[00:14] Okay.
[00:16] Hello and welcome to building the
[00:18] connected enterprise.
[00:20] I'm Demetri, director of product for
[00:23] Gemini enterprise agents. I'm joined
[00:25] today by Sammy.
[00:28] Hi folks. My name is Sammy Akram. I'm a
[00:30] product manager in Gemini enterprise. My
[00:33] focus is on context and collaboration.
[00:35] Now we have Florian.
[00:37] Hi. I'm Florian.
[00:39] I'm from Signal Iduna. I'm data
[00:42] scientist working on Gemini enterprise
[00:44] and building agents at Signal Iduna.
[00:49] Excellent.
[00:51] Let's start with a quick reality check.
[00:53] AI was supposed to automate the toil of
[00:55] our day-to-day work.
[01:01] And agent deployments are certainly
[01:04] accelerating.
[01:05] Check out this hockey stick graph from
[01:07] Google Trends.
[01:08] Last year
[01:09] 35% of employees cited broad adoption of
[01:12] agents and over a quarter of executives
[01:15] have increased spending by more than 25%
[01:17] year on year.
[01:19] So enterprise agents are really picking
[01:21] up. You can see this measured by
[01:23] interest, adoption, and spend.
[01:27] However, this is only half of the
[01:29] picture.
[01:30] AI deployments are creating new and
[01:32] unfamiliar challenges that I'm sure
[01:34] you're all aware of. End users have to
[01:36] deal with siloed data, multiple tools
[01:39] for the same job, and agents that run in
[01:41] many places.
[01:43] It's difficult to track budgets. End
[01:45] users lack real continuity across their
[01:48] workflows.
[01:49] And on top of this, cybersecurity and
[01:52] governance are messier than ever before.
[01:55] I'm going to pass it on to
[01:57] over to Sammy to cover how the Gemini
[01:59] Enterprise app addresses these critical
[02:02] pain points.
[02:05] Awesome. Thank you, Demetri.
[02:09] So, the solution to this fragmentation
[02:13] or tool fatigue
[02:15] isn't just another app.
[02:17] It's fundamentally new approach we need
[02:19] to think through.
[02:21] So, we call this concept the fabric of
[02:23] universal intelligence.
[02:25] It is the underlying framework that
[02:27] makes the Gemini Enterprise app the
[02:30] single front door for your whole
[02:32] organization.
[02:34] Instead of forcing you to learn yet
[02:37] another AI system, it adapts to your
[02:40] workflow.
[02:42] The vision comes to life
[02:44] through three core pillars.
[02:47] One, ask and create anything. Second,
[02:49] deploy agents to get job done.
[02:52] And third, harnessing your unique
[02:55] context.
[02:56] We thread these pillars together with
[02:59] meaningful personalization
[03:01] and frictionless collaboration.
[03:04] And we anchor it all
[03:06] with something non-negotiable.
[03:08] Enterprise compliance and governance.
[03:13] First, ask and create anything.
[03:16] We need to break down the creation
[03:17] silos. By leveraging Gemini 3.1 Pro and
[03:20] Flash, you're getting models explicitly
[03:23] optimized for high-frequency complex
[03:26] enterprise reasoning.
[03:28] We're giving you generative media to
[03:31] instantly turn ideas into images using
[03:34] Nano Banana 2 or to video using Veo 3.
[03:38] You get a real-time voice assistant for
[03:41] hands-free execution.
[03:43] You also get multimodal mastery.
[03:46] It analyzes PDFs, CSVs, and repositories
[03:50] without manual uploads.
[03:52] Now, let's get the details of one of
[03:54] those creation mechanism in Gemini
[03:57] Enterprise.
[03:59] The era of copy-pasting from an AI chat
[04:02] chat window is behind us. Meet Canvas,
[04:06] an interactive space built directly into
[04:09] the Gemini Enterprise app that lives
[04:11] side by side with your chat.
[04:14] It is smart enough to know when you need
[04:16] it.
[04:17] Start working on a long-form document,
[04:19] and Gemini automatically triggers
[04:21] Canvas.
[04:22] It instantly grabs your chat context,
[04:25] your connectors, and your sources to
[04:28] build a complete first draft.
[04:30] From there, editing is a breeze, no
[04:33] matter which ecosystem you use.
[04:35] If you are a Workspace user, we embed
[04:38] Google Docs directly into the interface.
[04:42] Full features, auto-saved in your Drive.
[04:45] If you don't use Workspace, you get a
[04:47] robust native editor equipped with AI
[04:50] sliders that you can tweak to change
[04:52] your tone or length of the document.
[04:55] You get full version tracking, one-click
[04:57] exports.
[05:00] But, we didn't just stop at text
[05:01] documents. Canvas is also built to to
[05:05] transform your ideas
[05:07] into slide decks.
[05:09] Using that exact same conversational
[05:11] context, Gemini can instantly generate a
[05:14] structured 8-to-12 slides
[05:17] doc
[05:18] deck
[05:19] on the fly.
[05:20] Just like with documents, Google
[05:22] Workspace users get an embedded Google
[05:24] Slides editor right inside the Gemini
[05:27] Enterprise app.
[05:29] You can make full manual edits,
[05:32] adjust text formatting, styles, colors,
[05:35] without disrupting your workflow. You
[05:37] don't have to kind of jump between
[05:38] different tabs.
[05:40] For those without Workspace licenses,
[05:42] our native HTML editor still allows you
[05:46] to generate those professional looking
[05:47] slide decks,
[05:49] easily export them
[05:50] uh into PDF or PowerPoint and share with
[05:53] your team.
[05:54] Now, Dimitri will discuss the second
[05:56] pillar of the universal intelligence.
[06:01] Great. Our second pillar is get it done
[06:04] with the agents. This pillar is really
[06:07] uh for me it's it's really close to my
[06:09] heart. Um
[06:11] this is the moment we move beyond
[06:14] passive chat and introduce an active
[06:16] execution engine.
[06:18] Think of what we're building as a single
[06:20] pane of glass for your entire
[06:21] organization.
[06:23] A central registry or set of registries
[06:26] where you can manage all of your agents,
[06:27] your tools, your skills in one place.
[06:30] You can manage
[06:32] agents that are Google native, all of
[06:34] our pre-built agents like the data
[06:35] insights agent,
[06:37] third-party agents, or even your own
[06:39] custom-built agents.
[06:42] The Gemini Enterprise app allows you to
[06:44] design bespoke workflows that mirror
[06:46] your exact business processes.
[06:49] And your deployment, it takes just one
[06:51] click to roll them out company-wide.
[06:54] Governance is built right in and your
[06:56] IAM roles are seamlessly maintained. You
[06:59] can assign gateways and policies. All of
[07:02] this works together,
[07:03] which is a really critical insight here.
[07:07] Um this is what my team's been working
[07:09] uh really hard on over the past couple
[07:11] of months, and that's overhauling our
[07:14] agent designer, the one that's
[07:15] integrated directly inside of Gemini
[07:17] Enterprise.
[07:18] Between our natural language approach
[07:20] and our template gallery,
[07:22] anyone in your organization can build
[07:24] agents fast.
[07:27] But simple doesn't mean basic. This
[07:29] interface still gives you total command
[07:31] over complex logic.
[07:33] You can effortlessly snap together
[07:36] conditional loops, agentic reasoning,
[07:39] specific organizational skills,
[07:41] and you can combine both agentic and
[07:44] deterministic orchestration to really
[07:46] balance the creativity that you get from
[07:49] agentic orchestration with the control
[07:50] you have from more deterministic logic.
[07:54] You can also enable human-in-the-loop
[07:55] checks to really build trust in your
[07:57] agents when they take critical actions.
[08:00] And there's much more for you to to try
[08:02] out. This is available now in preview.
[08:08] An agent building is it's not one and
[08:11] done. It's really an iterative process.
[08:13] This is This is something that we
[08:14] believe very strongly. And it's really
[08:16] important that humans stay in the
[08:17] driver's seat.
[08:19] So, users can instantly improve their
[08:22] agents just by providing feedback and
[08:24] the system will automatically suggest
[08:25] smarter prompts.
[08:27] You can play back agent runs, really
[08:30] step through each of the actions that
[08:31] took place to make sure that they're
[08:33] behaving as expected.
[08:37] In preview, we're coming up with a
[08:38] couple of other ways for you to be able
[08:40] to debug in even greater level of
[08:42] granularity.
[08:43] All of this is going to be in an EAP
[08:45] soon.
[08:47] And next, Sammy will cover the critical
[08:49] role that context plays in effective
[08:52] agent deployments.
[08:55] Thanks, Dimitri. By the way, I built a
[08:58] agent to look through my email to figure
[09:00] out which emails I should not miss.
[09:02] Um it worked wonderfully last night.
[09:05] Thank you for for your tip.
[09:08] Awesome, folks.
[09:09] Um so, this is a space that really I'm
[09:12] really passionate about. It's really
[09:15] understanding your context.
[09:17] Um let's level set for a minute.
[09:21] So,
[09:22] um the
[09:23] the current reality of the enterprise
[09:26] LLMs, they're out of the box super
[09:28] capable, right? They know the world, but
[09:32] they do not know your world.
[09:36] An off-the-shelf AI doesn't know your
[09:38] company's internal acronyms, your
[09:40] quarterly objectives, or the specific
[09:43] way your team pharmacy brief.
[09:47] And without that grounding, even the
[09:50] most powerful AI can only give you
[09:52] generic boilerplate.
[09:55] That brings us to our third pillar, deep
[09:58] integrated context.
[10:00] True intelligence starts with
[10:02] understanding who you are and how you
[10:05] work.
[10:06] We have to move beyond the manual work
[10:08] of constantly pasting reference
[10:10] documents,
[10:11] writing a lot of uh prompts in the chat
[10:14] window just to get a subtle like useful
[10:17] answer.
[10:19] Gemini Enterprise solves this by acting
[10:22] as a living extension of your workspace.
[10:25] It has access
[10:27] to Drive, Gmail, Calendar, Outlook,
[10:29] OneDrive, etc. It continuously learns
[10:32] your patterns, your meeting cadences,
[10:36] your important projects, your
[10:38] collaborators, and the topics that
[10:40] matter to you.
[10:43] And it goes beyond just inferring.
[10:45] You can give it explicit saved memories,
[10:48] like sort of the brand voice that you
[10:50] like to use,
[10:51] or different formatting that you like to
[10:53] use.
[10:55] And it will listen to you.
[10:57] Because Gemini Enterprise is integrated
[10:59] deeply into your workflow, it is
[11:01] grounded in your reality. For example,
[11:05] when you ask to draft a project update,
[11:07] it already knows who are working in that
[11:09] project and what is the current focus of
[11:12] that project. You get a specific,
[11:14] relevant answer.
[11:18] Everything we just mentioned is
[11:20] incredible for individual productivity,
[11:23] but work doesn't happen in a vacuum.
[11:26] Work is a team sport.
[11:28] And this brings us to the multiplier
[11:30] effect.
[11:32] Let's address the elephant in the room.
[11:33] Current enterprise AI has two major
[11:35] limitations. First, lack of persistence.
[11:39] Temporary isolated chats force you to
[11:42] constantly re-upload files and rebuild
[11:45] context every time you start a new
[11:47] session.
[11:48] Second, there is no team-wide
[11:50] collaboration.
[11:51] AI tools restrict users to solo
[11:53] interactions,
[11:54] even though enterprise work requires
[11:57] fluid teamwork.
[11:59] Without a thoughtful, shared workplace
[12:01] for teams and AI, knowledge is lost
[12:06] and alignment suffers.
[12:08] To solve this, we're introducing Gemini
[12:11] enterprise projects.
[12:13] This is a persistent, logically bounded
[12:15] workspace that acts as your team's
[12:18] knowledge hub.
[12:20] When you create a Gemini enterprise
[12:21] project, it establishes a strict context
[12:25] boundary. You upload or attach file to
[12:28] this project, and the AI's output is
[12:32] uniquely grounded on these provided
[12:34] files and chat that happens inside of
[12:36] the project.
[12:38] If a file is not explicitly added to the
[12:40] project, it does not exist to the
[12:43] project AI.
[12:45] You can share this secure space with
[12:47] teammates and deploy specific agents and
[12:50] skills inside of it.
[12:54] This transforms the workspace into a
[12:56] shared brain.
[12:57] If a late-joining stakeholder or
[12:59] executive enters the project, they don't
[13:01] need a manual catch-up meeting anymore.
[13:05] They instantly read a AI-generated
[13:08] project overview that talks about the
[13:10] current state of the project. They can
[13:12] have a dialogue with the AI inside of
[13:14] the project to get answers to their
[13:16] question.
[13:18] The answers will be fully informed by
[13:20] the project data
[13:21] and chats that has happened inside of
[13:23] the project so far.
[13:26] This frees up the core team from
[13:28] answering questions that may not be that
[13:31] relevant to make progress.
[13:34] Um
[13:35] Does anyone remember TPS reports?
[13:38] If you don't, just Google it. Yeah, no
[13:40] more of that.
[13:43] Okay, so now
[13:45] once your team is aligned, you need a
[13:48] place to actually execute.
[13:50] You need a team space to iterate
[13:52] together. Now, this iteration happens
[13:55] through chat, like we talked, like I'll
[13:56] tell Demetriou, "Hey man, we need this."
[13:58] Demetriou like, "Okay, let's create a
[14:00] slide deck and send it to me." All of
[14:02] this happens in many fragmented planes,
[14:04] many different tools.
[14:06] But now, enter co-creation. We have
[14:09] built a shared group chat directly into
[14:11] the project where humans and AI
[14:15] build assets side-by-side.
[14:17] Imagine your team iterating in a single
[14:20] thread in a project. It is shared
[14:22] thread.
[14:23] As you fire off instructions, like one
[14:24] person asking for a launch email or a
[14:26] doc, another person immediately adding,
[14:29] "Make it more urgent." The AI The AI
[14:32] acts as a real-time co-author. It
[14:35] synthesizes the feedback, instantly
[14:37] updates the artifact.
[14:40] This is the unmatched alignment of human
[14:43] creativity and AI execution.
[14:48] But as we scale from individual tasks to
[14:50] complex team-wide workflows, there is
[14:53] one non-negotiable requirement.
[14:56] Trust.
[14:57] Ungoverned AI is a security risk, full
[15:00] stop.
[15:01] This is why we built Gemini Enterprise
[15:04] to be a centralized governance hub just
[15:07] as much as an AI platform.
[15:10] We put administrators completely in
[15:12] charge.
[15:13] Every action is backed by deep
[15:15] observability and audit logs,
[15:18] tracking every invocation and tool
[15:20] calls.
[15:21] You are actively defended by model
[15:23] armor, which
[15:26] neutralizes prompt injections,
[15:28] filters harmful content, and
[15:30] automatically masks PII.
[15:33] And currently
[15:35] And
[15:36] And crucially, you don't have to rebuild
[15:38] your security architecture.
[15:40] Gemini
[15:41] simply inherits your existing IAM roles
[15:45] to enforce the principle of least
[15:47] privilege,
[15:48] giving you granular custom controls over
[15:51] every user, tool, and agent interaction.
[15:55] Now, Florian is going to show you how
[15:57] Signal Iduna has been leveraging the
[15:59] power of Gemini Enterprise.
[16:03] Thank you, Sami.
[16:08] So, uh
[16:10] Who is Signal Iduna? Um Signal Iduna is
[16:13] a large German insurance company and one
[16:16] of the biggest private uh
[16:18] private health insurance uh insurance in
[16:21] Germany.
[16:22] Um
[16:23] if you're not from Germany, you probably
[16:25] also know Signal Iduna if you're a
[16:27] soccer fan because uh we are a sponsor
[16:29] of Borussia Dortmund.
[16:31] Um and they play uh at the Signal Iduna
[16:34] Park, which is basically
[16:36] next door from our headquarter in in
[16:38] Dortmund.
[16:40] So, if you if you like European soccer,
[16:42] you probably
[16:43] uh already know Signal Iduna.
[16:49] As you can see um
[16:51] on the right-hand side,
[16:52] uh we have a long heritage
[16:55] um of over 120 years.
[16:58] And we come from a world of paper
[17:01] uh and strict regulations as an
[17:03] insurance company.
[17:05] Um governance is is no buzzword for us.
[17:09] Um there's actually my colleague uh
[17:10] around here somewhere who's who's in
[17:12] charge of of the whole governance
[17:13] process
[17:14] for Gemini Enterprise.
[17:16] Um
[17:17] it's not a buzzword, it's a fundamental
[17:19] pillar
[17:20] um for our customers' trust.
[17:23] And the challenge is how do we transform
[17:27] Signal Iduna into the age um of AI?
[17:33] And one building block um of our
[17:36] transformation uh into the the AI age is
[17:40] T-Mobile Enterprise. So, in October uh
[17:43] 2025,
[17:45] we introduced T-Mobile Enterprise as our
[17:48] central AI platform and our window um
[17:52] to AI for the whole whole company.
[17:55] Um
[17:57] but AI is not only a tool and T-Mobile
[17:59] Enterprise is not only a tool,
[18:02] but um
[18:03] we also have to drive drive AI literacy
[18:07] in the whole company.
[18:08] Um
[18:10] and that that's what we we're trying to
[18:12] do.
[18:14] Um
[18:20] As I said in uh October 2025, we rolled
[18:23] out uh Cozy, how we call um T-Mobile
[18:26] Enterprise internally, um to all of our
[18:29] users.
[18:30] And um
[18:33] it was not just
[18:34] an email, we're live, but we actually
[18:38] made a whole day event out of it.
[18:40] And I'll show you some uh impressions uh
[18:43] in the video.
[18:45] Um
[18:48] as I said,
[18:49] it was not just a tool launch, um it was
[18:52] with the
[18:54] total commitment uh of our executive
[18:57] board.
[18:58] Um
[19:00] we had a lot of different sessions the
[19:02] whole day about AI, about T-Mobile
[19:04] Enterprise,
[19:06] and agents.
[19:07] And
[19:09] yeah, just try to get get the word out
[19:12] that uh everyone can use German
[19:14] Enterprise in Cosi.
[19:17] That they're starting the day.
[19:19] We also introduce Cosima, which you can
[19:22] see on the on the right hand side, which
[19:24] is our mascot for everything related to
[19:28] AI at Signal Iduna.
[19:31] And in front of
[19:33] Cosima, you can see
[19:35] our quite impressive
[19:37] adoption numbers.
[19:39] So, we reached 90 over 90% of our
[19:43] employees
[19:44] with German Enterprise. So,
[19:47] almost everyone at Signal Iduna
[19:50] used Cosi and German Enterprise.
[19:53] We have over 80% monthly active users
[19:56] currently.
[19:58] And
[19:59] started with around 10 agents, which and
[20:02] the number is is still growing. So,
[20:04] we're ramping up
[20:05] agents basically as we as we speak.
[20:11] To drive the
[20:13] AI adoption, we also introduced
[20:17] Yeah, got 110 AI champions throughout
[20:20] the whole whole organization
[20:23] who are driving
[20:25] AI literacy, AI use cases,
[20:28] and also collecting use case ideas in in
[20:33] their lines of of business.
[20:38] And our strategy relies on a wide range
[20:41] of agents
[20:43] starting from very simple
[20:46] basically prompt-based agents
[20:48] going to really complex agents like our
[20:52] health insurance agent Cosica Vao.
[20:56] For the simple agents, one example is
[20:59] corporate language
[21:01] agent, which basically helps to
[21:03] translate
[21:05] I would say Florian
[21:07] to Signal Iduna German.
[21:09] So, that that the whole communication uh
[21:13] matches our brand guidelines and brand
[21:15] values.
[21:17] Um
[21:20] Complexity-wise, in the middle is
[21:22] something like our HR agent, which has
[21:24] access to um all of our um HR knowledge
[21:28] and answers questions related
[21:32] to vacation policies, benefits, and
[21:35] everything around HR,
[21:38] um reducing the workload for our HR
[21:41] teams uh dramatically.
[21:44] And as I said, uh
[21:46] when it comes to more complex agents,
[21:48] um there's Kosika Frau, which is our
[21:51] health insurance agent,
[21:53] um
[21:54] which is
[21:56] very capable in answering uh health
[21:58] insurance-related questions, and is
[22:02] deeply integrated into the core core
[22:04] business of uh Signal Iduna and our our
[22:07] health insurance.
[22:09] And it's supporting experts uh
[22:11] navigating the highly complex topic of
[22:13] health insurance,
[22:15] and uh yeah, answers questions around uh
[22:19] processes and and all other topics.
[22:23] Now,
[22:25] let's dive deeper into um Kosika Frau.
[22:28] Um
[22:32] the goal of Kosika Frau is mastering the
[22:34] complexity um
[22:37] of health health insurance.
[22:39] Um
[22:41] it has access to a lot of knowledge, so
[22:44] we have over 600 different uh
[22:47] insurance tariffs, or health insurance
[22:49] tariffs,
[22:50] which have
[22:52] yeah, completely different uh different
[22:54] benefits and and regulations, basically.
[22:58] And we also gave it access to over 800
[23:01] documents and data sources from
[23:03] different systems. So, it really has
[23:05] access to the the company's
[23:08] knowledge
[23:10] of of health insurance.
[23:11] And that's internal guidelines.
[23:14] It's systems like
[23:16] which which health insurance policies
[23:20] a customer has and so on.
[23:24] And it seems to integrate it into
[23:26] Chemona Enterprise to give yeah every
[23:29] every health insurance employee
[23:31] easy access to to Corsica FO.
[23:35] And providing that domain knowledge
[23:37] where it's needed. So, no matter if you
[23:38] are
[23:40] call center service agent working in
[23:43] customer service or managing claims,
[23:47] you have access to to
[23:49] Chemona Enterprise
[23:51] in
[23:52] to to Corsica FO in Chemona Enterprise.
[23:55] And it's grounded in truth. So, every
[23:57] answer
[23:58] is explained
[24:00] how Corsica FO comes to the the
[24:02] conclusion or the the result.
[24:05] And you get
[24:07] links to the data sources the the answer
[24:10] relies on or is based on.
[24:13] And there's always a human in the loop.
[24:15] So, currently we don't do
[24:18] automation here.
[24:20] That's that's really important for us
[24:21] that
[24:22] still a human is in charge of of
[24:24] decisions
[24:25] in in health insurance.
[24:29] So, how do we manage the complexity
[24:31] under the hood?
[24:33] Corsica FO consists actually
[24:37] of different sub agents.
[24:39] And they are broke down
[24:43] a breakdown basically the whole
[24:44] insurance journey. So, there's one sub
[24:46] agent who's who's in charge of all the
[24:48] operations topics. So, that's all topics
[24:51] related to your to your contract.
[24:55] Then there's
[24:57] benefits agent
[24:59] who
[25:00] knows exactly
[25:02] which
[25:03] services and benefits are ensured in
[25:05] your in your policy.
[25:08] And
[25:10] yeah,
[25:10] the benefits agents manage actually the
[25:13] the complexity of our
[25:15] 600 600 tariffs that are
[25:18] in
[25:19] available in in Kosika 4.
[25:23] And then then then there's the customer
[25:25] communication
[25:26] agent, which makes it easy for our
[25:28] employees to
[25:30] write empathetic
[25:32] and also efficient
[25:35] yeah, messages
[25:37] and correspondence to to our customers.
[25:44] Having a great agent is one thing, but
[25:46] the ultimate test is reality. So
[25:49] we're probably some of you also building
[25:51] agents and know that adoption is
[25:54] sometimes quite challenging.
[25:56] Um,
[25:58] it was the case for for Kosika 4 2.
[26:01] Um,
[26:02] but in Q4,
[26:05] we had
[26:06] premium adjustments for a lot of our our
[26:09] policies and customers,
[26:11] um,
[26:12] which was really a huge challenge that
[26:15] we could foresee
[26:17] because we we anticipated that we get a
[26:19] lot of customer inquiries
[26:21] um, about the the premium adjustment.
[26:25] And so what do we do in the times of
[26:27] agents? Of course,
[26:29] you build an agent supporting our
[26:32] supporting our your employees. And
[26:34] that's exactly what we did.
[26:35] Um,
[26:38] we build an
[26:40] another agent for for the premium
[26:42] adjustment, which we
[26:44] integrated into Kosika 4,
[26:47] which basically answers all questions
[26:49] around the the premium adjustment.
[26:53] Um starting from the uh
[26:55] being very efficient in
[26:57] answering recurring questions,
[27:00] um but also going really deep into
[27:02] providing information about really
[27:05] customized
[27:07] uh uh uh
[27:08] tips and and
[27:10] uh yeah,
[27:12] ideas maybe how to adjust your your
[27:15] coverage um
[27:18] during the the the premium adjustment.
[27:21] And
[27:22] as I said, we needed a solution that
[27:23] supported our um our our colleagues
[27:26] because uh of the high volume of of
[27:30] inquiries we we expected.
[27:35] And as I said, that's what we basically
[27:36] did. We built an sub-agent for the
[27:38] premium adjustment um
[27:40] and
[27:42] having built the the base with Kosika
[27:45] VAU and the other sub-agent, it was uh
[27:48] we could quite quickly develop the the
[27:50] premium adjustment agent.
[27:52] And giving it in into the hands of of
[27:55] our um domain experts, basically.
[27:58] And the impact was really really
[28:00] staggering
[28:01] um
[28:02] because
[28:06] before we had a a lot of knowledge in
[28:08] the in the heads of of our our health
[28:10] insurance experts. So, they didn't
[28:13] really need an agent to answer the
[28:15] questions because they had a lot of
[28:16] experience.
[28:17] With the premium adjustment and the lot
[28:19] of uh special special and and uh
[28:22] customized questions uh the customers
[28:25] had,
[28:26] um it was totally different.
[28:28] So, they really needed information they
[28:31] couldn't have at hand at that time.
[28:34] And
[28:35] not only could uh the the benefits agent
[28:38] uh the the premium adjustment agent
[28:41] assist our employees uh with the complex
[28:44] cases and and conversations,
[28:46] um
[28:48] but it also
[28:50] drove the adoption of of Cosiga FO by
[28:53] 400%. So, we see saw a dramatic spike in
[28:57] uh in usage there.
[29:01] But as successful the the agents have
[29:03] been,
[29:04] um
[29:05] we really want to integrate them uh more
[29:08] deeply more and more deeply in in Chemna
[29:10] Enterprise.
[29:11] And that's what we we're currently uh
[29:14] working on.
[29:17] And that's basically building custom UI
[29:18] UIs with uh A2UI integration uh in
[29:22] Chemna Enterprise.
[29:24] And so, what does it do?
[29:27] It basically gives you the the
[29:29] it gives us the possibility to have
[29:32] custom UI elements in Chemna Enterprise
[29:36] like a drop-down of for our uh 600
[29:39] tariffs.
[29:41] So, uh as you can see they are also uh
[29:44] often quite similar
[29:46] uh similar names of of our tariffs. And
[29:48] it's really hard to uh to do all that uh
[29:51] just in in a text conversation. So, we
[29:54] are really looking forward to building
[29:56] uh custom UIs with A2UI in Chemna
[29:58] Enterprise.
[30:04] So, what are our what are our
[30:06] key takeaways
[30:08] um for Chemna Enterprise and and uh what
[30:11] we saw with with the premium adjustment
[30:13] agent?
[30:17] It sounds simple, but capable agents
[30:20] actually get adopted.
[30:22] So, if you if you really need the pain
[30:24] and the the business need of your your
[30:26] uh
[30:27] company,
[30:29] the agents will will get adopted really
[30:31] easily.
[30:34] Second,
[30:35] um
[30:38] it's so something like uh introducing
[30:42] Gemini Enterprise or
[30:45] specialized agents for employees is a
[30:48] journey. It's not a project.
[30:50] So, you really have to start a movement
[30:52] in inside your company.
[30:55] And uh
[30:57] yeah, provide continuous information,
[30:59] continuous training, and support for
[31:02] employees to drive drive adoption and um
[31:07] drive adoption.
[31:09] Third,
[31:12] laying down organizational and technical
[31:15] and governance foundations feels slow at
[31:18] the beginning,
[31:20] but it actually makes it quicker at the
[31:22] end.
[31:23] So, if you have strong foundations, you
[31:25] really can
[31:27] uh
[31:29] uh
[31:32] you react you can
[31:33] easily react
[31:35] quickly to business needs and uh support
[31:39] your your whole company um
[31:41] and business demands.
[31:45] That's it from uh yeah, our customer
[31:47] experience talk.
[31:49] Back to Dimity. Dimity. Thank you.
[31:56] Thank you so much Florian for making
[31:57] that concrete. And to zoom way out and
[32:00] to wrap up, here's what I want to leave
[32:02] you with.
[32:03] AI agents are evolving rapidly and will
[32:05] soon reshape how every organization
[32:08] operates.
[32:10] But their success depends on the access
[32:12] to the right tools, data, and context.
[32:15] It's critical that users and admins
[32:18] trust their agents and love to use them.
[32:21] And Gemini Enterprise is the platform
[32:23] that integrates those essentials into a
[32:25] secure and governed place.
[32:27] We look forward to seeing how you use it
[32:29] to transform your organizations.
[32:31] And with that, we're happy to answer
[32:33] questions. Thank you again.
[32:44] >> We have a couple of mic runners if folks
[32:46] have questions about
[32:48] anything related to Gemini Enterprise
[32:50] app.
[33:01] Uh yes, thank you for the presentation.
[33:04] Just I got one question.
[33:07] Uh so we got Gemini Enterprise in
[33:10] private preview for a few weeks. And one
[33:13] of the problem we found out is that
[33:16] uh we needed to deploy MCP for every uh
[33:21] tools and internal app here that we
[33:23] have. And so we became the bottleneck of
[33:27] our users. And uh
[33:30] what I did not found on this is for for
[33:33] example with tools like Cloud Code and
[33:35] whatever where you have code execution
[33:37] and and uh
[33:40] uh skills. Yes. Uh
[33:42] you I did not see this kind of stuff in
[33:46] Gemini for Enterprise. So
[33:47] is it will this be available?
[33:50] Yes. Uh
[33:51] the short answer is on connectors, we're
[33:54] rapidly expanding our pre-built
[33:56] connectors and then we're making it
[33:58] easier to bring your own MCP through
[34:01] um
[34:01] our integrations with a number of like
[34:03] API gateway tools. Um on skills, we have
[34:06] a preview of skills across Gemini
[34:08] Enterprise both agent building and usage
[34:10] in assistance.
[34:12] Um
[34:15] and then uh code execution, we have kind
[34:18] of a basic version of it that's in the
[34:21] agent designer but we're going to expand
[34:24] and do much more with code execution
[34:25] shortly.
[34:27] So.
[34:29] Stay tuned.
[34:31] Awesome. Hey, thank you guys very much
[34:33] and enjoyed that. Uh
[34:35] Maybe a follow-on to that and and then
[34:37] an additional question. So,
[34:39] do the agents have the ability to take
[34:41] action within the other connections and
[34:43] systems, right? Like, you know, execute
[34:45] a snow ticket or something like that.
[34:47] And then along those lines,
[34:49] how does the the privileges work for all
[34:52] of these different tools? Like, if I'm
[34:54] connected to Drive and ServiceNow and
[34:57] SharePoint or something like that, how
[34:58] does the privileges flow through?
[35:01] Yeah.
[35:02] In general, the authorization follows
[35:05] the user. So, as a user that's invoking
[35:07] an agent, I if I haven't already
[35:10] authorized access to a given
[35:13] data source or tool, I'll have to.
[35:16] We do support a mix of both kind of read
[35:19] and kind of write
[35:21] or execution sort of base tools.
[35:24] Execution or tools that are
[35:28] potentially destructive
[35:29] require human approval. And so, that's
[35:31] all surfaced kind of seamlessly through
[35:32] the Gemini Enterprise app. But, we
[35:34] absolutely like feel very strongly that
[35:38] agents should be able to not just
[35:40] get back data, but also do useful things
[35:42] with it.
[35:43] And then lastly, we're we're adding
[35:45] support for triggers. So, you can run
[35:46] these agents on either schedules or
[35:49] external events. Like, an example would
[35:51] be when when a whenever a new Jira
[35:53] ticket is created, we're going to run
[35:55] the specific agents processes that Jira
[35:57] ticket.
[36:00] Hey.
[36:01] Hi, good morning. Here.
[36:03] So, I have like couple of questions.
[36:06] Like, first question is
[36:08] when when we have this like last year
[36:11] when Gemini Enterprise was in preview,
[36:14] you guys have
[36:16] used index-based search mechanism.
[36:20] And then recently, you are trending
[36:22] towards mostly federated search, right?
[36:26] One pain point that we have like with
[36:29] some of the products that
[36:31] people use here like Atlassian products
[36:33] or like some such products.
[36:35] Right?
[36:36] The search is not great, right? And um,
[36:39] the first thing the second thing is like
[36:40] when we
[36:42] when we send the request to
[36:44] um, uh, Gemini, Gemini is forwarding the
[36:47] request to
[36:49] uh, maybe Atlassian products example,
[36:51] right?
[36:52] So, if they send the response back which
[36:55] is not great.
[36:56] So, Gemini is going to use the same data
[36:59] and then it tries to provide the
[37:00] reasoning on information to user.
[37:03] Uh, is it not a great user experience?
[37:06] That's the first thing.
[37:07] The second thing is like
[37:10] there are like two types of use cases
[37:12] that we see in general with Gemini
[37:13] Enterprise in my organization
[37:15] especially. Like one
[37:17] where people
[37:19] uh, there is a public information where
[37:20] everybody can access like policies in
[37:23] the company.
[37:24] There is a private information which is
[37:26] probably tied up to a person, right?
[37:29] Federated search probably is okay when
[37:31] the user we ask user to go login and get
[37:33] the data, right? Through through forward
[37:36] forwarding the request. But when there
[37:37] is a a public data where for example,
[37:41] the user Y doesn't have an even access
[37:43] to let's say Confluence.
[37:46] With this
[37:48] federated search, are we not forcing the
[37:50] users to you know use
[37:52] or like get the license of the
[37:53] Confluence?
[37:55] Rather than I have this private data or
[37:57] public data available for my company,
[37:58] why are we even forcing the users to
[38:00] login?
[38:01] So, these are two questions I have.
[38:03] Thank you.
[38:04] Thank you for the questions. Uh, I I
[38:06] think the short answer is the the shift
[38:08] towards
[38:09] uh, federated connectors was largely
[38:11] driven by both performance on specific
[38:14] evals and making sure that we can kind
[38:17] of scale up to many other data sources.
[38:20] Um, if if you are experiencing issues or
[38:22] have instances where it's not performing
[38:24] as expected, I'm happy to like follow up
[38:26] specifically on that. Um, and then as
[38:28] far as like mixing public and private
[38:30] data sources, um, we do want to make it
[38:32] more seamless. So, uh, we'd love
[38:34] feedback on how we can
[38:36] potentially prevent users from needing
[38:38] to sign in when they don't need to. Um,
[38:41] but it it in general, I think we're
[38:43] going to support
[38:44] kind of more varieties of auth. I think
[38:47] right now agents, the way I think of
[38:48] them is they kind of run on your behalf.
[38:51] But I think the future looks much more
[38:52] like you have agents that run as an
[38:54] extension of you, and then there's
[38:56] agents that have their own sort of
[38:57] service account identities, where I
[38:59] think a lot of what you're describing
[39:00] would would sort of take place. Um,
[39:03] but yeah, that's definitely a vision
[39:04] that we're moving towards very quickly,
[39:06] so uh Right. It It sometimes, uh, you
[39:09] know,
[39:11] also it depends on the other side,
[39:12] right? Where they don't want us to
[39:14] ingest, right? Going that they
[39:17] they're also updating their policies to
[39:19] like, "Hey, don't do ingestion, just do
[39:21] federation." So, it's a mix of things.
[39:23] So, definitely would like to hear more
[39:24] in-depth feedback on this.
[39:30] Right.
[39:35] That's right. That's right.
[39:38] Hey there. I have two questions
[39:40] regarding to the Zoomdata Enterprise
[39:42] with the A2A integration. Uh, my first
[39:45] question is we uh we have a use case
[39:47] which our A2A server is dep- loyed
[39:50] within our private network. And when we
[39:54] trying to integrate the this A2A with
[39:56] the Zoomdata Enterprise, we find like
[39:59] the Zoomdata Enterprise unable to do the
[40:02] configuration to connect with our
[40:04] internal service within our network. So,
[40:07] this is like we'd like to understand how
[40:11] this one can be addressed. This is my
[40:12] question one. The question two is uh we
[40:15] also use the ADK to build our founda- uh
[40:19] uh foundation engine and then explore
[40:21] the ADK as the A2A engine. Within the A2
[40:24] ADK engine, we
[40:26] uh use that for example open API tool or
[40:28] NCP tool, which we define the native
[40:32] credential measure within the ADK by
[40:35] using the off ski- schema and off
[40:38] credential. But when we use this measure
[40:41] and we deploy our agent to the Gemini
[40:43] Enterprise, it looks like the Gemini
[40:46] Enterprise unable to understand the
[40:48] event come from the ADK and A2A
[40:52] integration. Uh are there any better
[40:54] solution for this one? Thank you.
[40:58] Yeah, I'm happy to follow up on the
[40:59] second question because I think I would
[41:01] need a little bit more context, but um
[41:04] by and large, I think the way to think
[41:05] about how to plug in external agents is
[41:07] going to be through what we're calling
[41:09] the agent registry.
[41:11] This is part of kind of the new
[41:13] expanded Gemini Enterprise agent
[41:15] platform. So, you can think of Gemini
[41:17] Enterprise as sort of two layers.
[41:19] There's the platform layer, which is the
[41:21] registry,
[41:22] it's kind of the high-code experience,
[41:24] it's a management plane. And then Gemini
[41:26] Enterprise the app is the end user
[41:28] experience. Um A- A2A, internal A2A, all
[41:33] of that should be able to plug into the
[41:34] registry,
[41:36] and be defined as kind of one entry
[41:37] that's governable that you can assign a
[41:39] policy to, and then then that's exposed
[41:42] into Gemini Enterprise.
[41:43] Um
[41:45] And so, whatever connectivity issues
[41:48] you're facing,
[41:49] um
[41:50] we should be able to address those at
[41:51] that like platform layer. Um so, there
[41:54] should be much less of a dependence on
[41:55] anything that's
[41:57] Gemini Enterprise app specific.
[42:03] Hi. Um so, first of all, thanks for the
[42:04] great talk to you guys. Um
[42:07] we are really excited about the overhaul
[42:08] of the agent builder. So, I have uh two
[42:11] questions regarding that.
[42:13] Uh the first one is um when it comes to
[42:15] the human in the loop feature. For for
[42:17] instance, if I have a team um I I want
[42:20] to build an agent to automate a decision
[42:22] process in the team.
[42:23] And I have for instance two two persons
[42:25] which are
[42:26] then
[42:28] um should should should have the
[42:29] decision in the end and be integrated
[42:31] into this human in the loop step. Would
[42:33] it be possible to build in a process
[42:36] where an agent um
[42:38] um results to for instance those two
[42:42] persons being um
[42:45] maybe prompted by an email to actually
[42:50] work on this decision and
[42:53] on this human in the loop step. Yeah.
[42:55] Yeah, I mean this is very in line with
[42:57] how we're thinking about collaborative
[42:58] projects
[42:59] where you can deploy your agents not
[43:01] just to kind of your own space but to
[43:03] your shared workspace with team members.
[43:06] You can jointly take actions or kick off
[43:08] agents all with shared context, files,
[43:10] skills within a sort of predefined
[43:13] context boundary. Right. And also like
[43:16] over time we we will build the
[43:18] governance like more complete. I think
[43:20] if you want to have multi-step
[43:22] permissioning, right? Something totally
[43:24] makes sense.
[43:26] Right.
[43:27] Not sure if it is in the road map right
[43:28] now, but you know.
[43:30] We'll definitely think through that.
[43:31] Great to hear.
[43:32] Uh second question would be
[43:34] so when we look at the agent builder in
[43:37] Gemini CX um it has uh functions which I
[43:40] haven't seen yet in this overhaul um for
[43:44] the workflow agent designer in Gemini
[43:46] Enterprise. For instance, there is a
[43:47] possibility I think to integrate custom
[43:50] code
[43:51] directly as as in as it as within the
[43:55] low code agent builder. So, is that
[43:57] something which will also come to the
[43:59] work workflow agent designer in Gemini
[44:01] Enterprise?
[44:03] It it's it's it's on the road map. Uh
[44:05] the
[44:06] you you'll hear here first, but um
[44:09] we we do want to play
[44:11] um
[44:12] much more nicely but with high code
[44:14] inside of the low code agent designer
[44:17] and to be able to run arbitrary code
[44:19] blocks is definitely something that's in
[44:22] line with our vision and it's something
[44:23] that we're prioritizing pretty pretty
[44:25] soon. And just one one last question.
[44:28] Um
[44:29] so in in terms of feature parity between
[44:31] those two products, Gemini CX and Gemini
[44:33] Enterprise app um will it be aligned or
[44:37] will it always be uh
[44:39] developed separately?
[44:41] I think in general we want to be a super
[44:44] set of what's supported in
[44:45] Gemini Enterprise for customer
[44:46] experience because
[44:49] the agent designer supports sort of two
[44:50] styles of workflows. It supports both
[44:52] workflows that are chat-based, that are
[44:54] interactive, and so we want to have all
[44:56] of the goodness that GE for CX has. And
[44:59] separately, we want to support workflows
[45:01] that are like kind of headless
[45:03] background tasks that are doing work um
[45:06] on a schedule, creating reports, or
[45:09] responding to events. So there's a
[45:11] different set of capabilities needed to
[45:13] support those kinds of workflows. So in
[45:15] short, we want to be
[45:17] have all of the features that GE for CX
[45:19] has and more
[45:20] uh within the agent designer.
[45:22] And I we're investing very heavily on it
[45:23] on kind of closing that gap.
[45:25] Makes sense. Thanks.
[45:28] Awesome. Thank you so much for for
[45:29] joining us here. Uh please ask us if you
[45:31] have any more questions. Uh we'll we'll
[45:33] we'll be around.
[45:35] Thank you, folks.