Tavakoli: I think it’s twofold. One, you should focus on the outcomes, not the tools. The number of people that say, “We're behind. I've got to get a bunch of agents up and running. I’ve got to show that I can do AI.” You get no points for saying I stood up AI, right?

Instead, what you really want to figure out is, What's the business outcome I want to drive? And usually that is “I have an existing process that I want to automate and make much more efficient,” or “There's a new set of capabilities that I want to put out,” and AI is what unlocks and makes it possible to do that.

The second thing is, everybody got so excited about AI, and they associated it with LLMs and which model are you going to use? And honestly, the biggest thing in the enterprise world is AI that is high-quality, accurate, trusted. And that is very much dependent on, “Do you have your data estate in order and do you have a governance strategy?”

It's not about the model; it's all of the other pieces. How do you get accuracy? How do you govern it? How do you figure out how you put it in production and measure it? And also, how do you do that in a space that's rapidly evolving? The majority of people you talk to who  launched an AI application even six months ago tell you that if they were to rebuild it today, they would build it completely differently, because there are new products out.

Tavakoli: The current revenue is a pyramid. At the bottom layer, you need a bunch of infrastructure, and those are chips. That’s an area where there won’t be a ton of companies, because the barrier to entry is very high.

On top of that, you have the foundation model providers. We started out with a lot and it’s winnowed down, mainly because of the capital you need to train some of those models.

The last layer is the applications on top. And today, because it's early days, that is not massive — even though Databricks just recently crossed over $1 billion in run rate of AI revenue, so it’s not chump change.

If you fast-forward five years from now, the pyramid is going to be much more massive, and it's going to invert. Much more of the revenue is going to be on applications that are leveraging AI to transform what folks do. And in that space, I don't think that there is a winner who takes all.

Tavakoli: Nobody likes that answer. But many processes you spend a lot of money on are not sexy. I'll give you an example. You're an insurance company. You get tons and tons of claim forms coming in, and the amount of horsepower and frustration that is spent on — “How do I take all those claim forms and extract the information that I need? How do I put it on an analytics form so I can run insights on it and then, based on it, take action?” Nobody gets excited about that — except the person who's sitting there and pissed off that their claim takes three months to reimburse. But if I can now go from doing something that took months and do it at a fraction of the cost — automated — that is a really exciting use case.

Or you're a manufacturer of semiconductors, and if you have something that could intelligently detect anomalies and improve your yield by 0.1% — once again, when was the last time someone got excited about fab yields? But it means a lot of money.

Huge in productivity, huge in cost, not ones that people associate with as being earth-shattering. I think that those are boring AI use cases. You can deliver meaningful improvements with AI, and that's what we've seen with our customers.

Tavakoli: The answer I always give is, by this logic, if we had said, “Hey, when ATMs came out, or when computers came out, they were massive transformations — will a lot of people lose their jobs?”

There are a certain set of things that people do today that will be automated by AI. However, on many of these things, just for quality, you still want a human in the loop. And the whole premise is that when you automate these tasks, they also open up an entirely new set of demands to do things that you could not do before. So, for example, now that you have ATMs and online banking, there are new sets of e-commerce roles that opened up that we wouldn't have ever thought would exist in the past world, and those generate a ton of jobs and productivity with it.

With upskilling and training, while specific job responsibilities will shift, there are a whole new class of new job responsibilities where companies are going to need people to drive. So I actually think that you'll see demand for labor increase. So it's more about, “How do you upskill?”

Tavakoli: Yes and no. Yes, we are in an AI bubble. No, it doesn't change the plans.

I oftentimes get asked the flipside of that question, which says, “Is AI transformational or is AI overhyped?” And my answer to that is yes. I think people still don't fully understand AI, and so the answer to any question is, AI is going to solve it. I walked around and there was a sign that said “AI-powered car wash.” And I'm like, I don't know what the heck that means. Everything is AI-powered now. There's always that peak of excitement that is bound to subside as we settle on what the real use cases are that people need. I think you won't see all of the companies right now in the AI space continue to survive.

Why it doesn’t change Databricks’ plans is that AI is great and we think it’s important in the future, and obviously we’ve leaned heavily into it over the past 12 years. But also a core part of our business is the data side of the house, like data transformation and operational workflows, which are proven, which are definitely not in a bubble and are growing. From a Databricks perspective, you adapt to what customers need. And we've already seen that movement from overhype to what are the key important use cases and outcomes, and we've supported them there.