Digital release of Engineering Executive's Primer. @ Irrational Exuberance

Hi folks,

This is the weekly digest for my blog, Irrational Exuberance. Reach out with thoughts on Twitter at @lethain, or reply to this email.

Posts from this week:

- Digital release of Engineering Executive's Primer.
- Thesis on value accumulation in AI.

Digital release of Engineering Executive's Primer.

Quick update on The Engineering Executive’s Primer. The book went to print yesterday, and physical copies will be available in March. Also, as of this moment, you can purchase the digital edition on Amazon, and read the full digital release on O’Reilly. (You can preorder physical copies on Amazon as well.)

Thesis on value accumulation in AI.

Recently, I’ve thinking about where I want to focus my angel investing in 2024, and decided to document my thinking about value accumulation in artificial intelligence because it explains the shape of my interest–or lack thereof–in investing in artificial intelligence tooling. I’ll describe my understanding of the current state, how I think it’ll evolve over the next 1-3 years, and then end with how that shapes what I’m investing in.

My view on the the state of play today:

1. There are three fundamental components: Infrastructure (cloud providers, NVIDIA, etc), Modeling & Core (OpenAI, Anthropic, etc), and AI-enhanced products (Github Copilot, etc)
2. Today there’s significant value being captured in the Modeling & Core layer, and many new companies attempting to compete in that tier. Valuations in this tier are extremely rich at this point
3. Infrastructure hasn’t captured too much value, except for NVIDIA who arguably should be split into their own bucket of “hardware” instead of lumped in with cloud providers. Cloud vendors have the scale of physical resources to participate in AI, but generally don’t yet have strong offerings. However, these companies have a structural advantage in preexisting legal contracts with companies to govern API and data usage, along with the economy of scale to rapidly grow these businesses once they find product-market fit in the AI segment
4. AI-enhanced product has relatively few sophisticated entries today. There’s a lot of handwaving and loud statements, but very few companies that have proven out their AI-enhanced products are meaningfully better than preexisting alternatives. I think this is a matter of time rather than exceptionally difficult, so we will see more value accumulate here.

However, I think this is a transitory state. Where I see things moving over the next several years (and generally I think the transition here will be faster than slower):

1. I believe Infrastructure will eat an increasingly large amount of the Modeling & Core tier. Even today, the cloud providers of the Infrastructure have significant ownership and control in the leading Modeling & Core tier. This will make it harder to perceive the shift, but I think it’s already happening and will accelerate
2. Because I believe AI-enhanced product will successfully capture value thoughtfully using AI, the interesting question is what sorts of products will capture the majority. Ultimately I think the question is whether it’s harder to get the necessary data to power AI (fast-moving incumbents capture majority of value) or whether learning to integrate and integrating products with genuinely useful AI-capabilities is harder (new challengers capture majority of value)

There’s no interesting way to invest in the Infrastructure tier in 2024 (the main players are all public at this point), and I think the Modeling & Core tier is shrinking (and largely over-valued by interest from folks with a different thesis on value accumulation), which means that the interesting place to angel invest in 2024 is, in my opinion, in products that are well-suited to adopt AI-capabilities. That’s a broad category–we’re still learning where these techniques are powerful–but I think it’s particularly any company that works heavily with documents, and any company where it’s product is capable of keeping a human in the loop (e.g. LLMs are cheap, fast and imperfectly accurate, but in a system where someone uses it to draft replies and review them by a human, you’d be fine).

Not angel-investing, but if you wanted to make a career bet, I think the interesting career bet is finding an established company with significant existing data and product workflows that could be enhanced by recent AI advances.

That's all for now! Hope to hear your thoughts on Twitter at @lethain!

This email was sent to you why did I get this?    unsubscribe from this list    update subscription preferences Will Larson · 77 Geary St · co Calm 3rd Floor · San Francisco, CA 94108-5723 · USA