Correctly Excited
A brief note before we begin - my heart goes out to anyone who may have had family or friends affected by the attack on Israel this past weekend. May peace be with you and that this conflict is resolved as quickly and justly as possible.
Ideologically I am opposed to betting against innovation, which makes sense given my job. It would be weird if you ever met a VC that consistently said, “Yeah I don’t buy the hype.”
Not to say that hype in and of itself is a good or bad thing, but more to say that if you’re a VC you’re going to have to be optimistic about the future, you’re going to have to be optimistic about commercial breakthroughs, and you’re going to have to (at some point) be excited about hype. As such, it seems to me that the earlier you are correctly excited about hype, the better off you are in the long run.
Case in point, see Instacart’s recent IPO.
Being correctly excited is easier said than done, but one of the things that I have observed about the history of hype (I mean innovation) is that a fair amount of breakthroughs are distributed via written essays. In 1989, the idea of the internet was published and distributed by Tim Berners Lee in Information Management: A Proposal. In 2008, the Bitcoin Whitepaper was released, and in 2017 Google released the seminal research paper Attention is All You Need.
There’s a direct through line between these papers and literal trillions of dollars of value creation. In all three cases, buying the hype early on would have paid massive dividends. And ideologically, I believe it’s going to happen again. As I’ve referenced many times in this blog/newsletter history doesn’t repeat, but it often rhymes. And I would expand on that further to say that innovation doesn’t spontaneously combust, it’s a longitudinal process built on the shoulders of giants.
It’s iterative and messy and has fits and starts. Craigslist walked, so Myspace could run, so Facebook could blast off into orbit, so TikTok could slingshot to Mars.
We’ve gone from a static internet in the early days, to a responsive internet in the mid aughts, to the globally dynamic, hyper personalized, on-demand, internet of today. Internet based businesses have been lauded as engines for social and economic advancement, and derided for prioritizing profits over safety. In either case, the internet transformed society and the idea of the internet was first conceived in an essay.
If history is any guide, the importance of the Attention is All You Need Paper is going to be revealed to us in the coming decade. We could be headed toward a global utopia, a barren machine ruled hellscape, or an incrementally improved version of today; I don’t know where we will end up on that spectrum, but just like many of you I am trying to figure out how, where, and when to be correctly excited about AI.
As such, I spent parts of this past weekend thinking about and researching startups that are building in AI. Here are some general thoughts and what I found:
Today, I view language models like an engine. Fuel (i.e data) goes in one end, the engine (i.e the model) transfers that fuel into energy (i.e an output). Both the quality of the fuel we feed the engine, and the independent quality of the engines are being improved
In theory, the tens of billions of dollars invested into these companies should lead to better engine performance
Better engine performance means increased productivity, which means greater aggregate wealth levels for a society. We will be able to create (and consume) more, with less
However, I suspect the distribution of the wealth creation will be highly uneven (i.e similar to what we’ve seen in the early 21st century)
A recent view of history tells us to not bet against the innovation curve
So if we assume that over the next decade, the $40bn invested into AI as a broad category results in measurable and material increases to broad based productivity, what happens?
My gut tells me that no different than historical pushes toward outsourcing post WWII, the private equity boom in the 80s, or e-commerce in the early 2000s; when the market identifies repeatable processes for generating returns; durable multi-decade trends of investment occur
First and foremost; we will see enterprises that have the capital base and engineering talent to experiment adopt these tools internally, or embed them into their current offerings
As these experiments at the top of the value chain occur and succeed, a few things will occur:
The companies that succeed early on with enterprise customers will experience compounding benefits to scale and ardently defend their position
My sense is that these companies will primarily be horizontal in nature and focus on job functions (i.e Customer Service, Sales & Marketing, FP&A, etc)
As a result, the next wave of entrepreneurs will look to replicate success in one of two ways:
1) Focusing down market on mid-market and SMB customers, but primarily providing the same product as the early winners
2) Focusing on verticalized opportunities and bringing in novel data sets that have been previously overlooked or recently emerged, and leveraging these new data sets as a means of differentiation
In many ways we can see the scatterings of the future in the present today
We have AI native, or AI enabled companies looking to make broad based job functions more productive
Lindy is looking to create a general purpose AI assistant
Julius is looking to create a co-pilot for data science
Lightpaper is looking to create a modern assembly line for knowledge work
Unify is looking to automate outbound tasks for GTM teams
ZipperTax is looking to create an AI copilot for tax professionals and taxpayers
Merlin is looking to automate user interviews for product feedback
TaxGPT is looking to automate general purpose tax returns
Blackore.AI is looking to automate general purpose tasks for the finance function
Betterleap is looking to create a co-pilot for recruiters
We have AI native, or AI enabled companies looking to serve SMBs
And we have AI native, or AI enabled companies looking to own a verticalized opportunity
Personally, if the promise of AI enabled software products works out - I expect a multiplier effect to occur where value is created both by the margin expansion for customers that adopt these tools and market share consolidating around winning tools in various verticals/functional areas
In either case, capital’s share of income will increase relative to labor
And we could very easily see the “outsourcing” of white-collar work in similar effect and scope of the post-WWII era of globalization
This could create all sorts of downstream effects such as long term headwinds for seat based software products, increased wealth inequality in areas with high levels of AI employment, and pockets of demand for specialized services such as industrial real estate development
I’ll leave you all with this. I think the true test of humility in venture capital today, especially if they are a new VC, is to ask them if they believe they could have predicted the success of the internet in the early 1990s.
By 1995, only 14% of the US had internet access and 42% of U.S. adults had never heard of the internet. Similarly, in March of this year 18% of US adults had heard “a lot” about ChatGPT and 42% had heard “nothing at all.”
As I said before, innovation is a longitudinal process built on the shoulders of giants. The success of the AI wave is not a foregone conclusion, the hype may recede and there could be billions of dollars washed away by the tides.
That being said, the history is clear and the opportunity is present.
You just have to ask yourself, am I correctly excited?
If you’re building an AI enabled vertical SaaS tool, are interested in working at an AI startup, or in general just want to debate on what’s next I would love to talk. Don’t hesitate to reach out to me at dez@firstmark.com