Why now? When innovations meet compelling economics
The term "artificial intelligence" was coined in the 1950s. In the (counts to self) 70-ish years since then, AI has been applied in various contexts, including expert systems, Lisp machines, and neural networks.
But given the long history, why haven't we witnessed a massive AI adoption until recently? And — zooming into the more recent years — why have only large companies like Google and Facebook primarily reaped the benefits of AI?
It’s a combination of the right innovations and economics that finally make sense.
Recent AI innovations are obviously enabled by advances in hardware. We now have enough compute to create compelling training models. We can also point to a specific technical innovation: transformers, a deep learning architecture that came out of Google. But another reason that AI has been rarely embraced outside of the realm of big enterprise players is that the economics of AI have historically been pretty bad.
The economics of AI finally make sense
When we look at the costs of AI in the past, there were some deterrents. First, you needed a lot of data. Second, you were competing with the human brain, which despite all the AI hoopla, let’s remember has proven itself to be quite adept at many tasks.
Add in the fact that it was expensive to assemble the teams required for AI projects, and you can see why AI was something normally only explored in big enterprises with big piles of cash on hand. But today, that's not the case.
“The current AI is tackling a set of problems with a new type of technology where the economics are incredibly compelling,” Casado said.
From generative art to self-driving cars
Consider an application like Midjourney, where you enter a prompt to generate an image of, say, a gorilla riding a Vespa. The inference cost to generate an image of our scootering silverback is something like 1/100th of a penny — and it only takes a moment. In contrast, hiring a designer to make the image might cost you $100 an hour, and it would probably take longer (and cost more). When you compare the two scenarios, you’re looking at a difference of about four orders of magnitude.
AI economics have also faced historical challenges because they tackled problems that humans excel at and where errors have significant consequences, demanding extensive investment. So while we’re “there” with things like rendering our moped monkey art or copy for a blog post, the economics still aren’t great yet for other AI areas.
Take self-driving cars: a technology with investments totaling around $75 billion, and still, the unit economics aren’t superior to those of humans. It’s just not yet competitive with an offering like Lyft or Uber. (Unless surge pricing starts really getting out of hand.)
“If you're looking at something like self-driving perception, humans are really good at that. We've got this part of our brain that's been around for 100 million years, navigating the real world, picking strawberries, and evading leopards, right? To compete with that is very tough,” Casado said.
But there are some other surprising areas where AI is already proving itself a powerful tool and ally.
"If you don’t have an API strategy, you need one. And if you think for your website or whatever your product is that the user in the future is going to be a human . . . it’s almost certainly going to be an AI."
General Partner at Andreessen Horowitz (a16z)
Business use cases for AI
“What's really interesting and exciting about the current wave of generative AI is the problems it tackles are things that we're not very good at. It's solving problems like creativity and companionship, and the part of our brain that does that is only about 50,000 years old, so it's not optimized at all.”
These applications of AI — like serving as a creative companion, a coding partner, or a brainstorming buddy — are carving out new use cases, significantly improving the economics of AI, and pushing computers into an entirely new area.
Is AI really that big of a deal?
In talking about the historical significance of this era of generative AI or foundation models, Casado compared AI to the microchip or the internet.
“This is a very much an epochal time . . . People sometimes compare it to mobile, but I think it’s bigger than that,” he said. “Anytime the marginal cost of something important goes to zero, you’ve got this explosion of value and productivity on the other side of it. And we’re already seeing that.”
Before microchips, we had people at desks and in rooms calculating logarithm tables by hand. After the computer — which is about 10,000 times faster (sorry, humans) — and the compute revolution, the marginal cost of compute went to zero.
Then the internet comes along and brings the marginal cost of distribution to zero. Selling software? You no longer need to box up a disk and ship it across the world. With the internet, the process of sharing data went from days or weeks to seconds. The price per bit went down asymptotically, and that created the internet revolution.
Casado said we can think of AI in this current wave as bringing the marginal cost of creation to zero. That is: we can create images, audio, and text for nearly nothing, which will lead to impossible to foresee breakthroughs and innovations yet to come.
“When you have new technologies and they’re super disruptive, it takes a long time for organizations to catch up... Don’t be like [companies in the ’90s that banned] the browser. You need to try to catch up and incorporate it.”
General Partner at Andreessen Horowitz (a16z)
How can businesses embrace AI?
When it comes to how businesses embrace AI, we’re likely to see companies make similar missteps as they have with other new technologies over the past decades.
“Remember mobile? ‘You can't take your mobile phone to work.’ Remember the cloud? ‘Oh, you can't use AWS,’” said Casado. “When you have new technologies and they’re super disruptive, it takes a long time for organizations to catch up. It’s almost like the individual gets it before the enterprise. That is absolutely the case with AI.”
Casado said the most important thing for enterprises is to understand the adoption cycle: users will use this tool, so rather than attempting (and failing) to ban it, enterprises must figure out how to address challenges — like figuring out what this means for data governance or regulations.
A potential mistake, however, would be to internalize this in central IT and have an AI strategy that is disconnected from the actual use cases because these are secular, organic movements.
“Don’t be like [companies in the ’90s that banned] the browser,” Casado said. “You need to try to catch up and incorporate it.”
The relationship between AI and APIs
Businesses need to embrace AI as a catalyst for growth. And the relationship between AI and APIs is central to transformation.
APIs are essential for AI. They act as hands, eyes, and ears for AI. And as the usage of AI increases, so will the number of APIs that enable them.
APIs will play a crucial role in facilitating communication between humans and AI systems, as well as AI systems and other connected tools. The future will see a surge in AI-powered agents and applications interacting primarily through APIs, to a degree where we can assume the users of the future will be AIs, and they’ll need APIs to accomplish their tasks.
Why businesses need an API strategy for our AI-centric future
For some back-of-the-envelope math, let’s just assume there are 4 billion people online today. With AI adoption growing as it is, there could realistically soon be 100 individual AIs per person. That’s 400 billion AIs, and every one of them will be online.
"Are you going to give those AIs a keyboard, monitor, and mouse? No. They're all going to interact with APIs," Casado said. "If you don’t have an API strategy, you need one. And if you think for your website or whatever your product is that the user in the future is going to be a human, it’s almost certainly not. It’s almost certainly going to be an AI."