one agent. For everything sort of theory. I think you'll have a lot of agents.”
Scott: Future AI product managers must be deep domain experts
“Your product managers are probably going to have to be domain experts, like people who sort of deeply understand something like medicine or, you know, drug discovery or early round venture investing, or, you know, like, you know, just sort of pick your thing.”
Scott: AI agent memory will improve significantly over the next year
“I think one of the things is going to happen because I know lots of people are working on it right now is memory is going to get a lot better over the next year or so, which means that as you're using an agent, like, and it remembers more and more about your p…”
Scott: AI agents will execute tasks asynchronously within the next year
“I think there's going to be more over the next year of you sort of dispatching your agent to go do something and it like goes and works while you are not paying attention to it.”
Scott: 95% of net new code will be AI-generated within 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.”
Scott: Microsoft Research launched an initiative to eliminate technical debt using AI
“There's a big research initiative. We've started at Microsoft research about a year ago where like the whole mission of the lab is like eliminate tech debt at scale using these new AI tools.”
Scott: The gap between AI capability and utilization is growing larger
“Like, I think honestly right now there's a bigger gap than there was even two years ago between what the most capable frontier models can do and what they're being used for.”
Shorts cut from this episode
▶ The Genius Strategy Everyone Gets Wrong · 20VC with Harry St (@39:25)
▶ "This advice stuck with me” 😮 · 20VC with Harry Stebbings (@39:25)
▶ Why We’ll Have Multiple AI Agents 🤖 · 20VC with Harry Stebb (@0:13)
▶ Is this the biggest mystery in AI? 🤖 · 20VC with Harry Steb (@11:13)
▶ “AI Models Aren’t Products” 🤔 · 20VC with Harry Stebbings (@3:00)
▶ “Now is the best time to be an Entrepreneur” 📈 · 20VC with (@0:00)