It’s still early

Twice before, I felt I was late to a technology wave, when, in hindsight, I was early to the next, bigger wave.

As we try to figure out where we are on this exponential AI curve, I’ve been thinking about this a lot.

San Francisco, 2006

The first time I moved to Silicon Valley was the fall of 2006. I dropped out of college to pursue a language learning idea. I remember reading Steven Levitt’s book about Google and feeling, wistfully, that I had missed the golden age. I missed Google. I missed Facebook, which had fewer than 50 employees at the time—a few people we hired decided between offers from just us and them.

How wrong I was.

Especially looking backward, it’s striking how virtually none of the startup infrastructure existed that we now take for granted. One of our engineers came from Amazon and had private beta access to this new thing called AWS. I remember streaming Steve Jobs’ January 2007 keynote introducing the iPhone; the transformative App Store was still 18 months away. GitHub was just starting up down the street from us in SoMa. And our first two engineers left to join Y Combinator’s third batch.

On that last topic: I vividly remember Anson and Lux explaining YC to me (“there’s this famous Lisp programmer named Paul Graham…”). They felt bad about leaving, and they were so incredibly gracious that they recruited their former bosses from Microsoft to take their places. One of the occasional benefits of not having widely known best practices is that people could wildly exceed what they would’ve been.

I remember chatting with our lead engineer Nate about Facebook’s Connect platform. Both of our next ventures would be built off of it.

Mountain View, 2012

I ended up going back to college, and then in January 2012 moving back out for a second stint for the Winter 2012 batch of Y Combinator. Not having learned, again, I felt like I had missed all the good startups. By then, Nate’s Airbnb was already a runaway sensation. Instagram had fully taken off since its launch in 2010. Uber was becoming a household name. I even felt like I had missed the boat on YC itself. My batch was ~60 companies, the biggest ever.

If I could go back in time to give my past self advice, it would be simply: look around.

Our office hours were with Paul Graham. Jessica was there every day. So was Sam Altman. Garry Tan. Two Irish brothers personally onboarded us to their new payment platform, Stripe. Future YC CEO Michael Seibel was in our batch. So was his fellow Twitch co-founder Justin Kan, and Justin’s brother, Daniel, who would start Cruise. I remember Kyle pulling up to YC with his jury-rigged Audi. My co-founder, Tom, who Paul introduced me to in YC, would start Anthropic.

Again, I thought I was late to the party, when I was really just early to the next one.

2026

How early are we in AI?

If my own experience offers any hints: we’re still early.

The nuance is: yes, it’s probably too late to rebuild the things that have already been built, like frontier models, but my hunch is that 99.9% of the value created by AI has not been created yet.

Like in all tech waves, the breakout products of the future will not look like the ones of the past.

AI creates the possibility for totally new products, just as the Internet made possible Google, broadband enabled YouTube, the iPhone created the rails for Instagram, Uber and Airbnb, and so on.

I tend to agree with (Nobel laureate, Google DeepMind founder and general legend) Demis Hassabis, who recently described AI as much bigger than the Internet or even the Industrial Revolution, and more on par with the discovery of electricity or fire. As he put it:

“We’ve essentially found a way to make sand think.”

It’s still early.