- Shuo's Snippets
- Posts
- Growing number of startups with 95% AI-generated codebases
Growing number of startups with 95% AI-generated codebases
3 trends, 2 theses and 1 tool from Shuo
Hello friends!
This is Shuo from IOVC. For the past while, I’ve been writing “Shuo’s Snippets” to document what’s new and next in startups and tech.
I’ve included you because some of the themes I’ve been thinking about align with discussions we’ve had, and you’ve made me a better and smarter person – thank you.
This is my way of sharing notes and sparking discussion, so feel free reply anytime – I’d love to learn what you’re seeing. No hurt feelings if you opt-out!
So, here’s what I’ve been seeing this past month investing in fractional founders* as well as teaching entrepreneurship at Berkeley and Stanford:
📈 3 trends in startups/tech/venture
🤔 2 theses on what’s next
🔧 1 tool I love
*a fractional founder is an entrepreneur who is transforming their part-time project into their full-time startup
💌 Someone shared this with you? Connect with me and join the email list here.
3 trends in startups/tech/venture
🚀 Growing number of startups with 95% AI-generated codebases
At pre-seed, 50%+ of startups I’m seeing have 95% AI-generated codebases. By seed (Y Combinator), a quarter still have 95% AI-generated codebases.
Startups are using AI for more than just coding. There are now AI models optimized for every business task.
Startups use 1) OpenAI's GPT-4.5 for coding, drafting emails, generating ideas, and research, 2) Anthropic's Claude 3.7 Sonnet for creative writing with consistent style, and 3) Google's Gemini for input across documents.
🤖 (Re)creating AI models for the cost of a nice meal
Researchers have been taking advantage of Huang’s law and learning from existing models to (re)create them for much cheaper.
Berkeley researchers recreated DeepSeek’s R1 model’s capabilities for $30. Stanford did the same, but for OpenAI’s o1 model for $50.
💰 Increasing concentration within venture (and public equities)
Diversification is the core of any successful and replicable investment strategy.
The S&P 500 is diversified, but it’s increasingly less so, with 36% (historic high) of its total index market cap accounted for by just the top 10 stocks.
Venture is also becoming more concentrated, with 9 venture funds collecting 50%+ of all money raised by US funds last year.
2 theses on what’s next
📦 It's not about innovation. It’s about distribution.
I'm looking for startups with distinct growth strategies — that 1) differentiate their products from general (and rapidly improving) AI models, and 2) land paying customers efficiently.
🕒 Venture investing will go even earlier
With founders increasingly building in public, the investors that generate the most alpha will be the investors that can find founders who are early enough that they have not yet even started sharing publicly.
We are riding the wave by focusing on “fractional founders”—entrepreneurs who are so early that they are still building part-time and have not yet even quit their full-time jobs.
Check out my notes about other trends in venture.
1 tool I love
✈️ AI for simulations
Rescale helps engineers and researchers build and scale simulations — for institutions including Microsoft, NASA and Nvidia.
Learn more about how Rescale helped Boom break the sound barrier.
What’s top of mind for founders?
Founders have been asking me a lot about valuation benchmarks. You can hear my latest thoughts below 👇🏼
Please hit “reply” with any thoughts and reactions, and stay tuned for more on what’s new and next in the coming month!
Cheers,
Shuo