AI Insights from Andrew Ng, Amazon Board Member & DeepLearning.AI Co-founder | AWS Executive Summit Recap

**SPEAKER_1** (0:03)
Welcome to the Executive Insights Podcast, brought to you by AWS, where we address vital questions and share unique perspectives from leaders at the intersection of business and technology.

**Ishit Vachhrajani** (0:20)
Welcome to Executive Summit at reInvent. I'm thrilled to be joined by someone whose work, and I don't think I'm exaggerating when I say that, has literally created the vocabulary and curriculum when it comes to AI, Dr. Andrew Ng. Andrew, all of us who have been to any tech conference in the last couple of years in this room, have seen a slide with your quote, that AI is the new electricity. But you said that, and you called that back in 2017, I believe. Can you tell us where we are in that journey? Are we still building the grid and the power plant? Or are we sort of ready to take the value out of it by plugging in our appliances in this world?

**Andrew Ng** (1:01)
Yeah. When I say AI is the new electricity, I meant that AI is a general purpose technology, meaning it's useful for a lot of applications, a lot of different applications, not just one or two color apps. And it's been interesting that right now, to those of us working on AI, it feels like the growth is very rapid. If I talk to my friends building agents and agentic workflows, we're all very busy. We see the stuff working and growing very rapidly. At the same time, we also see things like that, widely quoted MIT report, which I think has some methodological issues that I don't agree with, but that said that, claimed that 95% of AI isn't working yet for enterprise or something like that. And I think we're in an era where the practical adoption of AI is still at a relatively low base, but the growth is tremendous. And living in Silicon Valley, we've seen how this picture plays out. There are these things that everyone says was only, you know, 1% of workloads, and suddenly it's 2%, then 4%, then 8%, and we know how that goes. And that's why despite the caution and skepticism, I'm actually very bullish about the amount of tasks, about the amount of work that AI will be able to productively do over the next few years. But there will be years also. On the flip side, this hype that we have AGI in two years, or do everything, that's just hype. And I think 10 years from now, we'll still be working hard identifying workflows that we want to automate or assist using AI, using agentic workflows.

**Ishit Vachhrajani** (2:28)
One of the things, Andrew, speaking of agentic workflows, and you coined that term, that is a little counterintuitive, is that news cycles are typically driven by the bigger model. There's a lot of focus on the biggest model. For the leaders in this room, can you talk about when you say that agentic workflows are going to drive more progress than just the bigger model? How should we be thinking about that? What should leaders in this room should be thinking about in their org when it comes to enabling and thinking about those agentic workflows?

**Andrew Ng** (2:57)
It's been really interesting watching the news cycle of AI, because kind of two, three years ago when AI, modern genesis of AI was new, there was a handful of businesses that were at the forefront, or happened at the forefront, that almost got away with saying anything. And news media, traditional media, and social media were ineffective at fact checking them, because it was difficult to know what AI actually could and could not do. And so I think the information ecosystem for AI has become very polluted with a handful of businesses that were able to drive hype for PR purposes, or to try to lobby regulators, or whatever. This is what I see right now. If we look at the AI stack, so much of the attention has been on the technology layer, including the clouds and hyperscalers, and then also semiconductors, and then the AI foundation models. And despite all the capex going into these technology layers, almost by definition, there's a different layer of the stack that's got to be much more valuable, which is the application layer. Because frankly, for all the technology investments to even make sense, the application is built on top of them. The applications I think many of you will be working on building on, had better generate even far more revenue so that we can collectively afford to pay for all this capex. But for whatever reason, what happened in the last couple of years as well as where traditional media and social media tends to pay attention, the application layer just gets a lot less attention. And the other interesting dynamic I'm seeing is among investors, there is a recipe for spending a billion dollars or 10 billion dollars or whatever on capex to build data centers, to power model training or model inference. So a lot of media on that. The interesting thing I'm seeing is for building AI applications, which eventually has got to be even more valuable, the cost of placing a bet is very low. So there is no recipe for deploying a billion-dollar split applications. And the good news is, for those of you that want to place a bet, if you want to try out an application idea, you could do so very inexpensively. But because the bets are so inexpensive to place, it doesn't drive media the way that some supposed $1.4 trillion plan or whatever tends to drive a lot of media news cycle. So I find that if we ignore the hype and look at the business fundamentals, I'm very optimistic about what we can build at the application layer. And I'm seeing many, many green shoes. You just saw Matt's talk where he has talked about tons of applications that Amazon's building. And I'm seeing across many businesses tons of green shoes. But it's starting from a small base, which is why simultaneously true that, you know, 90-95% of whatever businesses have not yet seen a material change and grow because of AI. And at the same time, there's a small number that is rapidly growing.