Earlier this week, I attended the AGI Builders Meetup in San Francisco. The event had 5 product demos ranging from new AI features by Twilio to a YC S24 startup building an AI agent to handle calls on behalf of users.
These demo-based events are always helpful for me to track the latest in AI. Also ended up chatting with a few engineers and founders attending the event, to get their thoughts on what they are seeing in their respective domains within AI.
One key insight I got from this event was that LLMs aren’t the real future of AI. No one really knows what’s going on inside them. They hallucinate much more than desirable (especially for accuracy-driven enterprise use cases). They are prone to prompt injection hacking.
In fact, the presenting Twilio PM said that “working with LLMs is like getting toddlers to do something”. They don’t take instructions promptly. You don’t know what’s going on inside their heads. You have to proactively ensure their safety as they are doing a task. I found this framing really interesting.
Building on this further, I have an updated (but still working) POV on the 3 buckets of AI – infra, applications, and dev tools.
#1 Infra
Irrespective of where AI goes, hardware infra like GPUs will always be needed. Hence, it makes sense that Nvidia is doing so much capex.
Also, Big Tech software infra players (Microsoft, Google, Meta, AWS, etc.), as well as AI-native hyperscalers (OpenAI, Anthropic, etc.), will continue playing a key role in defining where AI goes from here.
Sadly, as a micro VC, I can’t play much in this bucket (except personally investing in the public markets).
#2 Applications
Application layer founders that are starting up today are leveraging the capabilities of AI from Day 0 to solve customer problems. Their core focus still remains commercial-first – using the best-available software capabilities to solve customer problems, rather than getting overly enamored by the research aspects of AI and where it’s headed.
In a sense, these startups are centered on customer problems, not AI per se. Wherever AI ends up going, LLMs and beyond, these founders will leverage whatever capabilities they can get their hands on, and modify their architectures accordingly.
As an investor, the key is to back founders who are starting up now with an AI-first mindset and therefore, are fresh enough and agile enough to keep evolving their software as the underlying AI capabilities evolve.
Therefore, it’s reasonable to expect that these AI-native application startups should be fairly resilient to changes in the overall AI landscape. Hence, I feel reasonably comfortable in backing them (eg. portcos like Confido Health, Loop, and Soulside).
#3 Dev tools
This is the bucket I am most confused (and concerned) about. From seeing these demos, it seems like dev tools startups are essentially using the mental models of the previous cloud & mobile waves to make assumptions on use cases.
Further, I have observed that many of them are solving short-term, immediate pain points that could easily become irrelevant due to where AI goes from here, and/ or from the competition (eg. open source alternatives, AWS quickly launching it as a feature, etc.).
As I was seeing these demos, I looked up how much capital some of these companies had raised. Many of them have raised anywhere from $10-35Mn. The capitalization of these companies seems out of sync with the durability of their underlying use cases and revenue.
Essentially, what all this means is that I have a macro “Why Now?” question around the AI dev tools bucket. A top Bay Area engineer who recently left a cushy Big Tech job to start up was recently saying – “Given how things are changing every month, I am really not sure what to build right now”. I feel this is an intellectually honest view, rather than a FOMO-based approach that many VCs are taking.
Based on what AI practitioners like this person are saying about the exponential rate of change in AI, I fear that a majority of these dev tool use cases won’t endure.
Again, this is just my working POV. Would love to hear your views on what you are seeing.
Subscribe
to my weekly newsletter where in addition to my long-form posts, I will also share a weekly recap of all my social posts & writings, what I loved to read & watch that week + other useful insights & analysis exclusively for my subscribers.