By playing, you agree to uListen's Terms and Privacy Policy and YouTube's Terms of Service.
The Twenty Minute VCKevin Scott, CTO @ Microsoft: An Evaluation of Deepseek and How We Underestimate the Chinese
Summaries, chapters, transcripts, and FAQs on this page are made by uListen, not YouTube.
At a glance
WHAT IT’S REALLY ABOUT
Microsoft CTO Kevin Scott Explains AI’s Future, Agents, And China
- Kevin Scott, CTO of Microsoft, argues that AI is in the early stages of a major platform shift, where true, durable value will accrue to products that solve real user problems rather than to models or infrastructure alone.
- He rejects the idea that we are near scaling-law limits, expecting significant further gains from better data, algorithms, and especially inference optimization, while emphasizing that high‑quality, reasoning-focused data is far more valuable than raw web tokens.
- Scott believes the dominant interaction model will shift from chat interfaces to persistent, domain‑specific agents with memory, deep domain product management, and asynchronous workflows, fundamentally changing software development and product building.
- He highlights China’s AI capabilities as widely underestimated, sees frontier models already outperforming average doctors on diagnosis, and calls for massive investment in education and deployment to turn AI’s capabilities into broad societal benefit.
IDEAS WORTH REMEMBERING
5 ideasPrioritize products that solve real problems, not just building models.
Scott stresses that models and infrastructure only capture value when connected to user needs through great products; entrepreneurs should ship, iterate, and be brutally honest with data rather than falling in love with technical artifacts.
Exploit high‑quality, reasoning-centric data and expert feedback over raw scale.
He notes that carefully curated data and expert human feedback, especially in post‑training, can be amplified into much more valuable training signals than undifferentiated web data, particularly for reasoning rather than fact recall.
Expect inference costs to keep falling as software optimization compounds.
DeepSeek R1 is framed as just one point on a long line of price/performance gains driven mainly by software and systems work, meaning larger, more capable models can still get cheaper to use over time.
Build domain‑specific agents with memory and asynchronous workflows.
Scott predicts many specialized agents rather than a single general one, with better memory, personalization, and the ability to handle long‑running, delegated tasks—more like a coworker than a chat bot.
Leverage AI to reshape software engineering and reduce tech debt.
He expects ~95% of new code to be AI‑generated within five years, with humans focusing on higher‑level authorship and system design; AI can also systematically attack technical debt, turning a zero‑sum tradeoff into a solvable problem.
WORDS WORTH SAVING
5 quotesModels aren’t products. The only thing that really matters is making good product.
— Kevin Scott
I can very clearly see what we're doing now and what we're doing next, and I don't see the limit to the scaling laws.
— Kevin Scott
You don't want the thing that's just a good email summarizer. You want something you can delegate increasingly complicated tasks to, the same way you would to a coworker.
— Kevin Scott
Ninety‑five percent of net new code is going to be AI‑generated.
— Kevin Scott
We should really, really, really respect the capability of Chinese entrepreneurs, scientists, and engineers. They are very good.
— Kevin Scott
High quality AI-generated summary created from speaker-labeled transcript.
