I'm John Furrier here at the NYC Cube studio here in New York City. This is our East Coast hub. Of course, we got our Palo Alto studio connecting tech and business money here, Wall Street and Silicon Valley. Arvind Jain is the co-founder and CEO of Glean, a rocket ship startup. Um, we've covered many times. If you've watched the Cube, you know Arvind's been on multiple times. They're continuing to innovate, bringing hardcore deep tech into the enterprise. They already cracked the code on search in the enterprise, doing a lot more. Arvind, great to see you. Thanks for being part of the NYC wired program in the Cube and our AI factory focus and our mixture of expert series. Thanks for coming on.
Yeah, excited to be here.
You know, we have been hyperfocused on the enterprise AI for a long time as you know, but obviously the people looking at the market, the reports go, oh, 10 out of nine out of 10 projects aren't failing. Okay, first of all, not sure the stats are right, but certainly not the explosive growth on the production holy grail workloads.
That being said, you know, search is working what you've been doing because they have data. So, this is a progression in my mind. So, you know, we're predicting 2026 we'll see a surge of growth. Enterprises aren't hyperscalers. They got high bars for resilience. Um, they don't have the capex budget to throw at GPUs, everything. So take us through the enterprise AI market because certainly there's a robust data business, but as enterprises start to see value, what's your take on where we're at because you're doing very well. What's the state of the enterprise AI market?
Yeah, well, I mean like so everybody's seen these studies, you know, which are showing that AI projects are sort of failing, stalling. But I think we actually see really good progress not just with our product but there are a few specific use cases that enterprises are seeing good success with today. So first, the biggest application of AI of course has been knowledge access so far. So enterprises, they're deploying good search solutions like Glean and this becomes a really important tool for everybody in the business. You build a product like Glean, give it to all of your employees and now you're bringing AI into the day-to-day work of every individual. So that's actually the first thing that enterprise is doing because what they're thinking about is that look, AI is happening and I need to actually bring AI tools to all of my employees and get that education going. So good progress on that part.
And we covered that before, that the RAG, the retrieve augmentation generation market was booming. Why? Because they have data.
Now, when you start to scale it across the enterprise and this is why I like your news that you guys have announced, making it personal. Well, I think you hit that last time we were on. Personalization is a killer app, but search too. The user experience is changing and agents, the promise of agents are certainly hyped up, but I think it's legitimately hyped up, but the infrastructure's got to get taken care of first. You can't just wish yourself or manifest agents. Talk about that importance because you guys have specific new things you're talking about here.
Well, I mean the last year has been all about AI agents and expectations have been sky-high and that's what you know actually when you see these reports, that's what's coming out. People thought that I could actually bring an agent, I can take any business process that's in my company and just automate that in one day, and that's actually hard. The agent technology has actually been developing over the last one year, there's a lot of infrastructure that you need to build to make sure that there are the right guardrails, the right security, monitoring, checking if agents are actually performing with high quality. So I think what we've seen in the last one year is enterprises deploying agents while the technology is getting built. So for that reason, there's been some disappointments but I think we're in a really good place now. And at Glean, what we have done is we worked hard on making sure that piece by piece we build that technology stack, the vertical set of tools that you need to build agents that can actually be production grade. So when you build an agent with Glean, first making it really easy, we're democratizing access to agent building. You could be a business owner, not a technologist, not an AI scientist, but you can still build something really cool with AI through our agent builders. It's becoming very natural. But then we're actually also bringing the rest of the infrastructure to help you evaluate, to test, to monitor, to put the right guardrails, make sure agents don't go wild in the enterprise. So that's actually the progress. So the industry is moving very fast. The use case you see, the preferred future that's happening very fast is the vibe coding of apps inside an enterprise or workflows, which I say vibe coding because everyone knows how easy it is, everyone sees the Replets and the Lovables of the world, but the problem is once you build your app, 'Hey look at my app', and then you hit a button and the API doesn't work. It's hard. Even vibe coding you get excited by the user experience but to wire it together is hard.
It's very difficult, it's even harder on the enterprise. Forget the web. So take me through how that works because you know if Glean on our last interview you said that Glean's the enterprise brain. Okay. Now you got the enterprise graph. So you have almost like neural connections to data. Take me through this process of how you guys see the Glean enterprise graph which connects to the enterprise brain and how that makes agents work.
Yeah. So firstly, if you think about any agent, an agent is doing some work, is trying to perhaps take a business process and automate it. So as part of that, you have to do a few things. One, you have to actually work with some information that's inside your enterprise. You're going to apply AI and its brain to actually do something interesting, something that humans could do before but now AI can. And then finally, you're going to actually take that work that you just performed and you're going to probably log it, record it somewhere again in your enterprise systems. So any agent that you build requires that connectivity to your enterprise knowledge. We do a good job of that because Glean comes with hundreds of these enterprise integrations to most common enterprise systems. So if you build an agent and you need some data that you need to access, it's very easy and seamless in Glean to do that because Glean's already connected to all enterprise systems. The second thing is that we actually now allow you to take actions. You want to send an email, maybe you want to update an object in your CRM, you want to go and file for PTO, which requires you to take some action in a system like Workday. So there are all these actions that agents need to take to actually perform work. And so Glean has actually come out with a very extensive actions library. We have thousands of actions across hundreds of different enterprise applications that you can take to make agents do work. So those are some of the core building blocks.
So this action library uses the Glean enterprise graph, which is new for you guys. And is it behind the scenes? So from I'm an end user customer or I'm the user, 'Hey I want to take PTO, take PTO, tell people', so to me it should be nothing. To the user, what happens behind the scenes? Or do I take action or just prompt?
Yeah, well you just talk to an AI system. So Glean assistant is your go-to AI companion. You go to Glean assistant, you ask it questions, you give it tasks, and it's going to actually perform that. And behind the scenes, it's orchestrating the right agents to do the right work. Now this example of 'Hey file a PTO for me' doesn't require much intelligence, it's basically you just need the ability to actually make the right API call.
That's pretty straightforward. But another request might be, 'Hey check Arvind's calendar on September 24th, is he available from 10 to 2, and is he in the area?'
That's right. And that's also reasonably deterministic. So you can actually do that. But there are a lot of other agents which actually are doing work that requires judgment. Even this one, for example, let's say that you want to schedule an urgent meeting and you want to do it in the next 3 days and my calendar is always full. So now you have to actually decide, well I need to think about, is there something that we can push out and make room for this new important meeting? So when the tasks start, the graph actually comes into play. You have to deeply understand a business how it works. And maybe I'll just add one more thing.
The agents so far, there's been struggle. There are basic frameworks available but agents have often been misfiring because they just don't have enough context and they have not learned how humans actually do work inside the enterprise. So it's not just about connecting with the enterprise knowledge, the mechanism, the mechanisms are APIs but you're getting at context.
Context and actually so one thing is you know of course reading the knowledge, reading the documents, understanding how the company works, but the other thing is actually understanding humans and their behavior and taking day-to-day tasks and how do people actually complete those tasks. So that's the new level of understanding that now Glean is building, which is where we're going, understanding every job that is to be done inside the business and we're looking at how humans are actually performing those tasks and then you use those as the baseline, as the learnings that AI can now use and learn from human intelligence and actually start to do that work itself.
I want to ask you about signal versus noise. In fact, Silicon Angle's original motto in 2009 was extracting the signal from the noise. Signal-to-noise ratio. You kind of pointed that out in that example of, 'Hey, I might want to push something out.' Yeah. What's more urgent? That's contextual, too, because you have dependent variables going on here. So I want to ask you about that ratio signal. And also, if you go back to the old search days, which you know, you invented it with Google in those days, there was always two variables, context and behavior. Contextual and behavioral. Some would optimize for one or the other, whether you're search, but now you have both going on. The behavior of the work and the people and the context are two critical variables. How do you look at that? Now, it's complicated in the enterprise. You want to schedule something and figure out what your priorities are, what your signal noise is. So how does it know what your noise is and what my signal is? Or is it just in the data or is there more work that needs to get done?
Yeah, I mean look, enterprises have tons and tons of data. If you have 10,000 people and they're working eight hours every day, the amount of information that gets produced, the amount of decision making that happens, it's at a very large scale. And you have to build certain things as core abstractions inside the company. For example, you can say that 'Hey look, I'm going to actually identify the hundred most common jobs where our enterprise collectively is maybe spending 30% of all of their human time.' So you get those hero use cases, the key tasks which are very repetitive in nature, and for those you start to build very deep understanding of how those things happen so that now AI can actually take them over and do them independently. But a lot of those things, AI is not going to automatically build it. It's not going to be the case that suddenly from today to tomorrow, humans are doing all the work and tomorrow AI is doing all of it. It's not going to be like that. You'll have to actually still go and take specific tasks, specific business processes, and you're going to have humans that are actually going to go and build that agent and actually inform AI.
There's a lot of data that's available but there's still some instruction that's going to come from humans. Well, that's what I was getting at with the signal and noise because that context behavior, you got to nail that in the day. That's reasoning, super intelligence road to super intelligence for the enterprise. But then it's the individual because you're announcing the upgrade to the assistant, you got agents now, and you got the data platform. Yeah. With the graphs. On top of everything you got, that means that you're going to be in a position to set the table for that use case. Where are you seeing the enterprises on their journey there? Are they? That's a great north star but you're going to ship this product. What are your hopes and ambitions for the product immediately? What's, because are they ready basically? That's my question, are enterprises ready for this?
Enterprises are very eager to deploy AI. This is actually one thing that's universal. It doesn't matter what industry vertical you are in, everybody knows that AI is going to change how we work and everybody wants to be prepared. So there's a lot of appetite for bringing AI into their company. And even though it's hard, it's difficult, and you're not exactly getting all the ROI on day one, that's not making anybody put brakes on their AI initiative. So there's a lot of appetite. And companies are thinking of it in a few different steps. First is, 'Hey can I just get AI education to happen as we talked about? Let's bring some tools that people can make part of their day-to-day work and they start to learn AI and leverage it.' Then typically what we've seen is businesses are actually now creating an AI center of excellence. The AI center of excellence basically goes and works with all the different business units and departments, they're accelerating there basically, and they're picking the top 10, top 20 agents that they want to go and build, and then work starts to happen on those. So there is still phenomenal progress that is being made and the results are showing up in pockets. For example, when you look at engineering, software engineering, customer support, business sales teams, go-to-market motions, now agents are everywhere actually making a lot of work that we used to do.
And they need the data platform to do that. So I want to ask you what do you think the future of the data platform is? If you had to kind of just throw out your vision and statement around, if someone asked you 'Arvind, what's the future of the data platform?' what would you say?
Yeah, well I think you will see, if you think about data platforms in the pre-AI world, we have these warehouses from companies like Databricks, Snowflake, Google, Microsoft, right? And you take your structured data in the enterprise, you aggregate it from all the different places and that has been your data platform that has powered a lot of business intelligence. But the future data platform looks very different. It's a data platform that's meant for AI use cases. So number one, this data platform actually aggregates unstructured information which is 90% of the enterprise's data. So the new data platform is actually going to contain both unstructured and structured information. It's also going to have a very different paradigm in terms of how you access this platform. You're not going to be issuing SQL commands to retrieve data. It's going to be very conversational, very informal where you can actually just express in natural language what you want.
And the enabler on the tech enabler, that is what graphs? What's the key technology enabler?
