AI 企业——Databricks 与 Glean｜BG2 嘉宾访谈

I think we have AGI. I think we have artificial general intelligence. We really have. You hear that 95% of projects fail, but that’s actually what you want.

I think the LLM is a commodity. People are not saying that, but it is a commodity. You can get gas from this gas station, you can get gas from that gas station—it doesn’t matter. Just compare the price.

Brad Gerstner

Is AI in a bubble?

There is an AI bubble.

Brad Gerstner

Okay, so then Glean is also in the bubble.

Everybody’s in the bubble.

No, I would say there is a bubble. I would say there are 3 camps. There is a superintelligence quest camp. I would be very worried there. There’s a second camp: the researchers doing the research. That’s definitely not in a bubble. They’re the sober ones.

Brad Gerstner

Yeah, they’re super sober, and nobody cares about them. [Laughter.] All right, and they’re probably the ones that are right, unfortunately.

And then there’s the third camp, which is us trying to make this valuable. We’re not in a bubble in the sense that we’re not spending huge amounts of capital on what we’re doing. We’re just trying to get actual economic value out of these organizations.

1. Consumer AI vs. Enterprise Reality

Brad Gerstner

Two legendary builders, Ali and Arvind. I’m so thrilled to get into this with you because both of you have seen every supercycle I’ve lived through: internet, mobile, cloud, data, and AI. Not just the supercycles, but also the hype, the trough of disillusionment, and “this time it’s different.”

Today, we’re going to chop it up on the state of AI. Let’s start with the 20,000-foot view. Take stock of where we are. We’ve seen consumer AI with billions of users. ChatGPT—the gun went off 3 years ago. Claude, Perplexity, ChatGPT: people use them in the room.

On the SMB and developer side, we’ve got hundreds of millions of users with Cursor, Codex, Claude Code, and so on. Enterprise, on the other hand, is a lot more divided. It’s hard to see; there’s a lot of fog of war. On one side, you’ve got models that are acing math benchmarks, science benchmarks, and engineering benchmarks. But on the other side, you’ve got the MIT report saying that 95% of AI deployments don’t work. What’s the reality? Bridge that gap for us. Lay it out as you see it—a view from the top.

2. Why 95% of AI Projects Fail

First of all, I think we should know that people use AI in both their personal and work lives. There’s not so much of a divide. Everybody in your company is probably using ChatGPT, Claude, and other tools on a daily basis.

The thing that I feel is happening in enterprises is that you hear these 95% of projects fail, but that’s actually what you want. When you’re actually experimenting with new technology, if all of your projects are failing, that means you’re trying enough at the moment.

So, I think when I read the study, it was not a surprise to me. We’re hopefully going to see similar stats next year, too, because you want everybody in the industry to be really eager to experiment and actually figure out how to mix this technology into their businesses and get benefits from it.

Brad Gerstner

This would make you guys, by default, the 5% of AI that’s working. [Laughter.] That’s 1 in 20. Maybe you go to the 5%. What is a use case that’s working—not just working, like saving me time, but working and transforming my company? Something you can take to the bank, to the CFO, while the CFO will not listen, but legal won’t shut it down.

Well, look, we’re seeing a lot of use cases that are working. You can’t just unleash the agent and expect it to work. It’s an engineering art. If you’re going to have a company that’s really differentiated, like my company or your company or anyone’s company, and you want to beat the competition, you can’t just quickly put something together and think that your competition isn’t going to do the same thing.

It needs evaluations, and you need to productionize it. It’s going to take effort, and you need a great team around it. But we’re seeing a lot of these use cases. I’ll give you some examples.

3. RBC, Merck, and 7-Eleven Use Cases

Royal Bank of Canada built agents with us that, as soon as an earnings report comes out, basically do the work of an equity research analyst. Their job is to put together reports that say, “This is a buy, this is a hold,” and so on. The agent gets the earnings report, all the previous earnings reports, all the competitors’ earnings reports, and everything that’s going on in the market. It does the full analysis—the news, everything—and puts it all together. It can get the equity report out in 15 minutes from the earnings call. The industry standard is 2 hours.

Of course, it’s going to get commoditized, and others are going to do that as well, but that’s actually a really important use case that we’re seeing in finance. There are lots of examples like this: sifting through hundreds of thousands of documents, SEC reports, and so on.

Let’s switch gears and go to health care. Health care is completely different. We have a customer, Merck, that in the life sciences space created a model called Teddy. Teddy stands for Transformer Enabled Drug Discovery.

This is a transformer model, kind of just like large language models that can predict the next word, but instead it can figure out which genome is missing if you remove a genome. So, it really understands the gene regulatory network and can start telling you what’s happening with gene expression and so on. This is really important for drug discovery. It’s the beginning, but it’s going to help us do things that we couldn’t do before.

Let’s pick retail, too. I’m picking different industries. Health care is one; I gave you finance with the RBC example. Let’s go to retail: 7-Eleven. They have agents that completely automate the marketing stack.

I actually think the marketing stack is going to get disrupted pretty heavily. These agents can prepare and segment the audience—this segment wants to hear this—and prepare all the marketing material that’s directly targeting you. They can put the campaigns together and do that.

7-Eleven was doing this before as well, but we’re seeing this at Databricks, too. More and more is being done by agents and automated, so you can do it faster and segment more finely. Before, you had to create the content for the groups. Content creation was heavy, human, manual labor. Now you can do that much more. You can have all your web materials completely customized for a target group.

These are examples where it is working. There are also lots of examples where it’s not working. Even at Databricks, we’re not just the 5%; we have some of that 95%, too. But those are some examples where we’re seeing success.

4. What Actually Makes AI Work

Brad Gerstner

Ali, a follow-up on that. These are great examples. If you were to take it a layer up, what is common across these use cases, these organizations, or these CIOs that’s making these use cases work? Is there something that we can pattern-match?

5. LLMs Are Commodities—Data Is the Moat

Look, I think the LLM is a commodity. People are not saying that, but it is a commodity. When I took economics classes, a commodity was something interchangeable. You can get gas from this gas station, you can get gas from that gas station—it doesn’t matter. Just compare the price.

LLMs have become that way. It doesn’t really matter. This one is better right now; next week, that one is better. You can’t even keep up anymore. They’re a commodity, so it’s not about that.

It really comes down to your company. What data does your company have that’s special, that your competitors don’t have? Can you leverage that, and can you build AI that really understands that data? Because that’s not a commodity. There’s not an AI out there that understands all your business processes, your secret sauce, and your data. That’s not a commodity. In fact, that’s closer to the 95%.

It really comes down to that, or to whether you have a complicated process that only your company has—this is how you deliver your products and services in your company—and that portion can somehow be disrupted with AI. If you can do that, now you can get ahead of your competition.

But it comes back to what makes your company special. Unfortunately, a lot of companies are just building commodity stuff. You should not be building that, because it’s something that every company can do. It’s not special to your company. I think that’s the problem in a lot of the industry.

Another problem in the industry is that there’s a lot of demo-ware. It’s really easy to make cool demos with AI, and therefore we’re seeing a lot of cool demos, but that’s all they are.

Brad Gerstner

Yeah. Well, something we say around Altimeter quite a bit is that your AI strategy starts with your data strategy. You’ve got to get the data house in order first.

There are a lot of reasons for use cases that we tried that were not working. Maybe give us an example of the 95% of an AI bet that either of you had at Databricks or Glean that did not work out, and why it didn’t work out.

6. Failed AI Bets at Databricks & Glean

It’s actually an interesting thing with engineering today: you build systems, and never before have you been in this mode where you start with a great idea and it doesn’t seem like a good idea anymore within 2 weeks, because we see a new development that happened.

We have numerous failures in engineering on that front. For example, some of our fine-tuning work—building models for a specific use case within our product—didn’t really pan out for us. Ultimately, the choice was that we could go with already-built models, whether they’re small open-source models hosted on Databricks or one of the large foundation models.

Internally, from a corporate use-case perspective, we’re also in this mode where a lot of our work actually—I would not say fails—but takes much longer to generate success. We’re trying to automate a lot of our business processes. For example, in our company, I want everybody to know exactly what their top priority for the week is and what they want to work on.

Maybe we want an AI agent to first tell them what their priority should be. We want it all to be documented, and we want a system that rolls it all up so I get a view every week where I can quickly see what all the different people in the company are working on and whether they’re aligned with what I want them to work on.

This is a simple thing. Companies have always tried to have this; as CEOs, you’ve always wanted it, and it’s always hard to make happen. We thought AI would simply, magically do all of this work because it has all the context inside the company to make it happen. But I still don’t have it.

Things do take time, to Ali’s point. AI is just one more tool that you have in the toolkit. It does not suddenly make building complex enterprise systems easy. It doesn’t mean that you can build them in 1 day, does it?

7. RPA vs. Generative AI

Brad Gerstner

Yeah. The last time enterprises got this excited about a tool was called RPA, and we know how that ended. It unfortunately fizzled out. Somebody in the audience yesterday asked, “How is this time different from RPA?” It seems like the same movie: bigger budgets, better actors. What’s different this time? How is the nature of the architecture of the technology different from the previous automation cycle? Either of you.

Yeah. Well, first of all, RPA didn’t capture my attention at all. I would not compare these 2 technologies at all. What we’re seeing now with AI is so fundamental. When we first saw it, it was basically magic, and we couldn’t believe that this was a machine doing this work.

Machines simply cannot do these kinds of things that we saw them do: writing on their own, having emotion, understanding emotion. It’s fundamental. It’s different, and that’s why I don’t think this technology is going to fizzle out.

You don’t have to be a financial expert or a deep thinker on business. This is obvious stuff. All of us know it, feel it, and can see the capability of this technology. We know it’s special, and it’s going to be around.

Bill Gurley

Yeah. You want to hear my RPA?

Brad Gerstner

Please.

Bill Gurley

It was rule-based, and the problem with it—especially if you want something that automates what’s going on on your desktop and automates the work that’s happening—is that too many unexpected things happen, and it’s hard and brittle to set up. It was never learning. There was zero learning. You told it exactly where the rules were, and if you got something wrong, you needed to go back and expand the rules.

100%. There have been many startups that failed in generative AI trying to replace RPA with generative AI models. There are many startups that failed, actually—some pretty high-profile ones that I know of. It’s because the paradigm we live in today with AI still has problems.

The biggest problem is that you bake a model, and that’s where it’s learned everything it needs to learn, and then you freeze it. Then you launch it, and maybe you give it some context, but that’s it: it’s frozen. So therein lies the problem: We need an AI that really can continue learning while it’s using the desktop and clicking around.

I do think this problem is hard to nail, but I think Arvind is right: There’s no comparison at all. It’s brittle, rule-based stuff versus a learning, agentic system. I think it’s going to nail it perfectly, but we haven’t really nailed computer use yet.

Yeah, we’re working on it.

Brad Gerstner

The number-one shift is this move from if-then-else statements to a more generative solution that figures out the solution. You’re trading breadth for maybe determinism. That seems to be the difference.

8. Advice for CIOs Planning AI Budgets

There are a lot of CIOs in the room here, and they’ve got budgets coming up to plan. If you were giving advice to them, something like, “Hey, based on everything I know from my customer base, here’s 1 thing or 2 things that you’ve got to figure out and align incentives on”—it could be a reliability problem or org design—what advice would you have for CIOs who are thinking about their AI budgets right now?

Bill Gurley

Well, spend more.

Brad Gerstner

On Glean.

Spend more, yeah, put it on Glean. But I think one thing that’s important in the AI market today is that it’s very new and there are many players. In fact, every software company is also an AI company now. You can go and check their websites.

It’s hard to figure out where to allocate those budgets, and what we tell people is that I think the winners are yet to be identified. Experiment with more vendors and do shorter-term contracts. While that’s easy to say, it’s hard to implement because every product that you try has a cost that you have to pay to even test it.

You also have to pick products that are easy to test—the ones that don’t require you to spend the next 6 months trying to implement something when you have no idea what’s going to come out after that. The products of today, the products that are built with the right AI, should work very quickly for you.

9. AI CapEx and the Revenue Math

Brad Gerstner

Crawl, walk, run. We’re going to take a peek into the future. Shifting gears, one of the things that keeps investors like me up at night is $250 billion being spent on NVIDIA on the semiconductor side of things. Assuming that is just 50% of the capex, you’re spending about $500 billion on capex, and then you’ve got to earn about $1 trillion of AI revenue for all of this capex to be worth it.

The entirety of the software industry earns about $400 billion of revenue. This seems like a physics problem at this point. How do you think this plays out? You’ve got to make about $1 trillion of revenue to justify this present spend that’s already happening. How do you think this shakes out? Maybe we start with you, Arvind.

Wrong person to start with, but I’m an engineer, and I shouldn’t really think too much about who’s spending what money. We’re here to build our product and add value, so in some sense, I’ve not really thought too much about this problem.

But if you think about AI, AI is not actually extending software in a marginal way. It’s a different product, and in fact, it’s going to grab a lot of revenue that’s actually in the services industry today, which is 25 times larger than the software industry. There’s a lot of spend that’s going to move.

The spend that you see happening on AI is, in some sense, those service dollars converting into AI or software dollars. With that said, maybe you have a more informed view on this.

Bill Gurley

Arvind, do you think that’s—just to build on what you said, “I’m an engineer and I want to just build something that’s cool”—I do think it’s not binary, right? It’s not like, “Okay, the physics doesn’t work out, so the whole thing will collapse.” No, there are going to be things that work, and so it is a good idea to continue focusing on the stuff that’s obviously already working. Continue expanding on that.

10. The Three Camps of AI

But if you zoom out, I think there are 3 paradigms, or 3 camps, and I put Arvind in the 3rd camp. I actually put myself also in the 3rd camp. But let’s start with the 1st camp. I think the 1st camp is this quest for superintelligence camp, and I think all the frontier labs are doing this.

All 3, 4, or 5 of them, however you want to count them. A lot of it still comes from the scaling-laws mentality, which is that whoever has the most GPUs and the most data is going to win the quest for superintelligence—intelligence that is almost godlike. It leads to recursive self-improvement of the AI, which, once you have it, can cure cancer and solve all economic problems. We can probably 10x GDP over a period of a few years.

So what the hell are you talking about when you say there’s a physics problem? All of your cost equations are going to pale in comparison to the economic value that this thing is going to provide. That’s 1 camp, and the way they’re developing it is through bigger and bigger clusters and more and more energy. That’s how they’re going about it.

That’s where most of the capital is going, right? That’s not the kind of capital you’re spending or I’m spending, but that’s that camp. How do they know that they’re succeeding? They’re not just saying, “Oh, just trust us.” They’re very smart people working on this.

The way they’re approaching it is, “We’ll throw the hardest questions we have at whatever AI we have now, and if it nails them, we’re making really rapid progress. So what’s your problem? Look at the Math Olympiad. We’re nailing these Math Olympiad problems, Physics Olympiad problems, and programming contests. It’s better than any human being.” That’s what they’re throwing at it—all the most intellectually challenging questions.

There’s a 2nd camp, which is the people who created the original technology—the scientists who created the technology and got the computer science Nobel Prize for it, called the Turing Award. That’s Rich Sutton, who created reinforcement learning, which a lot of this stuff is built on. You have Yann LeCun, who’s 1 of the 3 founding fathers, and many others.

These people have been saying for many years—in fact, I’ve asked them for years—that the 1st camp is not going to work. That’s not even the right approach, in their view. They’re like, “No, that’s just autoregressive next-token prediction. It’s just probabilistically predicting the next token. Usually, they will say that’s not how humans learn. That’s not how animals learn. We operate in a different way. Your brain is not that way.”

1 example is that even a child learns very quickly to walk, talk, and do things with very little data. Certainly, no child is reading all of the internet’s data 4 times over before they learn to speak. That’s camp number 2. Those guys, by the way, say it’s 20 years out. They’re saying, “Hey, it’s a physics problem, and it’s going to take 20 years to get there,” which, to me, is like, “I don’t know. Leave me alone. Let me do research.”

The 3rd camp, which I think is what we’re in, is that I don’t think we need superintelligence. I don’t think we need that superintelligence right now. Maybe they’ll get there. That’s awesome if they do. But I think we have AGI. I think we have artificial general intelligence. We really have it. We absolutely have it.

11. Making AI Useful Inside Enterprises

Anyone who says we need to get to AGI—that’s a false premise to start with. We already have AGI. I came to the United States in 2009, to UC Berkeley, not far away from here, and I was in an AI lab. It was called the AMPLab—Algorithms, Machines, and People. These were all AI people.

Back then, the definition of AGI we had was already satisfied. I know the discussions we had. I actually went back to some of those folks to see whether it was just me or what the sentiment was back in 2009. Everybody I talked to said, “Yeah, by those standards, we had AGI, but we’ve changed the definition now.”

For 30 or 40 years, we had a definition of AGI. We’ve already hit that. Now we’re changing it and moving the goalposts. But, very obviously, we already have AGI. Just use any of these LLMs and have it do some reasoning. Certainly, it’s smarter than a lot of friends that you have, right? Let’s not name our coworkers or whatever.

You already have AGI. Now we’re haggling over exactly how smart it is. Do you have a friend that’s smarter or not? If we already have AGI, we just need to make it useful inside the enterprise. We need to expand that 5% to 10%, 20%, or 30%.

That’s why I think Arvind’s answer is actually a good answer. We have the AGI we need. Let us just focus on solving the actual problems inside organizations. I think that’s enough to automate a lot of the tasks and get huge economic value out of it. We don’t actually need superintelligence for that.

That’s a good idea. If the superintelligence guys nail it, amazing. Then we’ve cured cancer. If they don’t, hopefully the 2nd camp comes up with a new thing in the next 20 years. That’s also awesome. We already have whatever we need. So, yeah, let us just do our engineering.

Bill Gurley

Right. Yeah. Yeah. That’s really good framing. The way this manifests in the world is that there’s a data layer, there’s the intelligence layer—which is where camp 1 is presumably producing a lot of great models—and then there’s the software layer, where users engage.

Where do you think value accrues if you were to design 100 units of value across these 3 layers: the data layer, the intelligence layer, and the software or application layer? Where do you think value accrues in the next 5 years?

All right. This is a tough question. I think all 3 layers are actually very fundamental. I thought you were going to add a few more, which you didn’t. As Ali was saying, the models are going to be available to all of us. They’re going to be commodities. It’s going to be hard to see more spend going to them versus these layers on top.

It’s hard to come up with where the most value will be. I also don’t know if it actually changes from today’s technology architecture. Again, think about a pre-AI world: any sort of enterprise application and data systems—you have data systems, you have the application layer, and I guess you don’t have enough of that intelligent layer today.

I guess some dollars will shift into it. We do think that the intelligence layer is actually going to be a pretty thick one. Maybe it’ll capture half of the enterprise value. Anything to add, Ali?

12. Why Apps Capture the Value

Yeah, no. I think there are more layers in the stack, depending on how you want to do it. But, as Arvind said, the LLMs are a commodity. You can get them. That doesn’t mean those companies aren’t going to be valuable. They can be very valuable—I mean, TSMC is very valuable.

I’m saying they’re going to be kind of like these fab-like companies. They’re interchangeable, and we’ve never seen something like that before. I have not seen that during all these years. People just switch LLMs in 1 day.

That’s not the case with your iPhone versus Android, your Windows versus your Mac, or anything versus anything. Google Sheets versus Excel is a huge religious battle inside our company. But LLMs are different because, as I said, they’re a commodity. They just speak English, or any language you like, and they give you different answers every time. You might as well just try the cheaper one, the cheaper commodity, or the slightly smarter commodity. You can’t even really tell the difference, can you?

What is special, then, is the data that you have. If your company has data that it has actually collected and your competitors do not have, that is valuable. Glean is amazing, but if you remove all the data from Glean, there’s no use to it, right? It’s all about the data that you have.

Can you secure the data as well? If we’re going to have agents running around accessing this data, it could be, “Oh, that’s his HR data. Oh, here’s the provost’s salary information. Oops, I blurted it out to all of you.” How do you lock it down? How do you make sure that there’s governance?

There’s also a lot of worry around this. What if it’s using a Chinese model? What if it’s accessing this information? What if it’s sharing this information with a competitor? What if it’s interacting with something it shouldn’t? The governance and security layer is going to be super, super important.

But I do think most of the value will accrue to the apps. I think that’s common sense. I just don’t know which apps. I do think Glean is amazing. Do you think of it as an app? I don’t know.

Now, we see ourselves as both an app and a platform.

Yeah. So I think it’s—let’s call it an app platform. I do think it’s amazing because it has the potential to automate so much of the overhead inside an organization. If you think about why organizations have hundreds of thousands of employees—some organizations have 50,000 or 20,000—a lot of it is the coordination overhead. So many people have to communicate with each other: “Hey, what happened? What did you exactly mean by this? Let’s do a meeting where you explain it to me.”

I ask some questions. Let’s—oh, let’s invite these other guys also, and then write it down. The coordination overhead of organizations is massive, right? It’s like this N-squared problem where everybody needs to communicate with everybody, and they’re communicating inside their siloed org chart. But how do we get it across? Through docs, Excel sheets, PowerPoints, and meetings is how we move companies and organizations forward. So much of that can be augmented and made more efficient with Glean. So that’s why I think Glean is amazing.

But this is kind of like 2000. You ask, what are the killer apps on the internet? By the way, back then, we thought it was Cisco routers, portals maybe with thousands of links on them. Actually, I had just started college, and we knew that the future of the internet would be portals, which were these web pages with 100 links on them, and you just click, click on the right link. This was before Google Search.

But the future of the internet actually didn’t look that way. It ended up being things like Facebook for friends, Airbnb for rentals, Uber for the cab industry, Twitter, and so on. Those became great companies. I don’t know what those are for the future. They will pop up, and they will be extremely valuable.

Brad Gerstner

Okay, so does that mean that Databricks and Glean then basically will die, and there’ll be a new set of companies?

No. Back then, there was actually an Amazon.com already in 1998. There was already a Google—it actually existed already in 1998, and it was only a $300 company or something like that, right? So it’s not binary. We’ll see what happens, but I do think a lot of value is going to go to the future apps that will emerge.

Brad Gerstner

Speaking of that, let’s double-click into it. The $300 billion companies of today at that layer—software apps like Salesforce and ServiceNow. There’s a lot of talk about software being dead. Satya calls them the CRUD apps. What is the future of this layer that today is called software, which seems to be heading toward becoming a database? Where do you see the value accruing to this part of the layer? Maybe start with you, Arvind.

Yeah, I think that’s an oversimplification. For example, even to say that Salesforce is just a database—it’s a full ecosystem of workflows and other applications built on top of that infrastructure. I haven’t really understood this concept that you have a database where all your enterprise data is, and then people can just go and create dynamic UI experiences on their own on top of that data. Every business can, for example, create all the UI by themselves on this. I don’t think it’s going to happen like that.

13. The Future of UI, Voice, and Data Entry

AI makes it easy for you to build. You can have a database and talk to AI and create a UI and experience that is exactly what you want it to be. But most times, you actually won’t know what you want. A lot of the good thing about software companies is that they think about how to take that data and present it in a way that lets people interact with it or modify it in a way that is natural and drives more productivity from a human. So I think, ultimately, software is an end-to-end stack, in my opinion, and all of these companies—I don’t think they’re going away. I don’t think they’re going to be relegated to becoming a database.

Brad Gerstner

Humans, over the last 20 years, got addicted to these screens. We scrunched over the screens, and we would input this information with our keys and the dropdown: “Hey, I met Arvind today, and this is what I learned.” It should really be, “Hey, Chat, I met with Arvind. This is what I learned. Remind me in 2 days to catch up with him.” That will happen. I think it’s going to happen in the next couple of years, and even with Glean, you won’t want to type; you’ll want to talk to it. But I think the big thing is data entry. How does the data appear in that database?

That’s today not completely automated.

Brad Gerstner

I think a company that would be well positioned to do that would actually be Zoom. A lot of people don’t think about it that way, but Zoom really should be the perfect data-entry application, because that’s where you’re having all the conversations and where all the information is coming out. If it could work with Glean and extract the most important information, store it all—not in a structured data table, but in the system of record—if you had that, that would be the full disruption of SaaS.

That’s actually one of the most common agents these days with Glean. You take these meeting recordings, figure out what you talked to the customer about, what the action items were, and then the agent updates the notes in Salesforce with that.

Meetings are—at Glean, we have this policy where we record every single meeting, internal and external, if our customers allow it, because there’s so much information in there.

Brad Gerstner

I joined a meeting last week; it was 4 humans and 6 AI note-takers. I heard about 17 note-takers in one of the discussions yesterday. This felt like the first scene of a movie where the AI takes over. Clearly, there’s a lot of sprawl—almost too many tools—and consolidation coming at some point.

But maybe your personal workflow: You guys are CEOs in the age of AI. A lot of CIOs are in the room; they’ve got more jobs than time on their hands. How are you using AI for both your personal selves, and how are you driving your organizations—both large organizations—to adopt AI and benefit from it? Maybe give us a glimpse of your leadership in the age of AI. Maybe, Ali, we start with you this time.

Yeah, we have agents for all kinds of stuff that we use. Everything from—we have agents that are really good at understanding our customers. We have an agent, Raffi. Raffi is the name. If I want to understand anything about any customer, I can ask it, “Tell me the best customer story on this.” I told you about RBC, Royal Bank of Canada, but I can just ask it, “I need a use case. I’m going to get on stage. I’m going to talk about the finance sector. Give me a use case that has these…” It’ll just find all the information and collect it. So it’s really, really helpful for me for these kinds of things, when I get on stage like this.

Also, if you go into a customer meeting, I want to tell Customer X about their biggest competitor, Y, and how they’re using Databricks. Now, maybe Y isn’t a competitor or isn’t using Databricks, so then I shouldn’t use Y; I should use Z, which actually is using Databricks. Maybe that’s the number 2 competitor. How do I get this information super quickly? All of those are prepared at Databricks.

On the go-to-market side, a lot of this is being completely automated, and we’re using it. The marketing stack I already mentioned is heavily automated already. A lot of the tasks that are happening in marketing are being automated, so we’re seeing that stack happening.

Then there’s engineering. That’s a whole big thing. I think there’s a whole change-management issue in how to do it right. Initial attempts to automate a lot of the software engineering at Databricks kind of failed. There’s nothing wrong with the AI; the problem is the humans and how we were organized.

Databricks is a big 6,000-person go-to-market organization and a 3,000- to 4,000-person R&D organization, and then there’s some back-office stuff. Those 2 big organizations are already seeing heavy automation using agents for all kinds of tasks.

Then there’s back office, so that’s finance and these functions. Finance is all on Databricks, and all the forecasting has moved to machine-learning-based systems. But it took them a long time because they had their Excel models, and they were very proud of them and didn’t want to change. Again, there’s a change-management issue there.

We actually had an external data-science team build the AI models, and then eventually they became good enough, and now finance has taken those over. Finance has largely moved from Excel to Python at Databricks. But it was a journey, because most of us speak Excel. A similar thing is now happening in HR and other departments as well.

In general, HR departments are not the closest to doing this kind of analytical work with Excel and so on, so maybe they’re not quite as far along. But yes, we’re seeing it everywhere.

Brad Gerstner

Anything to add, Arvind?

Same for us. I can share some of my own personal use with it. One of our agents is the daily-prep agent, which I really love, because every morning it tells me what my day is going to be, what I need to read, and what I need to prepare. For most of the meetings, I won’t have context; it actually brings the plan for those meetings for me.

So that's one of my favorite agents. It helps me feel more confident about how I'm going to do my meetings during the day. The other one, which I also shared yesterday, is that I've changed my instinct. Changing instincts takes a long time.

When you're the CEO, you're the boss, and everybody listens to you. You can just say, whenever you have a small question, “Curiously, just go and ask somebody,” and they're going to put 30 people on the task to get that answer for me. You're going to have a prep meeting before the prep meeting before the meeting.

All of that was sort of easy for me. I just got to ask somebody, and I changed that because I knew I was actually causing a lot of that, which was very expensive. Today, my instinct is that whenever I have curiosity, whenever I have questions, when I need to do data analysis, when I need to write something—my letter to the company every month—all of those things, fundamentally, I use AI, of course Glean in this case, to actually help me do my tasks.

More, more, I think you have to sort of have that belief. A lot of people won't do it. You have to have that belief that AI is a good collaborator. It's not going to do the work for you, but if you use it, you're going to produce better output eventually. Even if you don't save time for the first few months, you're actually going to improve the quality of your output.

14. Rapid Fire: Winners, Bubbles, Long/Short

Brad Gerstner

Fascinating. Well, this brings me to my favorite part of this conversation, which is rapid fire. Short answers are fine; long answers are welcome. Start with 12 months from now. Are the big AI companies that we know of today up or down? We'll start with OpenAI. Twelve months from now, are the stocks up or down? Ali and then Arvind.

Up. And I'll say revenue will be up. I don't really understand how stocks work.

Brad Gerstner

Anthropic. Ali or Arvind?

Up. Same.

Brad Gerstner

Okay, Arvind.

Because ChatGPT is going to continue growing, and it’s on fire, and it’s what everybody uses. So is Gemini, by the way. And then Anthropic, because more and more coding—we've only eaten into a small portion of that market. It's just started.

Brad Gerstner

Is AI in a bubble, yes or no?

There is an AI bubble.

Brad Gerstner

Like saying, okay, so then Glean is also in the bubble—everybody's in the bubble.

No, I would say there is a bubble. I would say those three camps.

Brad Gerstner

Yeah. There is a superintelligence quest camp.

I would be very worried there. There's a second: the researchers doing the work. That's definitely not in a bubble. They're sober. They're super sober; nobody cares about them. They're probably the ones that are right, unfortunately.

And then there's the third camp, which is us trying to make this valuable. We're not in a bubble in the sense that we're not spending huge amounts of capital on what we're doing. We're just trying to get actual economic value inside of this organization. So I don't think it's binary, but there is a bubble. I mean, there are startups with 0 revenue worth 10, 20, 30 billion. That's a bubble.

Same. I mean, I think there are quite a few companies where there's a lot of optimism and valuations that are well ahead of the business those companies have. I guess you can say, compared to non-AI companies, AI companies do have higher multiples. But I think that sort of comes from the fact that there's a good reason for it, because these AI companies are going to grow more than non-AI companies, for sure.

Brad Gerstner

My favorite game that we ask our CEOs is a long-short game. If you were to pick a company, a product, or an idea that you're long on—that you think is going to be a bigger deal than it is today—what is that? And then short, which is where there's more sizzle than steak, more hype than reality. Pick a long, something that you're really optimistic about. Same order: Ali and then Arvind.

I am very long on agents. I think I'm very long on speech—speech as an interaction. I think keyboards are basically going to disappear completely. We haven't actually nailed speech. I know it feels like we have, but we haven't, because you're still using your keyboard. As long as you're using a keyboard, we haven't nailed speech. But I think we're this close to completely eliminating keyboards. So I think that's a big one.

What would I say? I do think coding is a little bit overhyped. I don't know if I would short it. I mean, I think it's still the future, so I think that's one of them. I think automating customer service and support is a little bit overhyped.

I basically think the things that the industry thinks are amazing and where we've made great progress, we probably haven't made as much progress on them. A lot of the other things that are being ignored, we're going to have breakthroughs in those.

Brad Gerstner

Fascinating. Yeah. And for me, I think the products that are going to change the paradigm are the ones where, instead of you building a product and expecting people to come to you, you understand your user and your customer very deeply and actually bring the AI to them. That's the category that I'm excited about. I want to see more proactive AI products coming to the market next year.

That's what is going to take it from 5% of the users being power users to 100%.

Your favorite AI tool that you use in your lives?

I think Glean is awesome. I mean, if that was not clear.

Brad Gerstner

Let's go. So he uses it all the time.

I actually ask Glean a lot of the questions I would ask from the team. The thing you said you changed, I first ask Glean and then see if it nails it or not. Then if it doesn't, I'll spin up a 30-person team to go spend a week and have 3 meetings and all that to get the explanation of some simple concept for me. But usually Glean nails it.

Well, for me, I'm excited about note-takers. I've used Grain and Otter.ai myself and Fathom and a few others. But note-taking is actually fascinating. I feel like if you take those notes and then utilize them the right way—for example, what Ali was saying—that becomes the source of what then actually creates knowledge and saves data in your systems. That's going to change how companies work.

Brad Gerstner

In closing, I'd love to get your vision for your companies. We'll start with Ali's favorite tool, Glean. Congrats—you just announced crossing a big milestone: $200 million in revenue run rate. You're signing big deals, $10 million deals. You've got super users, and I'm seeing you're seeing casual users. Paint us the vision for Glean from here to $1 billion in revenue.

I think we're still doing annual planning, which some AI companies are telling me is old school. But we're doing it regardless. We're doing it. That's just because they're early startups.

Brad Gerstner

Did you do annual planning when you started Glean?

No.

But I think for us, the thing that I'm most excited about, again, is that we think a lot about AI literacy and how you get everybody along on this journey. We're not seeing it right now. Glean is a heavily used product, but there's still a big variance between the top users and the ones at the bottom. That's what we want to change.

For the future, we want Glean to be this very personal companion for every person in every company in the world. This companion with which you have a very confidential relationship, in the sense that whatever you ask this companion, whatever communication you have with them, it's fully privileged. Nobody else gets to see it.

But this companion knows everything about you and your work life. It knows your day, it knows your week, it knows who you're going to meet in the day-to-day, it knows your weekly goals, it knows what things you're not good at, and it knows what your career ambitions are. With all of that, this personal companion is sort of helping you now with your work. It hopefully takes the majority of your tasks automatically and works on them before you ask it to work on them. That's the vision that we're taking our product to.

We have most of the foundation for this in place already. Today, you have to come to Glean to get most of that work done. In the future, we want Glean to actually come to you and do that work.

Brad Gerstner

Fascinating. We can keep going for a bit, but I'm being called on time. Thank you so much for chopping it up with us. You got a lot of alpha and a lot of insights here. Really appreciate it.

Thank you.

Thank you.

Brad Gerstner

All right, gentlemen. Thank you so much.