Navigating AI’s iPhone Moment

A venture perspective on LLMs and what’s next…

The news today is full of the latest generative AI wonders. Twitter is packed with videos of ChatGPT powered robots planning meals, ordering groceries, writing API calls, generating CSS for landing pages, social media drip campaigns…you name it. One example even showed how auto-generated emails were sent to thousands of prospects, proposing customized software solutions to each one, and generating fantastic response rates that left the senders puzzling over how to deliver solutions fast enough.

An iPhone moment. In the words of the NVIDIA CEO, we are living through the “iPhone moment” of AI. The point he’s making is this: now that foundational AI models are readily available to everyone behind drop-dead simple UX (ChatGPT), the entire planet is suddenly experimenting with what these tools can do. Just as the iPhone put the power of semiconductors and the intuitive interface of a touchscreen in the palms of a billion hands, these natural interface AI models are suddenly putting the power of AI (generative and otherwise) in a billion hands. Mobile ate the world, software (and SaaS) ate the world, and now AI will eat the world as well — but what does that actually mean?

With that question echoing in our minds, we at Angular continue to review and process near-record numbers of opportunities to invest in seed-stage companies. Perhaps surprisingly (given the market conditions), the pace of seed-stage company formation seems to be higher than ever (our internal data shows that only 4Q22 was higher).

A dangerous time to start something new. We are investing right now (working several deals this week), but we also must admit that this feels like a pretty dangerous time to invest. The world is changing extremely fast right now — and while the magnitude of the change is clear, the shape of the change is less obvious. Here are a few observations that David and I have had recently:

• The data/AI stack. As we have for years, we continue to be excited by companies that are exploiting deep fundamental insights to provide foundational tooling for data processing and handling, including in ways that can power the essential “AI stack,” but not exclusively. Some of these might be relevant for generative AI, but others are more about achieving a 10–100x improvement in performance, ease, or reliability of basic IT capabilities. Our view is that the AI’s iPhone moment makes these companies more valuable than ever. Companies like Firebolt, Paradime, Tensorleap, and Valohai are in this category, as are a few we can’t talk about yet.
• Deep vertical expertise. We are also very excited about companies that leverage deep vertical expertise. Sometimes these companies benefit from advances in AI, and sometimes not. AI is just one of many inputs in their value chain, but their ability to capture value depends mostly on understanding industry workflows and sales processes. These companies are rarely threatened by advances in AI. Companies like Aquant, Crux, Fixefy, Planable, and Vault are in this category.
• Thick UX. Our thesis that “basic” AI models are commoditizing led us to invest in Levity, a company that is building a complex and powerful UX/UI layer on top of basic AI capabilities. The more powerful the foundational models Levity can leverage, the more valuable Levity can become — but the value and power of Levity is in a robust UX/UI layer — not the AI itself. Again, that’s just an input.
• Thin layers. We worry about companies that appear to be providing a very thin layer on top of a foundational model. We’ve seen a ton of these popping up lately. Where there isn’t a complex, powerful, or deeply industry-aware workflow to build, we wonder if these companies can really create enough sustainable value.
• Generative value is illusory. We worry about companies that are creating value by “generating” anything: code, summaries, images, emails, marketing text, etc. That capability is going to get commoditized pretty fast (we believe it already has), and our prevailing view is that something that costs nothing to create is going to be worth nothing to consume. When our email inboxes are full of AI-generated emails, we just won’t read them. That has interesting social implications — but I’m not sure the business implications are that interesting. Spam is spam, and more spam is even more spammy.
• Devaluation of code. We also worry increasingly about companies that are creating value primarily by helping human software developers write code faster and better. Our suspicion here is two-fold. First, it’s increasingly clear that coding “copilot” tools are going to be adopted very rapidly. We wouldn’t be surprised if the majority of code that gets written next year begins in a copilot authoring environment. Second — and perhaps ironically — we suspect that software development itself is one of the first professions to be most fundamentally upended by generative AI. The key issue here is testing…(see the next bullet).
• Testing, testing, 123. One area where we are actively looking but have not yet found the right company is software testing. Software testing is an incredibly important and expensive activity — that does not appear to have been fundamentally “solved.” To date, most companies have pursued one of two paths: Some companies have invested a ton of money and effort in manually-generated testing scripts, which create complex management problems of their own. Other companies have essentially given up on testing (QA) as a separate discipline and have attempted to move to some version of “test in production,” with varying degrees of success. Modern AI capabilities should enable the automatic generation, deployment, and evaluation of test scripts in realtime. This insight is not new: thoughtful entrepreneurs in the space have been talking about this for years. But so far, no one has managed to pull it off. It’s much harder than it sounds for a number of reasons that are very technical — but if you are working on this or think you’ve cracked it, we’d love to chat…

As I wrote above, we know the pace of change is faster than ever, and we know the magnitude of change is immense. We are eager to meet with and invest in those teams that have a strong opinion on how that change will manifest itself and how — in a world of increasing commoditization of so many things that were once very challenging — a given company has a chance to build barriers to entry that will last long enough to support a valuable business until a large outcome. Ultimately, while technology is constantly evolving, the true essence of building a technology business remains constant: building meaningful barriers to entry — true defensibility — around significant customer value. The creation of customer value itself is essential, but insufficient on its own. The business is in the barriers.

If you are building in this space with a thesis around defensibility, let’s talk.