Day 1 Afternoon Keynote | Glean:GO 2026  🔗 **Source:** [Day 1 Afternoon Keynote | Glean:GO 2026](https://www.youtube.com/watch?v=GuwoULakxdY) > [!note]+ Document Notes > --- > tags: [enterprise-ai, ai-agents, ai-governance, token-economics, knowledge-management, developer-productivity] > author: Glean > source: Glean (YouTube) > reference: https://www.youtube.com/watch?v=GuwoULakxdY > date created: 2026-09-21 00:00 > --- > > [!info] Generated analysis — not the author's words. Sections below are machine-written > > from the captured source. Only text inside `> [!quote]` blocks is verbatim from the author. > > Do not quote or attribute any other line to the named author. > # 🏷️ Title: Day 1 Afternoon Keynote | Glean:GO 2026 > ## 🧭 Table of contents > - [📝 Summary](#-summary) > - [📖 Main Content](#-main-content) > - [🗒️ Notes](#️-notes) > - [📚 References](#-references) > - [🗣️ Paraphrase](#️-paraphrase) > - [📰 Newsletter](#-newsletter) > - [✅ Action Items](#-action-items) > - [🧪 Hypotheses](#-hypotheses) > - [📊 Methodology](#-methodology) > - [❓ Questions](#-questions) > - [🔮 Future Research](#-future-research) > - [💡 Implications](#-implications) > - [🧠 Learning Styles Assessment](#-learning-styles-assessment) > - [📊 Accessibility Assessment](#-accessibility-assessment) > - [🌍 DEIB Assessment](#-deib-assessment) > - [✅ Logical Validity](#-logical-validity) > - [🛡️ DARVO Analysis](#️-darvo-analysis) > - [🎯 Conclusion](#-conclusion) > - [🔄 Refactor Appendix](#-refactor-appendix--day-1-afternoon-keynote--gleango-2026) > ### 🔗 References > - [Day 1 Afternoon Keynote | Glean:GO 2026](https://www.youtube.com/watch?v=GuwoULakxdY) — the source recording, 57:36, uploaded 2026-09-21. > - [Glean](https://www.glean.com/) — vendor hosting the event. > - Patrick Wendell's post on how Databricks optimized coding-agent costs, referenced on X at [00:26:41]. No URL given on stage; search X for the Databricks coding-cost note. > - OpenRouter token-consumption study cited at [00:39:21]. No URL given on stage. > ## 📝 Summary > - **What happened.** Three back-to-back conversations closed day one of Glean:GO 2026 in San Francisco: a customer fireside with Ericsson's chief data officer and eBay's CIO, a founder conversation between Glean co-founder Vish and Databricks co-founder Patrick, and a four-person panel moderated by Sharon Goldman with Cisco, Deloitte, and Glean's Arvind. The session closed with day-two logistics and the inaugural Glean:GO AI Innovator Awards. > - **Why it matters for a Glean customer.** This session ships zero product announcements. Everything Glean-specific is deployment evidence from three named accounts — eBay (three years in, search first, now "no data source internally... that's not connected"), Ericsson (POC fall 2024, rollout early 2025, 50–70% of employees active), and Databricks (10,000 employees, multiple uses per day). If you're cataloguing what Glean *shipped*, this is not the session; if you're building an internal case for expansion, the adoption and governance numbers here are the reusable part. > - **Key stakeholders.** Joel (chief data officer, Ericsson), John Felmmeer (CIO, eBay), Patrick (co-founder, Databricks), Vish (co-founder, Glean), Sharon Goldman (moderator), Arvind (Glean), Ashish (Deloitte), and the Cisco speaker rendered by the captions as "G2." > - **Risks and tradeoffs.** Every cost figure on stage is qualitative — no dollar amounts, no per-seat comparisons, no latency or benchmark numbers. The most quantified claims (60% of compute on inference, 14x agent token growth, 5x agent-to-human token ratio, FY27 liftoff, Cisco's 24-month no-legacy-code prediction) come from one speaker and are sourced to either an uncited OpenRouter study or nothing at all. > - **Bottom line.** Treat this as field evidence and market framing, not a roadmap. The durable takeaways are operational: eBay's per-user AI spend dashboard, Ericsson's shift from pre-screening checklists to runtime control environments, Databricks' pattern of using a second agent to review the first agent's code, and Deloitte's per-persona "workbench." > ## 📖 Main content > ### The session is customer evidence, not product news > No Glean feature was announced, dated, or given a GA/beta/preview status in these 58 minutes. Every Glean capability discussed — enterprise search, chat, connectors, the knowledge graph, agent building, and an agent publication and visibility workflow — is described in the past tense by customers running it in production. Availability status for all of them: **unstated in this session**, though production usage at eBay, Ericsson, Databricks, and Deloitte implies GA. > The one forward-looking Glean item is a cross-reference, not an announcement: eBay's CIO says at [00:17:08] that he heard "earlier in the keynotes about agents with their own identity," and Arvind says at [00:52:24] that proactive AI got time "today." Both point back to the morning keynote. Status of agent identity and proactive AI from *this* session: **unstated**. > ### Three deployment stories, three different starting points > eBay and Ericsson both entered through enterprise search, and both describe the same second act: search becomes the default entry point, then the connector surface expands, then the knowledge graph gets good enough that removing it becomes politically impossible. eBay started with roughly half a dozen connectors and now claims near-total internal coverage. Ericsson's differentiator in the buying decision was agents — Joel says Glean was "early on agents" before launch, and enabling employees to build their own was part of why Ericsson bought. > Databricks arrived by accident. Patrick, who owns internal security and application deployment, first met Glean as a security review he nearly blocked, then got converted by an employee who had already deployed it without asking. > ### Cost control is the shared operational problem > Three distinct cost mechanisms surface: > 1. **Visibility first, controls second** (eBay). A dashboard shipped in the last couple of months shows every user their own AI spend across tools. The stated purpose is behavior change, not blocking. > 2. **Power-curve triage** (Ericsson). Token consumption follows a power curve, so a small number of individuals account for most spend, and coaching those individuals is enough control for now. > 3. **Population-scale unbounded risk** (Databricks). Consumption-based coding tools multiplied by more than 2,000 engineers is the exposure Patrick spends his effort on. > None of the three gives a dollar figure, a per-seat comparison, or a budget. Contested claim worth flagging: eBay's argument at [00:13:56] that usage-based pricing makes tool sprawl cost-neutral assumes constant per-task token cost across tools, which is not true if two tools route to different models at different prices. > ### Where quality bottlenecks move > Patrick's argument is the most carefully reasoned segment. Coding models are good enough that writing an individual module is no longer the bottleneck, so the bottleneck relocates — to code review, and past that to architecture and taste. His explanation for why models still fail at architecture is a claim about latent context: an experienced engineer brings intuition from a decade of prior jobs that is not representable in the requirements document, and tools that add context help but are bounded by what can be written down. He names the feedback-loop asymmetry as the mechanism: small decisions get fast feedback, architectural decisions get feedback over years. > The Cisco speaker makes the same bottleneck-relocation argument at [00:49:11] independently, which is the strongest convergence in the session. > ### The market-forces framing > The Cisco speaker lays out four forces: inference is the new workload, agents are the new workforce, tokenomics, and control. The supporting numbers are the densest in the session and also the least sourced — see Logical Validity below. > Terms worth defining for a reader who wasn't there: > - **AI mesh** — Ericsson's internal name for the connective and governance layer between AI systems. Not a Glean product. > - **Workbench** — Deloitte's per-persona bundle of tools and agents. Not a Glean product. > - **AI front doors** — one of four themes named by the moderator at [00:05:35] as running through the whole day; the single-entry-point problem eBay describes at [00:14:37]. > - **Tokenomics** — th [00:00:02] [music] [music] Welcome back everyone. I hope you've had a great day exploring the breakout sessions, demos, and everything happening across Glean. We've got a great day one closing keynote ahead of us with a few different conversations that bring together the customer technology and industry perspectives we've been talking about all day. First, I'll have the pleasure of sitting down with two of our wonderful customers, Joel from Ericson and John from eBay. Then our co-founder, Vish, will sit down with Patrick, co-founder of Datab Bricks. And we'll close out the day with a panel featuring G2 from Cisco, Ashish from Deote and Arind. [00:00:54] We spent today talking about how enterprise AI is evolving. These next sessions are now a chance to hear directly from the people building it, deploying it, and shaping where it goes next. So, please welcome on stage Joel and John from Ericson and eBay. Three [music] high five. Beautiful. [music] >> All right, let's get into it. So, thank you so much for coming today. Um, should we start with some some introductions? So, Joel, do you want to go first? Do you want to just tell me a bit about your role and some of the top priorities you have at Ericson? >> Sure. So um so I'm chief data officer and I am also responsible for a program [00:01:44] to scale I guess what we at this conference call enterprise AI the broad use and our journey from microp productivity and starting to use AI to workflows and then moving towards enterprise level outcomes. So and in Europe exciting also all of the AI governance and the things that come along with it right which is can be quite hefty. So, >> and John, >> uh, yeah, I'm John Felmmeer, uh, CIO at eBay and, uh, been been at eBay for a long time in a lot of different platform and infrastructure areas, but in the current role, really responsible for all of the AI tooling that's used internally, uh, and, you know, bringing [00:02:29] those pieces together, uh, for our employees. And and while you've got the mic, John, can you tell me a bit about your journey with Glean so far and where did eBay start and what's top of mind right now? >> Yeah, so our journey with Glean started I was just thinking about it just about three years ago now uh when we were looking for a way to we had been trying to really uplevel the knowledge management internally. You know, it's eBay's 30-year-old company, and we, of course, like many people here, had information spread all over the place. And we were, we had tried a couple of projects internally over the years, uh, [00:03:07] to to to try and bring that information together. Uh, finally, you know, found Glean and we, you know, three years ago, it was a search experience. That's really where we started. You know, how do we have an integrated search? I crossed all of our knowledge. Uh, chat was just starting out. Uh but here we are you know again two and a half three years later and it's completely transformed uh and Glean has really been the the you know a major sort of foundational piece for us in upleveling our knowledge management. >> It's crazy to think eBay is 30 years old. >> You must have a lot of knowledge in that um in that company. Um and Joel, what [00:03:46] about you? What's the journey for Gleam been like for Erikson? >> Yeah, so for us I think we started summer two years ago. uh and we did a PC during the fall o of 24 and and we kind of got onto it and in the beginning of 25 and for us it was similar we started also from the finding information use case enterprise search like the internet search bar was terrible and that was the starting point uh I think decisive uh in the process was also that clean was early on agents right before it was launched uh we saw it was coming you had a good idea on how to do it and we wanted to enable people across the organ organization to build themselves. So [00:04:24] that was really part of the of the decision and then we've been adding it out, rolling it out. Now it is the new internet search bar as well. So that's at least very transformed and and a lot of of other things coming along with it. So I think we are now somewhere between like 50 to 70% of our um employees actively using. So that's very good. >> And when we talked earlier you were saying that it's not so much a business priority as a data priority. Can you just elaborate a bit on that? No, but I think it's I mean it's really been and I think others have said this here today earlier as well. You put the spotlight [00:04:59] on data challenges that you had and I mean you're 30 years young. Great. We're 150 years this year. So we do have a lot of legacy and it really puts a spotlight on those things that you need to be improving. Um, and I think that's been good for the company, right? To say, okay, our internal policies are not tagged to which legal entity they belong to. So, it's hard to find or oversharing. Yes, we've had oversharing or like we need to know where we have personal data. I mean, we used to think we knew, but maybe we didn't know. So, you can use some of these tools to to build that in. So that's kind of a I think it's a foundation because it's it [00:05:35] will help build trust for also more sophisticated AI that you can kind of handle these basics and in particular around unstructured data. >> Um we've talked a lot about um well a few themes today. So enterprise context, token efficiency, the role of governance and AI front doors. Um so maybe John I'd like to start with a question about context. So we all believe that models provide intelligence but enterprises have to provide that context. How do you ensure that AI understands your organization's context and what are the priorities and constraints at eBay so you can move more generic answers to meaningful enterprise impact. Yeah, I [00:06:17] mean that's really I think you know as I said at the start you know like why were we looking for a product you know to bring it all together um you know with you know only 30 years of knowledge I know you know we're not 50 or 100 but it it was you know we needed something to bring all of that you know information that was out there together uh and glean you know really like one of the things that really intrigued us from the start was all the connectors you know because that was one of the hardest things trying to build it yourself is how do you bring all these different systems together you know so when we first [00:06:51] rolled out Glean I think we started with half a dozen connectors but that quickly expanded uh and now you know there really isn't a data source internally I think that's not connected through glean and over the you know the two and a half years we've just seen that knowledge build up more and more the more glean is used the more you know it answers questions really Well, so so that you know that that enterprise context is what Glean is bringing for us right now. I mean more than anything I mean you know you know we we definitely want to go beyond just knowledge management and we've started to in a number of areas [00:07:30] but I will say you know definitively today if I tried to take glean away from from my employee base I you know I would have a revolution on my hands because it has made it so easy to answer questions internally that used to just be these crazy discovery exercises where people would bounce around between tools or asking their peers are asking somebody else. So it glean really is bringing that enterprise context because of the connectors and because of the way the knowledge graph is built up over the two and a half years. >> Thanks. Um it would be remiss of me not to talk about token and token efficiencies. Um Joel, how do you [00:08:10] evaluate the trade-off between the cost of tokens and the quality, speed, and intelligence of AI experiences? >> It's a great question. Uh we were also discussing here backstage actually tokens. um out of habit maybe uh I think it's not so much a trade-off maybe of the experience and the token I think right now we're at one visibility on token spend uh two ability to control it I think that's kind of the baseline uh and then the second conversation around token is one which is also sovereignity as much as u vendor lock in as much as cost right and for us it is important to have flexibility in the stack uh and as [00:08:50] a safeguard for to the future let's say right and I think that's probably how we see it uh but so far the token consumption is very much following a power curve as well so you see just a few individuals and maybe they need some guidance or some coaching uh and that brings I think a lot of control enough control uh right now at least uh looking forward we need more but that's sort of how we think of it >> yeah the challenge is how to size the workloads I think that's one of the the conundrums that we have at the moment on how is CB eBay approaching that? >> Yeah, for sure. I mean, you know, obviously token usage is growing [00:09:27] everywhere, not just in development. And you know, I think, you know, initially that was, you know, the focus is let's make sure, you know, the engineering usage is is appropriate, but now we're seeing token usage growing very broadly across the organization. And in an earlier session today, I talked about one of the things that we've implemented in the last couple months is a dashboard uh that shows users, all users inside the company their basically what they're spending on uh AI uh across the different tools. Uh because one of the things we learned, you know, is people don't really understand the cost of AI. [00:10:05] They just assume it's like any other enterprise tool, kind of all you can eat. you know, I've got access to my Google Docs or my Microsoft Word and I can just use it and they don't really understand that, you know, there it is a usagebased model primarily today. So rolling out visibility was a huge step and as part of that then we're also starting to put some controls in place not to prevent people from using AI because we want them to use it that's the value but to make sure that they understand and use it responsibly. So there's kind of steps along the way and and that's really having we've already seen an immediate impact where people [00:10:42] are like, "Wow, I didn't know I was spending that much or I guess I could have done that a different way that would have been a lot more effective." So I think it's it's really important to provide visibility. >> Yeah, just on the controls and governance, I mean that's super important especially as AI is moving more from just answering questions to being more proactive and taking actions. Joel, so how has Erikson's approach to governance and trust had to change because of that and this move to the agentic era? >> So I think we're probably in our second or third generation of revisiting governance uh since the introduction of [00:11:19] large language models. Uh so I think the first thing that happened and this is now a while ago I mean we of course as many large companies had a long let's say procurement compliance etc process. The first generation was us looking into that, leaning out like taking away duplicative steps and looking at this making a bit quicker to buy things. Basically, now what what we're doing right now is a more consequential shift. So, we're looking at with AI and what we expect AI to be able to do. We're not going to be able to rely on an upfront kind of pre-screening checklist, you know, paper or PowerPoint based process, [00:11:57] which is how much of governance has been done, whether that was a data transfer or a privacy review or uh an IT security review. We are going to have to rely on control environments, policies, and then active monitoring and managing the deviations. Uh so I think we're in the middle of uh that shift I would say and we're building something we call an AI mesh to connect things and to govern um but we already have uh these conversations up to the I mean the most senior levels to oversee governance in Ericson uh and we are using AI in governance. I'm very excited we have um for a responsible AI process you uh we built an agent you actually like if you [00:12:35] have an IT architecture you can just upload it and it will like populate the questionnaires and give you kind of a response on where you are. So I I think that's what we're doing. But really you have to manage at runtime the autonomy uh is where you have to go to. >> Yeah. I mean one of the things about governance um and that we talked about also today is AI sprawl and how important then governance is to manage that. So John as AI becomes more widely embedded across the business how is eBay addressing the challenge of AI spraw? Yeah, this is I think the the the million-dollar question for many of us is, you know, it a AI is sprawling. It's [00:13:15] everywhere. It's embedded with every product we're using. And you know, there was if you had asked me this question a year ago or two years ago, I had a very different answer than today. Um or I should say part of my answer is different. So on one side AI sprawl matters because the you know when we when we were licensing I said you know kind of mentioning enterprise licensing earlier when you're paying per seat for a product you don't want to have you know 10 products where you're paying you know for every employee to have access to that product that gets really expensive a as the AI you know world has evolved it's become very usage based as [00:13:56] we all know so on one And my view today is well sprawl from a usage standpoint isn't that big a deal because if if I shut down one tool, let's just say I'm using Claude and you I shut that down and all my users move over to Glean, my usage hasn't changed. I still have the same amount of usage or if they go from chat GPT to claude, they're still using the same amount, you know. So, so when we move to a pure usage model, which we're largely at right now, uh, then I don't worry as much about sprawl there. But, but where I really worry about sprawl and where we kind of started even on the beginning of this journey is the [00:14:37] customer experience, our internal customer experience is terrible with the more sprawl we have. It's very confusing to our user base you know to when you know when they have chat GPT as an option they have Glean as an option they have claude as an option they have Gemini as an option or any number of other tools and they don't know where to go you know so so you know one of our premises when we started out with clean is we want a single place for people to go to get an answer uh you know I'd like to say you know I've been successful there but I think that's still a big challenge for us is is trying to kind of [00:15:11] rein that in and have that user experience be very clear, very understandable and even when they want to you know build something like oh we want to build an agent to do something which tool should I build that in uh so so I think sprawl from a customer experience standpoint is still a big problem that we have to solve on the other hand I'm not as worried about sprawl when we talk about usage because I you know right now like I said I'm just pushing the ball around uh from a usage standpoint if I if I you stop one tool and move to another tool. So yeah, it's I think this it's sprawl is is definitely a problem, but the [00:15:50] problem is changing >> and and maybe to build on that a bit, I think one type of spraw is of course we have a grassroots movements to like put something together and I think in that respect Glean has been very helpful for us to sort of manage that part and and we have a publication process uh that I think that we've worked out uh together with you on how to sort of promote agents and what should be visible to whom etc. So I think that's really good for that sprawl. Um in the more broader sense uh our concern now is really a lot around the decision loops of the agents right and where will this sort of history acrewue right the traces tool [00:16:25] calls etc and who will be able to learn across your workflows and and how do we stay fully in control of that and that's a priority as well and but if you can stay in control of that doesn't really matter as much which consumptionbased tool that you that you have um yeah but it is complicated for sure. >> Yes. >> All right. Well, I'm going to wrap this session up with a future looking question. So, um it's not on the script, but I am a fairy godmother in my part-time. And I'm going to grant you two wishes. Like in the future, what AI outcome in two years do you want to achieve in your organizations? And then I'm going [00:17:08] to get you back in two years and we'll see if it's happened. So John, what's the future like for you? >> Yeah, I think you know for us I think with AI right now what what we really want to get to is beyond just this sort of you know knowledge work that we're doing today and truly having AI be part of a collaborative partner in the environment. So I think we heard earlier in the keynotes about agents with their own identity. I that's very much where we're going which is we want to have agents working side by side with the employees uh and collaborating you know just like two employees would work together today now it's going to be [00:17:49] employee to agent agent to agent agent to employee and and so really having like I said agents as a regular part of our workforce uh you know and so you know the way I often describe it to people is today we get work done with employees We also use contractors or outside consultants. And you know where we're going is we're going to have employees, we're going to have contractors, outside consultants, and we're going to have agents. And it's that combination that that really is how we're going to be working in the future and how we're going to get work done. And finding the right balance, you know, I think is going to depend on the [00:18:26] department, depend on the area. Uh, you know, and in some places we might have a lot more agents and in other places we might have a lot more employees. It really depends on the type of work we're trying to do. But that is the vision I see in the next two years and you know how we orchestrate between those is what we have to build. >> Yes. I think maybe three layers of that answer. I think the sort of the more >> you've got two wishes. >> Two wishes. No, but I have I have a two wishes but I I think the first one which is more the mundane a little bit boring like how would we want it to be at Ericson? Of course, we want to move to [00:19:02] outcomes that sit at the at the organizational level, like the domain level. So, you could say instead of we have some agents in a workflow, you would say we now respond half of the time one week faster in this process or or similar, right? We're much more um cost-ffective, right? I think that's a boring. But I think what is uh exciting, right? Uh is and what I'm excited about is changing. I will probably be a knowledge worker also in two years. like hopefully I am not replaced fully by AI but I would like to see work change fundamentally. So, so one example uh personal use case is um maybe my team hates me now but projects we've done [00:19:38] many many projects right and as kind of you can imagine um so one thing I did was I asked Glenn right okay this project what's the status and all people were on vacation last week um and gave me a status but I asked okay but is this a really executable plan that we have here like what you can see what are maybe the gaps right and he came back to me and said no you it's not an executable plan right we have some clarity on some deadlines multiple documents with different deadlines, not one single source of truth, etc., etc. Um, but if you take that one step forward, you could imagine as you build that, it helps you every day uh to see [00:20:12] and also identify not just is the status red, green or yellow, but like is there a decision that's stalled like maybe it's because of me by the way, right? Or I have been ambiguous and so on. So I think if you if you apply some of those things I think we'll up for a revolution that is kind of like the introduction of agile or something else which will be the tool but also how you come together and also where you store your stuff and how you like work. So, uh, like I'm excited to see that. And then, um, in the end, of course, I want these tools to be like this fantastic first Whimo experience, right? You get in and it's like, you know, now it talks [00:20:44] to you. Whimo yesterday talked to me in Swedish, like it drives you somewhere like you kind of want a little bit at at work as well, right? Not just where you feel overwhelmed and you have to like bot sit or otherwise, right? So, that's probably my wish. >> Well, I will grant you your wishes. All right. Thank you so much for joining us today. We really appreciate it. Um, please give a hand to Joel and John. [applause] All right, we've got another panel coming up. So, I'm very um excited to welcome up Vish and Patrick from Data Bricks. >> [music] >> Heat up here. Patrick, it's great to have you here. >> Thanks for having me. [00:22:11] >> Uh, so I know that so data bricks was one of the early customers of Glean and you were behind the purchasing decision. So can you share some stories around that? >> Yeah, it's kind of funny. Um, so I I'm in charge of a lot of our internal security and application deployment at data bricks and uh initially I didn't know what glean was but I got this uh security question from you know I have to approve when like there's a security thing and uh and the person said hey Patrick we need you to approve something. So I went to look okay what am I approving? Okay I've never heard glean I've never heard of this thing and [00:22:47] um this was many years ago and uh and I said well what do they do? What's the security question? And they're like, "Oh, they go in and index like every single thing of our data. All of our data at data bricks." And I was like, "What do you mean?" They're like, "You know, all those important applications, they just go and like make it index." And I said, "They do what? Like, are you kidding me?" Like, "No way. Like, why are we even having this meeting?" And um and then I kind of started pulling the thread more. Okay, wait, can you just say what is this thing? and uh and someone we had an employee that was like [00:23:16] really a champion and loved Gleam and they uh they pulled they pulled me aside and they had their laptop and they said look I've got it like already running up and running and they have like all this useful data and first of all they had already done it without asking me which was not great but um but this person was very ambitious and I was like okay show me give me give me this thing and they're like just tell me any question you have about data bricks anything you want to know and I forget what I was asking like hey this project why is it running behind it was something that was just thinking about and I typed it in [00:23:44] and like like it had perfect answer for that. Um and then I said okay maybe we should consider this thing actually you know this might be pretty useful. Uh so then we did go through all this security and you know we we looked at it very seriously and everything and uh and I still use glean like every day now that was many years ago. Um but that was that was kind of a that one ambitious employee who still is at data bricks. I'm not giving you their name because I know you'll try to get them in sales. [laughter] um uh that person uh really like showed me the light and you know our employees we have 10,000 employees I think on [00:24:16] average people use glean multiple times a day now to really run like almost every aspect of of their internal work at data bricks >> that's awesome I was on the other side of that and so you were one of our early customers and really appreciated you know how you made the product better like all those security questions were actually very good >> I'm sure for you it was like hey there's this founder this guy seems really pissed off about the security like we don't know if we can convince them. >> Yeah. So u yeah to jump in you know um uh one of the things is how uh we are curious to know how you are using AI in [00:24:52] your engineering organization because that's top of mind for uh all companies these days and so and data bricks is one of the leaders in the space so very interested in what are some of the best practices how what how are you using AI to transform engineering >> yeah so so Glee was one of our very AI tools and it's still pervasive um for kind of every knowledge worker in the company. But certainly the biggest change in the last year has been uh coding tools for engineers and uh a bunch of things have happened like the the quality of the coding models has gotten really good. There's been a proliferation of these like harnesses, [00:25:30] these kind of outer loop that's helping you uh write code and and almost every line of code today at data bricks is authored initially by some AI could be human reviewed and edited, but the first stab is usually AI. >> Um and so I've also been I don't know how I got this job, but I've also been in charge of the roll out of those tools. And um it's been really exciting and and productivity improving, but also there's big big challenges and um one of the big challenges is cost. you know the the tools are all consumption based. So in principle a single person who's you know aggressive about their use of a tool could acrue almost unbounded cost. [00:26:06] >> Absolutely. >> Um and when you maybe if you have a few trusted people with such a tool it's one thing but we have more than 2,000 engineers at data bricks. So uh it's a huge population multiplied by kind of unbounded cost that a lot of our effort has gone into how to think about managing that cost. M um and then the other thing is like the the the bottlenecks. We found different bottlenecks in our sort of software supply chain. You know the the writing of an individual piece of code is not a bottleneck anymore. It's so fast. Yep. >> So you start to bottleneck in other areas basically. >> Yeah. And and um for engineering leaders [00:26:41] in the audience, you know, Patrick has shared a great uh note on how data bricks optimize coding costs uh on X. So check it out if if you uh in your time. So I guess one of the things is like code generation becomes dramatically easier but one concern you know I have other uh engineering leaders have is you know are we compromising on quality. So how do you see the role of um architecture testing like you know production uh judgment all of that evolving in in this new era? Yeah, it's a great question because I mean if you look even ignoring code for a minute I one um unfortunate aspect of like the proliferation of AI generated content is [00:27:28] that in my view the quantity has gone up the quality has gone down um and uh and certainly like when it's you know on Twitter it's one thing but this is like our codebase this is our IP so we don't necessarily love this idea of quality going down you know it's what we're selling to our customers so we don't have all the answers, but some things we've found helpful are being honest about what coding models and tools are really great at and using them for that, but also being honest about where they're not really great. So, for instance, for big architectural decisions, >> I found that the the models are just still not very good. And and I thought a [00:28:09] lot about why this is. So, you know, because in in principle, I should be able to write down a design doc like I would otherwise and say the constraints and what I'm solving for. And I should almost start by writing all the problems and then I should say finish this design doc and it should write perfectly the architecture. You know a really um a seasoned engineer can do that. So why cannot the model do that? And I think it's because in practice when when a very um experienced person is tackling a challenge they're bringing a lot more to the process than just what's reflected in maybe the requirements as written [00:28:42] down. You know, they also are bringing huge amount of context about the company. They're bringing intuition from their prior jobs. And it just is for me actually insight into how human decision-m is working in that we ourselves are not just like okay there's some textual inputs and there's some output. We're actually taking into account many latent considerations that are unstated but are really really important. And unfortunately I I don't even know if in principle the models can ever do that. If they if they don't have enough information coming in then I don't even know if in principle they can produce that level of of insight and [00:29:17] decision-m you know things like glean that give them more context are helpful but ultimately they're constrained by like what can be represented textually whereas you know I'm my context might come from a decade of experience and it's hard to even put into a tangible written format. that's somewhere stored in my brain as what we might call intuition or something. So, so I think the biggest thing for us on quality is like recognizing tasks that are still in that category where humans re supreme, you know, and maybe they won't forever, but they do now >> and and really making sure humans do that and the right humans and then [00:29:51] separating that from let's say, you know, wrote coding of I need to write a small module that's well specified and the AIS will do that better almost every time now. So I think it's sort of this task delineation at the task level that is one key to it. >> Yeah, absolutely. I think I think there's something to be said for you know the the length of the feedback loop right like for a small thing it's is easy to get the feedback and >> exactly like like some of the big picture decisions we get feedback over years on whether we we ourselves don't even know for years whether it was the right decision. So how can you have a [00:30:24] tight AI learn that? It's very very hard. >> Yeah. Exactly. So, so um you know the other thing that we we think a lot about is what what are these metrics around quality that um that that you know that go beyond things like token usage or number of agent runs that happen like how do you what are the metrics that you watch out for uh when you look at the quality of the work produced and the velocity? Yeah, that's a very difficult question. And again, I don't know that we have all the answers, but one thing I find really funny is that somehow tokens became this thing you're supposed to like maximize, which is really funny because it's a [00:31:05] cost. >> That's generally not how companies view like you don't generally like give the person a reward who spent the most money in the company, but there was this like thing of for tokens that is how you reward people. >> I really wish it's like a great marketing that's a great marketing move by the model providers. I I wish that there I wish that happened for data bricks. Like the coolest thing you could do is like spend as much money as possible on data bricks. Can we all make that can we go make that cool? That would be like really awesome. >> Um but uh so I think there's some I can think of some bad examples of what to [00:31:36] optimize. At the end of the day really fundamentally we're trying to measure ROI of some sort like we're spending I think tokens are on the cost side of the equation and there's some benefits that come out and in truth I think we've only just started to approach some ways to think about measuring that. So for instance for engineering velocity we look at how much code is coming out and we actually have independent we have agents reviewing the quality of the code like often a different agent than wrote the code. M >> it turns [clears throat] out that agents if you ask them to evaluate their own output they think it's really good [00:32:06] >> exactly >> you know it's like asking any one of us like hey what do what do you think of that paper you just wrote like ah it was really good that paper I just wrote but you ask someone else you get like a different answer um so so there's some ways you can look at both quantity and quality in limited context like for when when it's code that's being produced but I don't think we have the answer yet for like hey I've rolled out some AI tools in my company is it working or is it not working like what is my ROI? What is my I I think it's still very early in that process right now. >> Yeah, we are we all trying to figure [00:32:38] this out as we go through it. Um but thanks a ton Patrick for joining us and sharing sharing your experiences uh in the AI transformation. Uh next up we will have um you know Sharon Goldman who will lead a conversation with Arvind Ju from Cisco and Ashish from Deote. Thank you. >> Thank you very much. Heat. Heat. N. [music] [music] >> [music] >> Hey everyone. So great to be here with you all. Thank you so much. We've been hearing all day today about interoperability, scale, context, governance. I'd love to kind of get underneath all that and kind of connect the dots. The promise that we're all talking about, of course, is this [00:34:15] true AI transformation that goes not just from a co-pilot, but to true agentic AI. But the reality when I speak to enterprise executives, I have to say that I still hear a lot of FOMO, a lot of pressure, but still struggling to kind of, you know, get there to those scaled outcomes that we've been talking about all day. So I'd love to talk about what it actually does take to transform. Arvin, I want to start on the far end here. Why is this still so hard? What are the bottlenecks? Is it really is it the model? Is it the data? Is it the context? Is it all those things? >> Yeah, it's it's all those things. I guess we've been talking all along since [00:34:55] the morning that to make agents truly work in the enterprise and add business value like first you have to make them have access to as much information as much experience you know as your current like you know owners your human experts have. So that has been like the biggest bottleneck I would say in terms of like you know why businesses are still struggling to drive that business value with agents. Um and then like a few other things you know also come into play. One is um businesses do need more help like in terms of take any department um or any vertical industry. What are all the things I should be doing with AI? You [00:35:38] know this technology is so open-ended. It's so powerful. You know in some ways you can do everything that you want with it. But that also makes it hard to then figure out how to prioritize and and how to actually not always be at the frontier and be the one who's experimenting. Uh you also got to learn from your peers and what things have worked um at other companies. what are the success stories and so that's all like so there's there's that level of help needed um which I guess like you know we for example like we're focused on that you know when we work with customers we are sharing those mutual learnings um what are the big success [00:36:13] stories you know that we've seen elsewhere that could actually also work for you so I I would say those two things are actually you know um is what does come in the way of people like you know succeeding the way they should be >> ashish I want to ask you can we drill down a little bit into what we mean By scaling in this context, you know, we're talking about going beyond chat bots, you know, saying something, getting something back. Are we talking about delegating? Is it really about delegating the actual work? And then what does that change for enterprises as a whole? >> It's really reimagining, you know, ways [00:36:50] of working, right? Like so if you look at sort of what we're doing with Glean and and how we manage and deploy our resources, right, which is really how do you recognize somebody's skills, their resumes and their availability to get them to the right job or right role to the right client, right? And there's 180,000 people in the US firm in Deote and 450,000 people globally. You can't imagine a human being attempting to read 180,000 resumes one at a time to position the right person in the right role. Right? So when you talk about scaling and we talk about you know what their imagination look like it doesn't look like what the work a talent [00:37:25] deployment manager did versus what AI does for them. There's always human in the loop that's sort of making the right kind of judgment calls and sort of the workflow which we had to reimagine to deploy this. So when we talk about at scale we're really talking about how the ways of working and the ways of deploying that at scale have changed at least for us. Um and it's true for most of our clients right when we it's very hard to take a existing process and say I'm going to identify this part but leave this intact it doesn't work right because there's a sequence of human beings in that process today where work [00:37:58] sort of moves from one human being one department to the other you cannot randomly identify it and pretend that it works it does not >> it's hard for me to wrap my mind around though and I'm going to turn to G2 here if you're talking about delegating to dozens, hundreds of agents, you know, not how do you govern a swarm? I guess is just the basic question. It it just it's hard to wrap your mind around that. >> So, um, first firstly, this is an epic venue. Is is like everyone having a good time? Like, >> yeah, [cheering] it's [applause] great. Um so I think like if you if you were to take a step back and say what is [00:38:40] happening in the market right now as we move from this era of chat bots to agents um I think there's like four driving market forces that are really interesting that are shaping up in a in a way that are that we've never seen before. The first one is inference is the new workload. We're going to have we have now gotten to the point where the global compute capacity 60% of that global compute capacity is actually now consumed by inferencing not by training which means that you've gone past the kind of let's hope this thing works. We have actually now started seeing the token consumption being you know kind of completely moved to um the [00:39:21] inferencing side. That's number one. Number two, if you think about agents as the new workforce, um in February of last year, Open Router had a really good study where there was this kind of fundamental uh shift that happened where the consumption of tokens by agents exceeded the consumption of tokens by humans in February. Since then, it went up 14x. And so now the consumption of tokens by agents is five times that of the consumption of tokens by humans. The slope of the curve of agent consump agent token consumption is like almost a vertical line. The third area is uh this notion of tokconomics where everyone's [00:40:04] worried about cost of tokens um and that has to go down and as as that goes down like you're going to have a whole sort of kind of a whole class of factors that will kind of play into effect. And then the last one is that control really matters. You know these models need to be within your enterprise governed by you because your intellectual property matters, data sovereignty matters. Um people are starting to get more and more open to the open weights, open source models. And this is the first time I think in a major platform shift that's happened where you no longer have to be at 70 80% adoption by humans in order to create a [00:40:49] super cycle. Right now, sub 5% of the humans are power users of agents. And the amount of agents they've created have fully consumed all the capacity that we can create from an infrastructure standpoint. And so just take take a moment and reflect on that that you you might never have more than 5 10% of human power user adoption and you would still be able to completely satiate demand, you know, because the supply just would not be available. And that's that's kind of bonkers when you start thinking about it because imagine what happens when instead of 5% of humans 10% 15% 20% 30% of the humans start to take start to [00:41:29] become power users when you get to 80% of humanity becoming power users we might literally not have enough infrastructure to go out and fulfill the needs of AI. So that's one. And the second area that's really important is this notion of the ability to securely delegate work to an agent versus the ability to just delegate work to an agent. Is the difference between market leadership and complete abject bankruptcy, right? Because it's a business ending event if an agent just starts to go rogue. And so monitoring an agent, dynamically, applying guard rails to the agent, ensuring that you can make sure that and and the this notion of security [00:42:09] and observability will actually fuse together because it's going to be hard to tell if an agent was someone that had been influenced because of a prompt injection attack or the agent just felt it was a good idea to jailbreak and go out and start doing something on Hugging Face because they wanted to go out and steal the results of an eval, right? And so the the the ability the these markets will start to merge between security and observability. So net net I feel like there's a tremendous amount of opportunity for all of us to participate in this market. Enterprises will um massively have a liftoff moment in FY27 [00:42:51] and inferencing will probably be about 70 80% of the workload over time, not training, but scaling loss will still persist and you will still see that the absolute dollars spent on training will continue to keep going up. And so it's kind of an interesting dynamic that you're seeing happen. >> It is bonkers. I completely agree. >> There you go. [laughter] Everything's bonkers about life right now. None of it makes sense. >> Arvin, but isn't there a tension then? I mean, you know, the governance of all of these agents, you're also trying to reduce costs. Can you talk about the management of all that? How how does [00:43:27] context that you've been talking about all day, you know, come into that? >> Yeah. Um the development of agents like in most large enterprises you know is very democratic in nature. Um like different companies choose different architectures or um or philosophies. You know in some companies every person inside the organization is allowed to build agents. In others it is maybe a select few like who are agent builders and others who can actually use those agents. So we see different like governance um different philosophies which then all lead to different governance models u and your platforms that you use to build these agents they [00:44:13] got to support that like you know there has to be enough controls available um in these platforms that allow you to sort of implement the policies and the like you know whatever your for example you may have regulation based needs you know that that need need you to have those kind of um um sort of controls available to you um But I think what is most important is that you may end up you know building agents in many many different systems. Um and and it's actually a good idea like that way you're not dependent on any one uh vendor like you know where you've built all your agents you make sure that um the you can switch you know from one [00:44:53] provider to other depending on who is actually making more progress in the future. So while you can have distributed development, many different systems where you build agents, there are certain things that you should be centralizing for the reasons of governance. Um, one of them is of course context like you know how do how does how do these agents interact with your enterprise systems? How do they actually get the right information, the data, how do they get the ability to take actions in those systems. You know it makes sense to centralize that. Um when it comes to uh security and governance um making sure that agents can only access [00:45:27] information um that that they that they can based on like who they run on behalf of. Centralizing those controls in one place will actually help enterprise um accelerate agent development because otherwise you know you will you will see you know bad things happen and it will sort of um really shut down agent development like you know we've seen these examples you know with our customers and then finally observability like you got to make sure that you centralize um like what are all the different agents you have in the company uh what impact they're having much do they cost what departments are running them um and [00:46:02] is business value that's being driven from them. So there are a lot of these things you know which you want to centralize while still letting people like across the or like you know go and build agents and actually use different platforms as as much as they want. Okay, Ashish, turning to the to the humans here. Um, what do you think this transformation moving from the co-pilot to the agentic era? What does it mean for the humans doing the work? If we're talking about managing all these agents, does managing agents become part of everyone's job? Is the entire organizational hierarchy upended and changed? Is this just a fundamental [00:46:38] shift in how people work? >> Just a simple question for you. >> [laughter] >> I often have my chief people officer ask me do we need a job family for agents. So um I don't think I'm mentally I'm not there yet. So the answer always is is that um every persona depending upon the role that you do is going to have a workbench and think of that workbench being tools and agents that sort of do work on your behalf. Right? So in reality what that becomes is what do you as human in the loop do and what does the agent do and do you have the right guard rails in place for the actions or the outcomes that you intend to act upon in the workflow when [00:47:23] an agent is acting on your behalf. So I almost envision that more so that instead of a job family called agents or an agent having a digital ID which is all true that it's going to happen human beings are going to start to determine sort of when they are the human in the loop in the workflow what actions are they taking as a result of the signals from an agent and what is being delegated to the agent to do and that's sort of still being discovered I don't think there's a clearcut answer on what we would do for deoid or for our clients as a result of that. But that's something that we're taking fairly seriously. At least the workbench part [00:48:00] we've got figured out, right? So every every person at Deoid has based on their persona, the role that they do, they have a workbench, right? And then the workbench is executing certain things as a result of the harnessing of the tools that we've done um on the back of you know sort of everything we've been talking about LLMs, model routing, contextuality, those kinds of things. So I think um eventually there will be such a thing where you have a console or a dashboard that I think I've started to see the beginnings of this even from some of some of your announcements that you guys are making. So when you walk in [00:48:33] what happens on your behalf and what are the things that you do today that sort of you did by yourself without the help of an assistant. >> Can I just add to that there's two philosophies right now that I think would be worth discussing here. One of them is that most people think that all human jobs are going to go away. we're just going to be staring at the beach and have no value to add to society and agents are going to do everything. I think that's such BS at such epic proportions because every single time there's been a step function improvement that I've seen in AI over the course of the past two two and a half years [00:49:11] a massive human bottleneck has emerged. So coding gets automated, code review becomes a bottleneck. Code review gets automated, taste and judgment and what product to build becomes a bottleneck. So I actually don't see in the foreseeable future that these jobs get every 100% of the jobs will get refactored. Um you know some jobs will be lost but there'll be entirely new industries that get created that we have actually not seen created yet. that that that in my mind seems like it's very hard because today AI is still very difficult to con um to um it's it's very bad on the last mile right everything can happen but at [00:49:53] the end of it if you actually don't have humans getting engaged you just get gibberish and I we used to have this thing when when claude first came out like it was so freaking annoying because what would happen is all our people would go out and start generating 50-page documents and all of a sudden I got 150 page documents, but none of them had freaking read the document themselves. [laughter] They just keep sending it over. I'm like, "Dude, have you read it?" So, I I I do feel like there's a level of human finesse that is required at the end and that last mile is pretty hard to go out and, you know, fulfill. [00:50:26] And and then the second area is that all entry- levelvel jobs are going to be lost. We don't need to hire young people anymore. uh and we're just going to keep hiring people, you know, like all entry- level people will be um will have nothing to do and and it would be the stupidest thing for a company to do that because what you would end up doing is you would just not get any fresh thinking. One of the most important things we do as a society is we learn to learn and then we learn to unlearn. And learning to unlearn is just as important as learning to learn. And the learning to unlearn happens when someone pushes [00:51:03] you enough. And so like student interns are my biggest coaches. I spend so much time with student interns coming into Cisco because what they do is they just think about the problem differently, you know. And so I do think that those two dynamics for cultural shift, we should not get carried away by articles in Techrunch and we should just think about the fact that jobs aren't going to get lost. In fact, you will have more demand for jobs, but you'll need dexterity in AI in order to be qualified for a job. And two, entry-level jobs are not going to be gone. In fact, they'll be more valuable now than ever before because [00:51:43] you need more people that are AI native that actually instinctively understand the texture and fabric of AI that can then make sure that they can infuse it into everything that you're doing. >> Well, I want to actually jump off of that with a with a closing question for all of you. Like if I I I feel like no one wants to give real predictions anymore like a year out but put your prediction hat on 6 to 12 months out like what do you think Arvin will be something that we would delegate to AI that we're not yet and something that we still will not be? >> Yeah. Um the I I I think proactive AI like you know we spent some time on it today is [00:52:24] going to be a bigger and bigger part of how AI works for you um in the next 6 to9 months one year um lot of basic things like when you think about like how do you plan your day um where should you be spending your time you know a lot of people actually like one of the biggest complaints I get from people is that I don't prioritize my time you know that I don't you you know spend times on what is most impactful and and I always tell them that like you know like I also like to do things which I like to do like give me a little bit of slack you know I can't always be working on the most high impact things um but but those [00:53:01] are the kind of things you know which I'm which I will hope that everybody all of us will have for example a really good um guide coach chief of staff you know which directs us like helps us spend you know time in the right place that's one thing that's going to happen like right now it's not happening right now AI is something that it's only a tool that you go and um ask it to do some work for you but it is not the one sort of like directing your day for you. So that that's one thing that I'm super excited about. >> Ashish, we just have a 30 seconds but I'm going to give you the last word. >> I don't believe that the judgment that a [00:53:38] human being learns over a period of time given the role that they do that we're going to have a delegate to AI. Right? So this whole thing about everything is identified there's a lot of professions in which it can happen like if you look at regulated entities and what happens in in drug discovery or clinical trials and the things that where human judgment is required right that happens in underwriting that happens in every industry I don't believe that that can be agentified so you know all this talk I hear about and you know the same sentiment that you were at coin G2 right like the apprenticeship model is gone [00:54:09] you only hire experts it's not possible >> it's not possible You cannot bring somebody without apprenticeship that looks like they have 10 years of doing anything. That's not possible, right? Like so it sounds like a great sound bite, but I don't believe it's true. I think like human in the loop is a very important component of how it's going to look like. Just the question is, you know, how much of that human in the loop is so centric to sort of, you know, how we evolve from there, right? So I think I think that's >> I don't have the answer. I don't think anybody else >> I I have to you can't like not let me answer like what's going to be like [00:54:42] Screw the clock. [laughter] Okay. So, 6 to 12 months from AI. In six months, Cisco will have half a dozen products 100% written with AI. No human lines of code. In 12 months, 70% of our $72 billion in revenue products will be written with AI, no human lines of code. In 24 months, Cisco will have no legacy code. And the crazy part of this is we will have more employees in engineering then than we have today. >> I like that. Okay. [laughter] Thank you so much. [applause] >> Awesome. >> Thank you. [laughter] >> Oops. We made it. What a great way to close out day one. A huge thank you to all our speakers this afternoon and to everyone who [00:55:41] joined us throughout the day, both here in San Francisco and online. We'll be back tomorrow for another full day of Glen Go, including our AI native showcase. We'll hear from some of the best and brightest upand cominging AI native companies about the solutions they're building and the problems they're tackling and where they see the next wave of AI innovation heading. And maybe you'll catch a demo or or two. Before we close, I'm thrilled to introduce the inaugural Gingo AI Innovator Awards. [applause] Woo! [cheering] These awards celebrate the people inside our customer organizations who are turning AI into meaningful change with [00:56:33] Glean and beyond. They aren't simply early adopters. They are innovators building new workflows, scaling adoption, enabling their people, strengthening enterprise foundations, and sharing what they learn. So, please join me in celebrating these wonderful AI innovators. Congratulations to all of you [applause] [applause] and don't forget to join us after the keynote for to get collect your awards and we'll grab a photo. All right. And for everybody else, you're warmly invited to join us at Gateway Pavilion. Just a two-minute walk or five minute swim. just stealing that joke from earlier for a reception that's starting now. Thank you very much., Day 1 Afternoon Keynote | Glean:GO 2026, Glean, 01m32jrxy7r7dy3zzjfkvnfzcw.