Glean:GO AI-Native Showcase  🔗 **Source:** [Glean:GO AI-Native Showcase](https://www.youtube.com/watch?v=kIfYH2JIDus) [00:00:02] [music] [music] [music] Good morning. We will be starting our keynote in about 2 minutes. Please take your seats. Thank you. Heat. Heat. N. [music] [music] Heat. Heat. N. [music] >> [music] [music] >> Heat. Heat. [music] Heat. Heat. Heat. Heat. >> [music] >> Morning. Good morning everyone. Welcome to day two of Glean Go. Woo. >> [applause] >> Yesterday you you heard from some of our customers and partners and we shared a number of advances our teams have been building in AI. As VP of partnerships, I often think about how Glean does not work in isolation. That's our architecture. We're the layer that sits across everything a company knows. Which means [00:03:08] we're as only as good we're only as good as the ecosystem we plug into. Every model we serve, every system our agents act in, every control that keeps customer data where it belongs. Someone else in this room built that. We're trying something different today with this AI native showcase. We know you are evaluating many AI tools for your environment and we want to be a place where you can see some of that great innovation. So today we are bringing you eight amazing AI native partners, eight companies, eight minutes each backto back. 8 minutes is not a lot of time. It's not much room for eight teams who spent years getting to this stage. You're [00:03:52] going to hear about what they do, the outcomes they're delivering for their customers, and a little bit about how they work with Glean. So with that, let's get this started. I'd like to bring out Dimma Juggov, co-founder and CTO of Fireworks AI. [music] [music] Hello everyone. Good morning. I have an honor to be the first talk of the day. So let let's get started. I am from Fireworks uh and I'm going to talk about future of specialized intelligence. So when we think about future of AI, the common narrative is AGI. There is a single god level model which is good at all tasks and solves all problems. It's probably concentrated to a handful of [00:04:42] frontier labs and everybody else is just rents their intelligence from them. There are several issues with this vision. One of them is that majority like intelligence is basically representation of data and estimates say that only maybe 10% of data is publicly available of on the internet. 90% is locked in in all the great companies and all the private data which ex exist beyond the internet. Second, you should really ask yourself whether this future is something which you feel great about where when there is such a big concentration or maybe as some people say in extreme that intelligence is provided by only one company which [00:05:22] represents entire economy. At Fireworks, we believe in alternative future. the future of specialized intelligence where every company uh is able to use their data which they own and are able to customize models and tune them for the use cases which they care about. They are able to leverage 90% of data which is private to them. And the future is not a single AI but millions of distinct specialized intelligences. What powers this vision is open weights models and in case you haven't been paying attention uh right now open weights models especially in the past few months are rivaling the closed frontier models from the labs. We see it [00:06:02] across many use cases from coding to design to general to general writing to uh legal benchmarks etc etc. In fact, in Fireworks, we run a lot of evals across multiple models and we've seen models like GLM or K3 actually rival Opus and F and Fable on many tasks. This is not a just forward-looking pitch. This is uh something which we are do today with many of the AI native and digital native companies. To give you some example, we work with companies like cursor for example to post trainin their composer to empower ar infra for that uh in healthcare with companies like doximity where specialized models can open for outperform GPT and cloud [00:06:48] for for their tasks or latest examples with Harvey where we partner on post training uh postraining open weights models to uh to build their tenet uh uh tenant model which shows frontier level legal intelligence on many benchmarks. I have to say that this open weights open source is core to our team. Before starting fireworks, I was one of the core maintainers of PyTorch which is for those you don't know is open source deep learning framework which pretty much powers a lot of AI these days and many of my co-ounders were also coming from PyTorch team. So we built a open foundation for current deep learning or AI revolution [00:07:32] and we are bringing the same philosophy to building fireworks. In in the past few years we scaled company a lot. So today we are running tens of trillions of tokens per day. I think the last number we published was around uh 30 trillion which is at the scale of [snorts] uh many cloud pro cloud providers of Frontier Labs. So what is what does actual product look like under uh underneath? Uh comes in three buckets. Fireworks inference which is best performing inference platform for open models or customized models. Fireworks training which provides you a complete solution for any level of expertise to to optimize model for your [00:08:12] use case and on your data and fireworks nexus which is our hardness routing play which allows you to improve internal productivity by intelligently switching between models. So on inference side uh we proud pride ourselves in day zero support of uh all open uh all open weights models for example GLM 5.3 flash from yesterday uh what what is really core for us is specializing to use case even when it comes to in just pure inference uh we jointly optimize many levels of the stack to get to the best cost uh speed and accuracy for your use case and finding the right para frontier point which is import which is important to [00:08:53] And of course scale it with enterprise grade security and full full scale of options to deploy it on your on your hardware or on multiple cloud providers. Uh what it brings in practice is that we often can tailor deployment configurations for your use cases and get multiple x wins uh either on cost or speed or both. Coming uh coming down to post training uh important thing to realize that post training is kind of hard and Frontier Labs uh do training on uh kind of doing the full control of their of their stack from tra from trainer to roll out to inference and making sure that all of that works in works together well from numericics to [00:09:38] optimizations to training algorithms and produces the good result. With Fireworks, we are bringing the same level of advantage, the same level of quality uh for building your own models. So, we optimize training and inference. We offer multi-level APIs whether you're just trying to try out post training or you are expert researcher and want much fine grain level of control. And we have many options of how to deploy and run this. Well, it means a practice that instead of having to go procure your own GPUs and figure out how to set it up, you can focus on what's important for you, which is curating data, figuring [00:10:14] out what is the right evals, what is the right task to optimize for and speed up iteration cycle for getting to the great model for your use case at and at any point if you need more level of customizability, you can drop down to to the level below and be able to tweak tweak the details. Finally on Nexus uh Nexus side our our focus is to enable internal productivity in the companies use cases like coding or co-work and uh and the trade-off between kind of cost control and innovation. You probably seen a lot of shift in in the past half a year from token maxing to value maxing right where people actually ask what whether [00:10:54] particular uh spend is has high RAI or not or not. So our approach here is intelligent routing. So we plug in whatever harness you prefer and whatever use case uh use case you have and do intelligent routing between open open weights models and frontier models depending on the use case with zero workflow disruption and full control and ROI analytics which you would expect from enterprise uh software. uh by eliminating this trade-off between do I want to use just close or open and simplifying the workflow, you can still use the same ai agent powered for close which you have in in your company but now with much lower much lower cost [00:11:34] without changing details of your harness and uh and other bits and finally if you get more uh if you get more data for particular use cases you can jump into post trading and customize your models further. So I encourage you if you haven't been using open models give it a try. Uh go to fireworks try to some of the open weights models out of the box try post training. Uh the exciting thing with Glen is that we are powering uh inference for Glen customers through Glen. So if you're if you want to use it through Glen you can reach out to your account executive to discuss more. And uh for everyone here uh you you have [00:12:12] Glen Go uh code which should give you some credits on fireworks if you want to try it out and start building. That's it. Thank you. [music] Thank you Dimma. Now, we're going from the technology powering AI to what happens when AI AI helps people turn ideas into something real. Please welcome Grant Lee, co-founder and CEO, and Diana Berlin, GM and head of product at Gamma. >> All right. Hi everyone. My name is Grant. I am the CEO and co-founder of Gamma. >> And I'm Diana. I lead our business and product teams at Gamma. We are so excited to be here and to partner with Glean. With Glean, you already have all of your context in one place. And the [00:13:08] question then is how do you take all of that context so that you can make decisions to drive your business forward? At Gamma, we believe visual communication is the last mile. Human progress has always been pushed forward by our ideas, but we often times struggle with our ability to communicate them. When we think about how do we help get what's in our head into someone else's, how do we help others see what we see? That is the challenge. We know that ideas themselves do not win on merit alone. So when we think about our ambition, it is about really helping companies transform the way they communicate their ideas. Our goal is to help you take all [00:13:53] that context that's in glean and make it just much more tangible, make it much more actionable. So we've been building an endto-end platform centered around visual communication because we want to be this last mile. We want to help you bridge that gap. Today, we already power 40,000 teams with a better way to tell the story of their work. And we're here because we want to help you tell your story. Next, to summarize, basically today, we think many teams really face the same challenge. And with Glean and Gamma, you're basically bringing together two really powerful forces. You have all the context and data that runs your business and now you [00:14:41] have a powerful visualization engine to be able to make that much more actionable. With that, hand it over to Diana for a quick demo. >> All right, so you know, a live demo really benefits from being real and we really truly do have our go to market all hands later today. So let's take a look at how we can use Glean and Gamma together to tell a story to our team that drives progress forward. So using Glean, as you all know, all of your business context is in one place. that turns heads, gets results, and gets everyone focused on the right thing, whether that's the board or your team or a keynote. So, we pull all that context in to our [00:15:34] go to market all hands. And this is where the magic starts to happen. With templates and gamma and your brand in place shared across your whole team, you can pull all of that context from Glean and put it onto slides in a way that is by design. It's on purpose and it's something that you can guide with template instructions, with workspace instructions, and with the ability to really give context on how you want your context flowing in to that end product of beautiful slides in a repeatable way. On this side, we have data from financial sheets as well as pipeline information from HubSpot all flowing into a single slide [00:16:12] automatically. Let's take a deeper look. Let's break out that pipeline information into a table. You know, visual communication is interesting because it is a language all its own. It gives us cues for what to focus on and what kind of conversation to have. A table creates a conversation where everyone is looking at the same information and forming opinions on what's worth talking about. Stoplights like this help to queue people in to what is worth discussing as a problem worth solving. And it really is that neutral, you know, everyone on the same page, being able to visually see and be queued to have the right conversation [00:16:50] and to drive the right decisions that is the last mile where visual communication makes the impact. you know, bringing in stories as well. Visual communication is such an incredibly powerful tool for storytelling, and storytelling is what moves individuals. It's what moves teams toward the action that matters. Here we have sales wins from Slack as well as that same pipeline information coming together backed by a slide level template instruction that knows exactly what kind of story we like to tell, but then pulls that together based on real-time data in the moment. And then bringing it back to numbers, you know, the world runs on numbers and [00:17:29] we've got to share that source of truth that just lets us know how we are tracking. Fortunately, in this uh demo, we're tracking well. Um NRR is high. We're pulling together financial data that pipeline information again and then more of those stories as well so that we can all understand from the highest level to the ground level how we can drive the next right action as a team. And then really bringing that back to team motivation. There's nothing like being able to see a goal and then knowing how you're tracking toward it. And so always bringing it back to those stories. You know, I truly feel that um stories are what matter. It's they [00:18:05] they're what make the difference. And being able to pull data and stories together into a format designed for storytelling is how that last mile takes place. Again, the last mile is where the difference is made. With Gamma, you get on brand production every time. One client of ours saved two months of work per person per year. Insights unlocked companywide. One client's reporting cycle went from 3 months to 5 days. And real outcomes across teams. One client sales rep saw 4.5 times improved win rate. And what's so cool here is that you don't have to do it alone. The best outcomes happen when your entire team is [00:18:42] on the same page. But we are also on your team to help drive that. You know, part of why, you know, I believe it's been such a shame that slides have been bundled inside and even buried inside the productivity suite and now the everything apps is that they really are their own strategic initiative. And you know, because everyone cares about them. Everyone's got an opinion on slides. Everyone's got an opinion on brand and storytelling. But so far without a dedicated vendor to take you end to end, all of that just dies on the vine because it requires aligning. Literally aligning as a team to figure out what is that story, what is that brand that [00:19:15] you're all going to align on and automate. But the upside is that when you have a tool like Gamma and you have a way to connect that to glean, you're so close, you can taste it. You're so close to that outcome of all finally being aligned in an onbrand onssage way that it's finally motivating for every team to work with us to put in the work to get to that state where you have repeatable flows that everyone can get excited about and that unlocks so much energy in your business. So, I'll kick it back to Grant for what's next. Awesome. Thanks, Dana. So yeah, really quickly, uh, we have a lot of really big updates coming in a few [00:19:54] weeks and we're going to be super excited to be announcing them ourselves. So please stay tuned. Our team is here and if you are interested in getting a sneak preview on any of that, we love a chance to show you. And uh, thanks so much for for having us. Thank you. Thank you, Grand Diana. You can see how Glean brings the the whole intelligence together, right? It's not just about models or the harness or the applications. We bring that intelligence to a great product like grandma that can then serve beautiful visuals for you. We've seen now let's turn to an equally important question. How do we make sure the data behind [00:20:37] those ideas remain trusted and protected? Please welcome Anu Sharma, co-founder and CEO of Skyflow. >> [applause] >> All right. Well, thank you so much for showing up in such large numbers. Uh, I'm Anu. Uh, I'm the co-founder and CEO of Skyflow. We are a company focused on only one thing, protecting the data at runtime. So think about when you're using glean and you're querying the data, you're ingesting the data. How do we make sure that you get very very fine grain controls work for you? And what I'm going to do today in the next 6.9 minutes is give you a couple of customer stories and then talk about how our glean integration works. [00:21:32] So what does this mean? Well, we're in the business of securing the flow of data across data stores, models, and agents. So, if you actually think about what's going on in the market right now, a little bit about my background, by the way, um is we finding that companies of all sizes, starting from the tiniest companies to the largest companies, are basically trying to do two things. So, some of you work at very large companies the size of companies like Walmart and Karnney and Anderson who are trying to deploy AI and you're thinking I have so many employees. Some of them have no idea how to use this thing. Some of them [00:22:11] going to make mistakes. Some people you have in your company are extremely smart and savvy and they may be able to both use these products in very intelligent ways but they can also do some very intelligent bad things. Right? So imagine you being able to query any and all of your company's data. A lot of good things can happen and our job here is to make sure that the bad things don't happen. There's a second class of companies which is startups. Some of you work at startups. You're either selling to the same enterprises that are trying to deploy AI or you're building entirely new things like this company Hark that's basically [00:22:50] building a new kind of product in the market. So when we talk to the largest customers, they basically ask us about two questions. They're saying when I'm deploying products like Glean to you know build my AI uh footprint in the company, I want to be sure that I'm following the rules around security and sovereignity. The challenge is that agents break data security. How many of you here have heard about the hugging face incident a couple of days ago? Right? Who amongst us can say that my model is not going to reach back out and do something nefarious? This wasn't the case in the world that I came from. I spent most of my career at [00:23:31] companies like Oracle and Salesforce. And in the old world, life was simple, right? You could even barely search for your expense report. How many people here have tried searching for an expense report in Concur or an invoice in Aribba and spent years? Well, those were the good old times from a security perspective because if you literally can't search for something, bad things can't happen. Rule based access control at the level of rows and columns. Data is so locked up that you can barely use it. And then comes green and products like that that basically make it easy for actually to reach out and get the data as we need it. So you can actually [00:24:10] find your expense report. Unfortunately, that intern that you just hired who's just curious can also now say, you know, let me just see how many stock options does my manager have. Let me just see what is the last contract that we did that got cancelled. This is a problem that's nondeterministic because agents are nondeterministic, models are nondeterministic. The second problem is the data is no longer cleanly separated in rows on columns. You can't say column number seven has social security numbers, column number three has credit card numbers. If you think about AI from my perspective, it's literally a computer [00:24:50] for PDFs, right? How many of you here are feeding PDFs to your AI platform, right? When you do that, there is no word level security. There's no rows and columns in a model. There's no rows and columns in a PDF. There's no rows and columns in a vector store. So somehow we have to go out there and solve that problem. If you're running a large enterprise and you're deploying these products, you're basically trying to figure out this is the new architecture. I want to be able to connect. Thanks to green, you can do this in a holistic way to any model, any data store, any cloud. Fun backstory. A year ago, I s showed [00:25:31] this slide to my board and they said, "This is not going to ever happen. We're going to live in a world of one model per company and one data store." I think right now the jury is out and everybody's decided that actually this is where we are headed and thanks to that companies like Glean and Skyflow have an important role to play in your architecture. So what do we do? We do three things. If you're sending your data into a rows and columns paid data store, we can protect the data by encrypting and tokenizing very specific values like social security numbers, credit card numbers, phone numbers, email addresses. We do it [00:26:13] in a manner so that it can be still computed upon. You can still query and use it except the bad guys can't get access to it. Same thing for training data or the data that's going into your glean at the time of ingestion. What SkyFlow does is it goes through your PDFs, it goes through your CSVs, finds all the occurrences, basically scanning and finding the sensitive data, locking it up. We call that tokenization, which means we basically take your real email address, replace it with a fake number. We take your real social security number, put in a fake social security number. The value of the product is while you're doing that you [00:26:54] can still chat with the customer service agent that you're building on top of Glean or you're building on top of another platform. You can query your data. One of our customers who's actually here is Kernney a shared customer with Glean. What they're doing is they have millions and millions of PDFs that their consultants need to be able to use the knowledge of. Now just because a consultant can use the knowledge that's across my company doesn't mean that consultant can also see the details of another client's data. So think of this as a client separation problem. You know every one of you here works for a company that [00:27:34] either needs to separate data on a client by client basis or a user by user basis. We've been partnering with Glean for about a year. We were able to do this partnership. We got this customer signed up and they were live and working within weeks. In fact, we just did a press release with them yesterday and they are here in case you want to talk to them. We also run our platform in 30 regions. What that means is you can meet your data residency and sovereignity needs. What that means for you as a green customer is you can run one instance of green in North America and still serve the needs of employees that are sitting [00:28:17] in Saudi Arabia or Vietnam where they have local data residency requirements. You don't need to replicate your snowflake, your glean, your service now in every country that you operate in because we can keep the sensitive data isolated and protected in region. All of this has been built with the largest enterprises in mind. You know, some of our largest deployments are customers like Walmart and Flipkart that have literally hundreds of millions of users data. We protect it at runtime in line, which means it has to be super fast. It has to be super secure and it's always work. That's what we do. Thank you so much for [00:28:59] your time. >> [music] >> Thank you. Thank you, Anu. As you can see, you know, as you know, Glean is built natively secure. We have Glean Protect to add an extra layer, but we're also an open platform allowing you to plug in whatever security tools and products that your organization needs as a Skyflow is a great example of that. So the next question is, how do we turn powerful AI models into reliable productionready experiences? Please welcome Philip Kylie, author of inference engineering and head of AI education at base 10. >> Hey everyone, thank you so much for being out here this morning. Uh, Fort Mason is actually one of my favorite [00:29:48] venues in the entire city. So I was super excited to be invited here to speak today. I'm going to talk a little bit about influence engineering. So, my name is Philip. I work on inference at base 10. And I'm here because a few months ago, I kind of messed up. I made a mistake. Um, and I I published a book, like a paper book with words printed in ink about inference, a technology that is changing faster than just about any technology in history. Uh so you know out of respect for the trees uh I wanted to come out here and give a little update to some of the inference engineering techniques uh that have become more apparent in the in the last [00:30:31] month especially as they apply to a handful of Frontier models that have been released recently. So we're going to talk about a couple of principles that haven't changed and then we're going to go into Kimmy K3, Deepseek V4 and GLM 5.3. At the end there's some free books. Cool. So, principles of inference engineering. You know, we talk a lot about model performance techniques at base 10. Um, and I I write about them a lot. And fundamentally, some techniques manage trade-offs and others push out the entire frontier. So, inference really is all about trade-offs. You know, are you going to get lower latency or higher throughput? [00:31:12] Are you going to get lower cost or higher quality? And being able to identify the point along this frontier that you want to be at and apply the correct techniques to reach that point is inherently valuable. But as inference engineers, we also want to push the frontier. uh we want to develop new techniques uh both by applying new hardware, new runtimes, new konels, new optimizations, new research that give us global gains that we can allocate as needed towards some combination of latency, throughput, quality, cost. So let's look at what that those techniques are in the context of some recent models. Starting with Kimmy K3. [00:31:55] Uh Kimmy K3 is an open frontier model. Uh we've we've often times talked about models that's like, "Oh, this is the best open- source model." Or for an open source model, this is pretty smart. Kimmy K3 is just straight up a good model. It's it's a front tier model. And the uh challenge with running it is three-fold. First off is size. 2.8 trillion parameters, 896 experts with only 16 active, which uh 16 routed experts and and two shared. Um, this unprecedented size requires both new hardware and new performance techniques to run reliably in production. The second is architecture. To scale a model to this size, the Moonshot team shipped [00:32:38] with a new attention mechanism, stable routing for sparse experts, and a two-part KV cache, which means that every aspect of the inference engine needs to be updated to support this architecture. And the third piece is context. When we have a 2.8 8 trillion parameter model already straining the system to its limit. It's hard to do that and support a full 1 million token context window. So this required full stack system updates uh across the entire inference service. Um a lot of it actually was at the front end. So Kimmy K3 has some unique parsing logic. It has some new architecture features. It has some new special token mechanisms. And if you're [00:33:24] trying to support longunning agentic tasks, if you're trying to support function calling, if you're trying to support highquality inference, you actually need to do quite a bit of work on the deterministic front end in addition to the inference engine that actually generates the tokens. But we were able to do that. We were able to work with the Kimmy team um to offer day zero support for Kimmy K3 on base 10 as a launch partner. Um, this is one of my favorite AI models ever because it is so incredibly powerful and we're able to serve it at a at a speed, a cost, a scale where it's actually useful. So, I definitely encourage you [00:34:00] to try out Kimmy K3 and do some uh read some of the stuff we've published um on the technical reports about the work that went into actually getting this thing up and running because it was really hard but really cool. Another great model, DeepSec V4. This is actually a family of two models. Uh there's DeSseek V4 Flash and Pro. Um Flash is about 16 to 1/8 the size. Uh they're both very high quality models and they're MIT licensed which means you can take them, you can run them anywhere. You can post train them. You can pretty much do whatever you'd like with them. The thing about DeepSeek is that as a research lab, DeepSeek also [00:34:42] really goes into specialized architectures. If you think about DSA, deepsea sparse attention, there you go. Um, so there's quite a bit of work that happens at the inference engine level to support these models as well with their novel architectures. Um, but what's cool about these recent releases, something that that DeepSeek does very reliably is they release checkpoints called preview checkpoints. And we were able to work with these preview checkpoints back between April May of this year. And we were able to understand how to run these architectures. So when the full trained models came out, the flash at the end of [00:35:23] last month, the pro earlier this month, we already have an inference service ready to go that can run these models with frontier speeds and high quality. Uh the the DeepSeek V4 Pro model was a especially good release. Um we see for the first time native MXFP4 weights which is very good for uh large scale inference. It means we don't have to do any post uh uh any uh post-training quantization in order to in order to run these models efficiently on Blackwell. And it also shipped with its own speculator u a first-party speculator that uses diffusion language modeling to generate sequences and we were able to achieve leading [00:36:06] performance on these deepseek models because again we've been investing in this architecture for a long time. Uh that's that's something that we're consistently seeing as a theme throughout the inference industry is a winning architecture is going to get built on and built on and trained on. And so if you can invest in getting good at winning architectures, you're going to be able to run an efficient service for a long time. The final model I want to talk about is not actually out in open weights quite yet, although the flash version is, and that's GLM 5.3. So, ZAI released this model on their APIs and said, "Hey, it's [00:36:42] coming soon to open weights." And this is another example of a really popular architecture getting better with post-training. So, we did a ton of performance work on GLM 5.2 across the layers of quantization, speculation, parallelism, disagregation, inference service configuration. We even added vision to the model. Uh, one of our research engineers trained an 80 million parameter projector that is able to use the actually the Kimmy K2.6 vision encoder and project that information into the unchanged weights of the GL1 5.2 model. And we figured out how to serve this at multiple points on the curve. We were talking about that [00:37:23] trade-off between throughput and latency. We were able to create a standard API as well as a fast API to give multiple options along that curve. What's incredible about GLM 5.3 is it's the same architecture. The weights are the same shape. They just have different values because of post-training. And when you're in that situation, every single piece of work you did on the previous instance of that architecture reapplies. Now, do you have to retrain the speculator? Yes, because the hidden states changed. Do you have to, you know, rerun your model optimizer? Yes, you do. But the patterns and the techniques stay the same. and you can [00:38:01] pretty confidently apply everything you did in the past to this new version of the model. So, thank you so much for your time here. If you want to learn more about inference engineering, uh, hit up that QR code. Um, we've got a free book that I wrote called Inference Engineering, um, that's been doing the rounds in the industry. Um, and if you want a paper copy, I didn't bring any with me today, uh, but hit me up on Twitter or LinkedIn. I'll get one sent over to you. Thank you so much. Have a great time at the conference. Thank you, Phillip. [music] Thank you, Phillip. And thank you, B 10. They are one of the providers that help us bring open source [00:38:42] to you every day. Um, there's a lot on models. As you know, we provide the best models to you. We find the best model for the best job and we have teams of experts that are constantly optimizing the different models. So all that, you know, as he gets to all those details, don't don't worry, we got it covered. Okay. Um, now let's look at the connective tissue that helps AI work across systems that business rely on every day. Please welcome Roma Cestier, founder and CEO of Stack One. Thank you. >> Thank you. Thanks. >> Hi everyone. I'm Roman. I'm the co-founder and CEO of Stack One. Stack one is the tool gateway for your AI [00:39:26] rollouts in the enterprise. What does that mean? Most companies have AI systems like Gle, but they're not connected to all of the systems that their company runs on. And this is becoming more and more important as Glean announced at this conference that they're uh launching things like proactive agents where you really want agents to be connected to your entire systems in your company. Now, I want you to think about this for a second. Glean calls it your AI coworker, but imagine if you hired an intern. Um, they're very qualified, super eager to get started, and on day one, they hit the ground running, and they start working on all [00:40:05] the tasks that you've assigned to them. But the problem is that at the end of the day, they get back to you with a report of everything that you now need to do in all of your systems because they just weren't given access to them. That wouldn't be super helpful. And in a real world, your IT team would have provisioned a special access right that's scoped to only what they need to do in those systems like your ERP, your CRM, etc. And this is where we are with AI today. Companies are moving from being AI supported where agents were mainly reading data from their systems to becoming AI native where the agents are actually completing the entire [00:40:45] workflow from end to end. And this is where the true value actually resides in AI. But the reality of deploying this in enterprises and um being a sponsor at the conference, I've spoken to many of you guys already and it's it's a common pattern is that there are three big problems to overcome in order to make this a reality. Number one, you need to be able to connect your agents to take action into all of your systems. And oftent times the reality of this is that you have to spend months and months building custom connectors into all of your systems and even once you've built them you need to maintain them which is [00:41:23] an enormous task in itself. The second problem is that if you think about it giving agents right actions into your systems brings in an entire new category of problems that your IT and security teams are extremely worried about. And what I keep hearing is that many companies are allowing agents to read but not yet to write because they don't have a solution to enable the agents to safely write into their systems and take action. And finally, what we're seeing is that as agents complete entire workflows, they the the token costs increase and there need to be more levers, including Gle's awesome product with the model [00:42:05] routing. There need to be even more levers that allow you to save on tokens to make this an economic reality because it doesn't make sense to automate a workflow if it costs too much. Stack one is the platform that bridges all of these three gaps. If Glean is your control plane where you're already building agents and your teams are building agents as well as the context layer for these agents, stack one is the action layer that lets Glean take action into your systems safely and efficiently. As you can see on the video here, what's happening is that stack one actually lives natively behind the scenes with a single MCP that gets added [00:42:44] into Glean and connects it securely to all of your systems. So the user experience as you're deploying this to all of your employees and all of your teams is completely seamless as you're already accustomed to with the Glean platform. But stack one was designed for IT and security teams to manage the build, the customization and the governance of connectors in a central place so they can roll it out to all the employees safely and this connects seamlessly into the glean platform. So if you think back to these three problems that I was mentioning earlier, stack one is the platform that solves all of them in in one. Number one is [00:43:24] connect. Out of the box, we have built 450 connectors to some of the most complex enterprise systems like SAP, workday and netswuite and more. These connectors carry deep read and write actions into all of these systems like updating employees or updating financial records. But in addition to this, we know that your businesses run on heavily customized enterprise tools as well as customuilt internal tools that have APIs. So stack one has an AI connector builder that allows you to build custom connectors in hours with one engineer resources instead of months with an entire team doing this. Now what does that mean? That means that you're able [00:44:09] to actually completely cover the entire state of your company with all of the systems that your company actually runs on and connect it to the systems the AI platforms like Glean that you're already using. The second part is secure. And what we've built at stack one is a brand new set of governance features that were designed from the ground up with the use case of agents taking action in mind. It's not just inheriting read permissions from what's already in your tools. That would be too easy. It's actually going down to what fields can be updated and what data can be written into these fields. Let let me give you [00:44:50] an example. If you're building an agent that can provision user access into your tools, you want them you want the agent to actually be able to create a new user in the tools, but you may not want it to be able at all under any circumstance to give users admin access. And this is what IT and security teams have in mind when they're blocking AI rollouts and they're saying, "We want read, but we don't want write." It's because they're worried that these use cases don't currently have a governance layer that allows them to control and safely roll out agents that take action. The second thing is that at Stack One, [00:45:27] we have built our own AI uh engineering and research team and we're proud to have developed the leading model for prompt injection detection uh that is now open sourced and built into all of the stack one connectors. What that means is that your team, your security teams and your IT teams can deploy these connectors with more confidence that very common um risks like prompt injection attacks will not be a thing in your company. And finally, last but not least, optimize. As MCPs grow and stack one is one of the most uh heavy providers of actions for most agents with the 30,000 action catalog that we have, it becomes [00:46:11] unthinkable to load all of these actions into the agent. We've designed something called tool discovery, which is a high quality semantic search that actually allows the agent to search the catalog of actions that stack one has instead of preloading everything into the context window for you. That means that you can actually build very powerful agent without the overhead of token costs. And you don't have to take my word for it. We have hundreds of companies that are currently running on stack one across industries like healthcare, retail, manufacturing, professional services, SAS, and more to accelerate the deployment of AI within their [00:46:50] companies in a safe way. Here's an example. We recently onboarded a leading uh cancer research network uh that has more than 200 employees and they had purchased AI platform licenses already but they weren't able to deploy it to their employees because the AI platform was not connected to their systems. They needed two custom connectors to UKG and to Netswuite. One because they needed it to be HIPPA compliant and they needed to filter out PHI data. The other one is because uh Netswuite was heavily customized in their company. Stack one was able to build these two custom connectors plus 13 others within the span of two weeks [00:47:31] and roll out their AI tools to their 200 employees. And just for context, this is something that they were planning more than six months of build internally before they met stack one. That's how much they were able to accelerate their deployment. If you want to see an example of this, you can actually log into the attendee portal right now and see three examples of agents that we built on Glean with the stack one layer added to it. One for new higher provisioning, one for access and IT ops, and one for month and close um that uses all of the features I talked about. Now, I'll leave you with a question. What would change in your [00:48:07] business today if you could connect your AI tools safely and efficiently to take actions and complete entire workflows? If you want to see this in action, you can come and meet us in the other hall. We're an exhibitor in the expo hall and you can get a custom demo as well from uh Jacob and Alex who will uh do a demo for you. Thank you so much and enjoy the rest of the conference. >> [music] >> Thank you, Raman. Once AI can connect across the business, it can do more than move information. It can help us understand what customers are really saying. Now, I'd like to welcome Verun Sharma, co-founder and CEO of Interpret. [00:48:54] Verun. There we go. Come on. There we go. All right. Uh, good morning everyone. My name is Varun. I'm the co-founder and CEO of uh, Enterpread. Uh, we are a customer experience platform. Uh, and I'm here to talk about self-improving customer loops. So, before I get started, I'm trying to figure out, should I stand up or down? Okay, I'll stand up. Um, but before I show you how good my marketing team is at making PowerPoint slides, uh, I want to show you a tweet. So this tweet is from an account called Nissan Algyab which has some 50,000 followers. Uh and in August 4th they tweeted it's insane how lacky chat GPT and codecs get when you have a [00:49:39] long conversation. Then 10 days later OpenAI fixed this problem which is OpenAI is about to ship a huge performance upgrade to extremely long chat GPT and CEX conversations this week. And then they said they say a simple post can move mountains. And then one of the tech leads of Codex confirmed, "Yeah, we had no idea till you said it right now." What's interesting is, and I cannot believe I got auto logged out of my account. Uh um so what's what's interesting? Uh here we go. Sorry. Man, this is [panting and sighs] security protocols. Okay. So what's interesting is that we were able to detect this signal uh up to three months uh earlier um uh for for [00:50:54] them. Um and then like we there were there have been thousand mentions of this up to 3 months prior uh for uh for this solution. Oh man. So um um so there have been uh a thousand mentions of this up to 3 months prior that we were able to detect for them. Uh the idea is that like people who pay the bills for a business, they're telling us in real time exactly what they like, what they don't like, and those signals are, you know, once they detect it, it's easy to hand it off to AI agents to to do the work. But the idea is can we detect these signals as quickly as possible and at interpret we're able to do that because we have solved this [00:51:51] problem of like uh no matter the data sources of customer context we're able to deeply classify that in real time across at a very granular level and plot trend lines. So in this case like you know we have a five-level schema we're automatically able to create which updates in real time of like performance reliability task and execution performance uh you know long execution agency uh long execution latency uh task execution takes too long and then the the sub reasons beyond that. So we're able to do that in real time and because you'll do pro trend lines we're able to catch the signal significantly faster and those signals are available in any [00:52:28] control plane that you have whether it's like glean or any other solution that you have. So in this case we're able to detect the signal in real time. It would have been triaged and then all like the codeex team would have had to do is just hand hand off this bug or regression to codeex who would have automatically fixed it and done closed the loop. Right? So any customer work that happens it follows the same exact pattern of like detecting uh that something has changed in customer experience triaging it to ensure this something is real worth actioning on then handing the work which can be done by a human agent or an AI [00:53:05] agent depending on the context verifying that it was actually done and then closing the loop right and if you can do that like this workflow hap has been happening for ages this is how we serve our customers. What we take control of is like if agents can do all the work who's going to create the work and people who what so we take responsibility of like detecting and triaging customer signals in real time because we are ability to classify so deeply in real time and those signals can then be handed off to a human employee or AI agent to then do the loop and if we can do that then we can have self-improving customer loops that can [00:53:42] power like know like every company in every use case like another example If I could quickly show you and I got sorry I'm logged out here also again. Um um so uh and we can do this for like bunch of different companies. So like we have like companies like whisperflow who who use us to detect product regressions in real time. We have companies like Straa who have a self-healing help center because we're able to catch regressions on help center in real time and they can like have a self-healing help center. Uh, and if I could take a moment to tell my favorite customer signal from from the last few weeks, uh, we work with a Fortune 500 retailer. Uh, [00:54:20] and we were able to catch a signal that for one of the stores in Tennessee, they had a broken bathroom latch which was automatically routed to the right, uh, store manager and they were able to fix the latch. So, if you think about the diversity and the breadth of customer experience in real world, it could be it ranges from self-healing help centers to like self-healing bathrooms basically, right? like like catching bugs for an AI agent company to like in physical world improving that. Um and then the way we work with glean is like you know our ability to detect these customer signals which are present and which will happen [00:54:54] no matter what happens because your customers will talk about your product or service like all of these signals are available inside of glean uh that can then be acted upon uh as a control plane. Um yeah that's all thank you so much. [applause] >> [music] >> Thank you, Veroon. I've never heard that one before. Detecting uh bathroom latch is broken. Looking up the manager of the store, I'm sure in glean because that's where all that information is and then being able to send a ping saying go fix that. Um so that was that was very interesting. Listening to customers is powerful. Acting on what we learn requires making [00:55:38] sure the right information reaches the right people and stays [snorts] protected. Um we've talked about how we're an open platform. We have our security but also we let other people connect and brings additional security into our platform. Uh please welcome Vishal Gari CEO of Secllor. Thank thank you Zuben. Uh great to be here guys. Um my name is Vishal. I'm the CEO of Cycllor and we are a data security platform with a mission to allow enterprises to accelerate AI adoption. Uh I have 8 minutes today. So I'm going to spend a lot of the time talking about the bigger picture problem. How do we actually embrace AI [00:56:18] faster in the enterprise? I'll spend a few minutes just telling you guys what we do but a lot of the time we'll talk about glean and how glean's open platform allowed secur to really drive security of data into the enterprise. So first of all I think uh if you attended yesterday's sessions if you attended uh the panels and you heard what I heard which is that data is becoming the operating layer in the enterprise. Uh I'm a security professional and the security point of view we have not kept up with the way how fast AI has grown in the enterprise. We still think about network security firewalls. We still think about endpoints and and uh devices [00:56:57] in the enterprise. We think about cloud and identity. But the right way to think about uh really embracing velocity is to think about security and the data itself. And I think as the control point in the enterprise has moved from parimeters to data um the security has to move with it. And the problem's very large um and I'm just showing some stats here but 80% of enterprise data in the today is an unstructured format. It's not in databases. A majority of it is dark which means you don't know what's in it. Nobody's touched it for a long time. And you know it's not governed actively by anybody. lot of it is actually outside the [00:57:38] enterprise. How much of sensitive data actually leaves the perimeter of the company? It's hard to tell. And the average enterprise has 40 million different unique permission sets across the enterprise. Impossible to manage. And many of these are not set correctly. So with all this sprawl, um IT leaders, security leaders, they're asking the same questions. How do you understand what's sensitive? How do you actually make sure that you address the problems you have? How do you get velocity? And I think the most important question that they're still asking is how do you get velocity of AI adoption in the enterprise while keeping data secure? [00:58:16] And to do that, you have to really connect care, you know, very deeply with the system itself. The security and AI have to work together. And that's how secure and glean partner together. Um I'll tell you a little bit about us. The way we solve this problem is using our Armor data security platform. What Armor does is it actually discovers sensitive data in the enterprise. It understands what's in it by using small language models. So sensitive data actually never gets to public uh frontier models. We enforce security and we we show how you're actually um uh achieving compliance and security. Um and we do it using our data security posture [00:59:00] management product. Now this is a product that actually discovers we have connectors just like Glean uh to different data sources in the enterprise. We're able to discover information but we're also able to understand risk. We're able to understand what compliance regulations are in the enterprise. We're able to understand geographically what sort of um regulatory frameworks are in place and give you analysis of risk and sensitivity. We understand the content which is uh whether there's BII PCI in the in the data that you have. We understand the context which is is it a is it a business document? Is it a legal [00:59:33] document? And we understand the intent which is why was this document created? Is it to send to a vendor? Is it a PO that came in from somewhere? Is it an HR um you know hiring memo or something like that? And I think that context allows us to then actually apply labels and classification directly at the source of the data. So if you have unmanaged large volumes of data and you want to get velocity in adoption on that data, you need to be able to set the right context to the data and that is done by actually classification and and and labeling and tagging the data so that downstream systems can actually read that information. We're also able [01:00:12] to take the most secure information that you have in the enterprise and uh provide persistent protection by using digital rights management. That technology allows encrypted data to travel everywhere around the world inside the enterprise, outside the enterprise and allow you to control what you can do with the data, who can access it, what they can do with it, how long they can access it, where they can access it from, and to retroactively revoke access. That's that's the uh that's EDRM. And then we also have um a masking product AIDLP which is a data loss prevention product that allows data to be masked as it goes to LLMs. So it [01:00:53] allows for understanding of prompts and any documents that are going out and it does semantic preserved masking and tokenization so that the data that leaves the enterprise is actually uh desensitized. when the data comes back there it's unmasked and then given back to to uh the EI applications. So and on top of that we have um a visibility layer that tells you what's risk in the enterprise uh how you should meet security and compliance what's most sensitive and more importantly what's happening to data who's accessing it what kind of controls are being applied and how are they being followed I'm going to spend a few minutes talking [01:01:33] about glean and seclore um now glean has um provided us with a very open integration platform you know that allowed us to really be able to get access to information that they have as well. And I think we work with Glean in three different ways. The first is before Glean indexes data. In that scenario, you know, we would deploy the armor connectors onto data sources, we would understand the security and the sensitivity of the information around it and then insert labels into the files themselves. that allows the label data with metadata and understanding and context to move into Glean. And then Glean can of course then uh access it [01:02:15] safely, understand uh the security posture and be able to deploy uh role-based access control that they do extremely well and employees can use it safely and because of permissions that actually do work this time around. The other way we work with Glean is um after Glean has already indexed the data. So using you know using their their index. So what Glean does really well as we all know is to get deep enterprise understanding not just the what and the why and a lot of the data is in the glean index. What seclord does is we connect and integrate with the the the index itself without having to deploy any new connectors, without having to [01:02:54] deploy any new systems. And based on that, we can do what Armor does well, which is we can actually um remediate that information, understand the context, apply labels and protection to information and allow Glean to enforce the rules. And Glee can then take that information and respect the labels in the metadata and give you the right access. The third way that cycllor and glean work together is in um in at runtime when AI actually queries informations sends inferencing tasks to LLMs. Our AIDLP product sits in the AI path and it masks information that's sensitive using small language models keeping all that sensitive information [01:03:35] inside the enterprise. we can understand and mask that information semantically, send it out to the large language models and gets the inferencing information back, unmasks it and then based on the rights, based on user context, based on policy, glean can then show that information to the end user. Uh what this unlocks for enterprises together with glean cichlus allows you to uh embrace uh lean faster roll it out faster in the enterprise bring more data sources online faster you know reduce the manual work of understanding sensitivity and security uh protect information that you have and get you ready to embrace AI faster safer and [01:04:13] quicker. Uh in summary, I think Glean understands and seclt protects. So Glean really understands your documents, permissions, relationships. Secure is really able to read that risk, decide what's sensitive, apply controls on on onto onto information, and together and glean are able to allow you to get your your AI um uh accelerated in the enterprise. That's all I had today. Thank you so much for the time. [applause] >> [music] >> Thank you. [music] Semantic preservation through tokenization. You guys all get that? It's good stuff. Thank you, Michelle. We've heard about building, connecting, understanding, and protecting AIdriven [01:04:59] work. Let's close this showcase by looking ahead at what comes next. Please welcome Nicholas Scopesi, co-founder and CEO, and Nicholas Lopez, co-founder and CPO at Horizon. I like to refer them as do Nikos. Welcome. >> Thank you so much. [applause] >> Okay. Okay. Uh last one, so bear with us. I'm Niko, co-founder and CEO of Horizon, an enterprise intelligence platform. Let me start first with a question. You got pain, you go to the doctor. What would you prefer? like having no idea what's going on or a clear map of what they want to do. I think we all would prefer to know what to do, right? And the same goal for [01:05:40] enterprise transformation. You can't fix what you can't see and this happens to every company. We were doing transformation for the last 15 years and we know how hard it is to really understand where to implement technology. This problem is the same over the last 20 30 years maybe. What is going on on top of this is that the industry is moving faster than companies can really change. But all these executives, they already know they need to change to remain relevant in this new world we are living. We were in your shoes. We know how hard it is to actually know where to start. It's going to be operations, marketing, customer [01:06:16] success, where it's going to be. So the only way is talking with employees. Bringing in a consulting firm six to 18 months is really painful and really long, right? So we find a way to do this faster at scale and frictionless. We engage employees into the transformation. Our models talks with every employee in the company and they love it. 97% of all these people, they will talk once again with the models if that's needed. And they told us, "Your AI understood me better than my boss and better than the last consultant that came to do these same questions a few years back." You might be thinking, "All right, what is corres?" We are a team of [01:06:54] agents that talks with employees to understand your processes and rank exactly where to automate first. It's super simple. Three different levels of the value proposition. Discovery agent first we get all the context from the company. This can be documents processes or even for Glenn we train the models. Those agents goes talks with employees. We map the processes find the frictions inefficiencies and then we store all of this information into the graph of the company. That is the brain, right? And then this model will study the patterns. This is the second one. We'll study the patterns, find what is broken, what [01:07:29] matters the most and help you to prioritize and go through the future state reinvention. And then what we do is we leverage all the knowledge that was already available in the company through presentations, documents, processes, but most important trap in people's head is the human judgment. So we leverage that human judgment into the graph for you to take the best decisions in how and where to improve the company. How this works in action is super simple and I will explain from a customer point of view. So our biggest customer it's an e-commerce marketplace in Latin America. They went through 2,000 interviews in [01:08:08] the finance department, 15 countries, all Latin America region, and they found 50,000 hours after the agents talk with their employee, map the processes, find what is going wrong, and the model study the patterns. From those 50,000 hours, the model struck what matters the most. They prioritize and they get a pipeline of $2.5 million for improve the company. And then it's super simple. like they will get those right-sized projects and move to action in two ways. Big initiatives where the model will create the request for proposal, the one pure, the strategic plan, all you need to move to action. But then in the quick wins, [01:08:46] the model will create the agent blueprint, the agent specification for you to copy paste in whatever platform you are using like Glean agents. Now I will hand it to Nikico that he will make a demo for explaining how our customers are using our platform together with Glen. >> Hi everyone, I'm the other Nikico. I'm the CPO. So here I explain how it works uh working together with Glenn. So most teams start on the far right asking what agent should we build. We start one step before. We take from Glen the govern enterprise context. Then we launch horizon discovery. Basically with this we map how work actually happens and [01:09:28] then we fill the gaps between policy and reality turning this evidence into implementation ready context that we can basically after that teams can take into Glen and build and govern those agents but let me show you how it works in action. So it all starts with a plan here. So the before asking what agent should we build we need to ask ourself what does what's the capacity of my team what are the most impactful processes and where should transformation begin so we start with a plan and before we like the the first step is we need to launch a discovery and the first thing when creating the plan for the the interviews [01:10:13] is for example in this case Horizon is asking us to run this query England and paste a complete the result here. So basically here we're taking all of the context that is already in the company's documents and that Glenn is helping us structure when pasting that into Glen. Thankfully I already put it here and basically this is outputting everything that we know about how the department works. So we can take this we can copy it here and we can put it here in Horizon. Right? So what will happen here is that Horizon will build all of the context and take what it already knows in order for us to launch the plan. Now I'm going [01:10:52] to show you like a plan that already has been built. So you can see how it looks like here. So as I mentioned, we launch the discovery, the employees get an email, they see the interviews, then we generate the processes, then from there we find the insights which are the most impactful opportunities inside those processes. And then I'll show you what the executive team sees. Oh, live demo as always. [laughter] Uh here. So this is what the executive team sees. How many hours were mapped across how many processes and how many hours of those are recoverable. And here we can see the map. So here we can see like the classic uh opportunity metrics [01:11:35] where we see the most impactful processes meaning where should we start first. So if we open this up for example we see all of the process the list of processes that were mapped and where's the highest opportunity you shouldn't start on this one right where it has very low impact so we should start always from the most impactful ones this this could be done in one day so this is a process that used to take months before what it does it helps the business and the transformation teams align themsel on where they should prioritize and where they should start moving to action but now I'll go into detail on how that works and how the [01:12:09] process looks like. So, as I showed you, we learn to discovery and what happens with that is every employee gets an email. When they open that email, they see something like this. So, it says, "Hi, Sam West. Sam, I'm Nico Horizon Consultant. To start, how would you describe your role today and what usually fills your day-to-day?" After doing more than 40,000 interviews, we're confident that our models are better at interviewing people than probably your best analyst. This could be done through voice if they wanted or could be done through text. Depends on on how people want to do that. After we do all of that, what happens if we put all of that [01:12:45] structure knowledge into what we call the company brain? This is our ontology layer. Basically, we connect the processes, what people does, the tools that they use and we structure it in a way so we can find patterns in order to where to move to action. And what some teams are already doing this is they're connected through the MCP and you can connect it directly with Glenn and complement these two. But going back to the plan, so how does it look like? What happens after we do the discovery? We basically generate the processes. So the processes are the way for us to understand where are the biggest bottleness, [01:13:27] the biggest process, the partner payment process and here we have the most inefficient process. If we open it up here for example we have the BPMN that we already mapped from the conversations and most importantly we got the insights. So the insights is our unity of inefficiency. Here you can see what's the problem. You can see the evidence in here who said what and you can see the impact but at the same time we need to move to action now that we understand where are the problems. If we click here basically we'll generate an initiative. to move forward. I'll go into an initiative that was already created. So I'll show you here for example. This is [01:14:06] one that was already created and here we have everything that we need in order to move to action and even we have the one pager that basically will help the technology team understand what they need to build. But we can download this and send it to Glen agent builder so we can move to action really really fast. So in this case I'm in Glen agent builder. I put the RFP I already built and Horizon and and Horizon together with Glenn build the agent. So this is how we turn together with Glen enterprise context into action. Going back here. Thank you so much Nico. So based on this presentation, this is like the biggest customer we got. As Nico [01:14:47] mentioned, we made more than 40,000 interviews to all of the employees of these companies getting amazing results. If all the things we've been explaining resonates with you, just gave us one process, one week and you can have ready to deploy agents in Gle. Thank you so much. All right, let's go. Thank you. What an incredible showcase of what's possible when people, technology, and imagination come together. To everyone who participated today, our speakers, customers, builders, presenters, and of course, our behind-the-scenes teams, thank you for sharing your ideas, your experiences, and your vision for what comes next. [01:15:37] Um, as you know, Glean is an open platform. We brought everyone here. We showcased a ton of technologies. It's because we do believe in allowing our customers to use the best of the best and not restrict you as a closed organization. So hopefully you see that in today's showcase as we you know eight of our hundreds of partners get to showcase techn technology in front of all of you today. The the partner ecosystem is essential to bringing AI native experiences to life. Together, we're helping organizations connect their people, knowledge, and systems in ways that create real impact. And we're building a future that none of us could [01:16:18] create alone. And to everyone that came to Gleno, thank you for being here and for making this community so special. Your energy, your curiosity, and willingness to learn from one another what make this event possible. We still have a lot more to go. We have final breakouts starting at 11:00 am as well in our as well as our expert lounge. You want to go deeper into the tech, please do in our gateway pavilion. So, please let's make the most of our time together. Talk with your colleagues here. Swap stories. Thank you again for joining us here at Glean Go. Can't wait to see you next year. [applause] [music] >> [music] [01:17:08] >> hey baby hey hey [music] [music] [music] I [music] do [music] hey [music] I >> [music] >> Heat. Hey, Heat. >> [music], Glean:GO AI-Native Showcase, Glean, 01m32jrxezt9bxeyema2cx58ap.