Value in AI models, Eliminating tech debt, Hitting a scaling asymptote and Creating co-workers [20VC, March 2025]
Today’s issue features Microsoft’s CTO. A heavy, super important hat worn by a humble guy from rural central Virginia who worries about the world’s inadequate access to high quality healthcare.
Dear reader who doesn’t have time to listen to VC podcasts,
Today’s issue features Microsoft’s CTO. A heavy, super important hat worn by a humble guy from rural central Virginia who worries about the world’s inadequate access to high quality healthcare.
Kevin Scott said “I don’t know” 5 times during the episode. But far more often, he shared what he knows, predicts and wishes for.
Scott leads Microsoft’s AI and technology strategy at global scale and he played a pivotal role in Microsoft’s partnership with OpenAI. 20VC’s Harry Stebbings hosted Scott on March 31.
Enjoy🍿
[Time saved: 40 minutes]
N.B.
- It’s All Confusing Right Now And That’s Normal: We’re living through a big tech paradigm shift and at the beginning of every new cycle things get less clear. Remember the early days of internet? The early days of mobile? Confusing times.
- Ada’s Paradigm Is Over: Since Ada Lovelace's first program 200 years ago, computing has followed a single paradigm. AI is now transforming this by enabling users to interact with devices without programming knowledge. AI agents understand user intentions and execute tasks accordingly, eliminating the need for product teams to anticipate user expectations.
- Don’t Get Swept Up in the Technical Bits: The key focus should always be on creating a quality product. In the early stages of development, technical individuals often lose sight of this priority amidst other tasks and challenges.
- Are Models Valuable?: They are but your product is the most important. Most value lies in your product. Infrastructure isn’t being built for its own sake; it’s built to serve people who wants to make a product.
- Reasoning Over Information: While your data may seem valuable, its actual utility in building effective models is difficult to determine. Models aren't merely collections of facts; they must be able to reason with that information in ways that provide practical value to end users.
- From A Missing Memory To A Co-Worker: Current AI agents lack robust memory, making their interactions transactional. He anticipates significant improvement in memory capabilities within a year, enabling agents to recall past interactions. Also, Scott believes that we’re progressing towards the user delegating increasingly complicated tasks to the agent (like to a co-worker), where the agent the deliver outcomes over time rather than instant responses.
- Not One-Agent-For-Everything Theory: Engineers will continue developing the underlying infrastructure, but the user-facing interaction will shift to AI agents. Scott envisions a future with numerous agents, supported by product managers who are domain experts in specialized fields like medicine or venture investing. These experts will design feedback loops to enhance the agents' ability to assist end users effectively.
- No More Tech Debt: Tech debt is a painful trade-off for Scott (he describes it as his “mortal enemy”). While launching quickly often means postponing technical fixes, this unresolved debt eventually causes problems. Scott believes AI could eliminate the need for such trade-offs, enabling more balanced development without accumulating tech debt.
- Hitting A Scaling Asymptote: Scott doesn't believe that AI will keep scaling endlessly into "weird territory". He believes there will be a limit—a scaling asymptote—where advancements in AI lead to diminishing returns. Beyond this point, the cost of further improvement may outweigh the value it brings, especially if its utility for users isn’t clear.
- Microsoft Doesn’t Have Enough Imagination: No single company, not even giants like Microsoft, can fully grasp or imagine every potential innovation. A vibrant ecosystem with diverse contributors exploring areas of value is both essential and exciting. Additionally, tools, infrastructure, and platforms are currently more affordable, accessible, and user-friendly than ever before, creating unprecedented opportunities for exploration and creativity.
Quotable moments””
- Launch, Gather, Iterate, Be Brutal: “Launch stuff, gather data, iterate and be super, super brutal with your own self about what you're seeing.”
- Open-Source: “You’re going to have lots of open-source infrastructure products.”
- Who Will Create New Code In 5 Years?: “95% is going to be AI generated. That doesn't mean that the AI is doing the software engineering job. I think the more important and interesting part of authorship is still going to be entirely human.”
- Hype Around Agents?: “Usage always follows utility. You make useful things, they get used a lot. Clearly, with software development agents, we're getting a lot of adoption right now.”
- On DeepSeek: “We should really, really, really respect the capability of Chinese entrepreneurs, scientists and engineers. “Oh my God, this is coming from China?” – that shouldn’t have been surprising.”
- Better Than Your Average GP: “The frontier models are probably better health diagnosticians than your average GP is. It's a good thing to realize and act on as quickly as possible because we have a whole world of people who have inadequate access to high quality healthcare, including my own family in rural central Virginia where it's just not good.”
- In Scott’s Ideal World: “The thing that I would want [to see] is that we invest super heavily in education. I would love to see every child feel as if these new tools that we're building right now are for them, accessible to them, expressly built for them to go accomplish the things that they think are most important.”
- Lessons From Working with Satya: “His core leadership principle is that you have to simultaneously, for people, create energy and you have to produce clarity.”
- The Best Advice Scott Received: “The mistake that people always make is they focus on trying to improve at the things that they're worst at.”
And that’s 45 minutes squeezed into 5. Full episode here.
Take care,
Dominika
For those of you who can find the time to tune in, here’s a selection of my recent finds:
- 7 Business Ideas from a Unicorn Founder by My First Million (March 27, 56 minutes) (Spotify)
- Why AI Voice Feels More Human Than Ever by a16z Podcast (March 18, 43 minutes) (Spotify)
- Snapchat CEO: Exact Formula Used To Build A $130 Billion Company! by The Diary Of A CEO with Steven Bartlett (March 24, 148 minutes) (Spotify)
- VC React Podcast (E31): Consolidation, Investing In Hardware, Traitorous Eight-MENA Version & More! by VC React (April 12, 53 minutes) (Spotify)
- Trump's tariffs swift impacts on the global supply chain by StrictlyVC Download (April 5, 21 minutes) (Spotify)
Find out what you've missed out on from the most popular VC podcasts – hit the Subscribe button!
Disclaimer: This article was written by a human with some help from AI.