Name: Claude for Financial Services teams: Putting agents to work
Uploaded: 2026-05-07T16:31:44.511Z
Duration: 59 min 20 s
Description: Claude for Financial Services teams: Putting agents to work

Transcript for "Claude for Financial Services teams: Putting agents to work": Hey, everyone. Thank you so much for joining us today. Earlier this week, we released several new capabilities to really make Claude an even better thinking partner for financial services teams across many different types of use cases. You'll see that we launched several new agents as well as the ability to deploy those agents via both Cowork and Cloud managed agents, new add ins across the Microsoft three sixty five suite, and new connectors to really ensure that Claude has all of the same context that's needed to do really effective work. So in this session, we're excited to really go into all of those capabilities in detail and walk through how you can set up, and start taking advantage of everything that we released. So so excited to be joined today by Nick Lynn, our head of product for financial services, and Owen Scully, one of the leaders from our applied AI team. Next slide. Great. So first, a few quick logistical notes. So, the first question that we always get is, can I get a recording of this session? So, very happy to share that, yes, we will definitely be recording, and we'll be sending this to you, after the session. So if you need to drop at a point or want to share the content with a colleague, we got you covered there. Second, we do want this to be a very interactive session. We're really going to be going into detail, showing how to take advantage of all the features. So please use the q and a tab in the webinar, and, we're going to, be doing our best to address as many of the questions as we can during our dedicated q and a time at the end. You can also upvote some of the questions from your peers. And finally, we'd love to get your feedback, and we'll be sharing a short survey toward the end of the session, and it really does help us make these sessions better. So if we move to the next slide, we also really highly encourage you all to use the ask Claude feature at our docs to get, any questions that you may have answered at any time as well. Next slide. I think I'm handing it over to Nick, so carried away. Thank you, Nancy. Hey, folks. My name is Nick, and I lead our product efforts for Cloud for Financial Services. I'm also a recovering investment banker and private equity investor. So I remember all of those late nights, you know, ticking and tying, like, cell models and making sure that the text boxes on my PowerPoint slides are the same exact shade of blue. Now we think that finance is just a few months behind coding when it comes to the exponential. And I've been witnessing a lot of this progress firsthand, so I'm pretty excited to to show you all what is to come. And, my job is to make sure that the exponential actually shows up inside of your organizations. Now what is Claude? We think Claude is one thinking engine. At the foundation are the models. Right? We are a frontier research lab, so nothing else really matters if the models are not the best. On top of those models sits a platform. So APIs, tools, manage agents, essentially, an ecosystem and operating system you can build with. And on top of this platform, we build applications. We do it together. Now we think that means two things. One is Claude alongside your people, so really accelerating and transforming how we all work. Second is cloud really inside your systems, really fully customizable to you and your workflows. Now this whole engine really brings and comes together to create agents. Agents that we think can do useful things for your org. So within creating pitch decks, discounted cash flow models, valuation reviews, many of the same use cases that you all are familiar with. Now every one of these agents is really built on top of the same foundational Lego blocks, if you will. Now let's imagine we're bringing Claude into your organization for the very first time. What do you give it? Access to a computer with tools like S and P and Moody's, training on how a DCF model is done at your firm, and a set of instruction manuals on things like calculating the discount rate and building comp sets. This is what we mean by connectors, skills, and sub agents, the foundational building blocks that really help Claude get the job done. Now what's new? We took those little blocks and did a lot of the heavy lifting for you. On Tuesday, we released 10 agent templates for some of the most important and time consuming tasks in finance. You know, pitch meeting prep, model building, accounting, and more. And you get to decide how these templates show up. If you want them working alongside your people, install them as plugins in Cloud Cowork and Cloud Code. If you want them working autonomously inside of your systems, deploy them as cloud managed agents. Now let's dive into one of these templates for smarter employees. You know, we're all knowledge workers. Right? We do a few things like retrieving data, running analysis, and creating outputs. What does that actually look like in action for an investment bank? Let's meet Sarah. Sarah is an executive director at the natural resources group at Arcos Bank. She's out at an industry dinner when an urgent email lands in her inbox on a possible tape private transaction. Now every banker probably knows what happens next. Right? Three analysts just lost the weekends, and tonight is going to be a pretty long one. Except tonight, Sarah has Claude. From her phone, Sarah forwards Claude the email through voice mode and CoWork dispatch, and Claude is able to get to work immediately. Under the hood, Claude is running the pitch builder plug in inside of CoWork, one of many we have in the library. The pitch builder is prewired into FaxSet, SMP, and LSEG. It is also pulling our comps analysis skill, which really encodes how a banker actually builds a comp set and operates Excel. By the time Sarah's heading home, Claude already has prepared a preliminary work plan with a detailed execution, recommended next steps, and a first cut of the comp sheet and leveraged buy off model already in motion and ready for her to review. Now Sarah is back at at her laptop. Inside claw for Excel, she can start to refine on these outputs. Starting with a comp set, she pulls out one period that doesn't feel quite right. The formulas flow straight through. She double checks the methodology and reviews that the narrative still looks good. And then comes the LBO. She has some intel of this the sponsor is willing to stretch on leverage and asks Claude to rerun the scenario more aggressively. Claude updates the model in real time as well as the summary narrative with every source tied in every number cited. And that's across the operating model, the desk schedule, and the returns analysis. So no more manually checking scale, and Claude is really able to do that heavy lifting for you. Alright. Finally, the deck. Now the part that all of us has probably lost many Saturdays to, instead of building in PowerPoint, Sarah redirects Claude to Claude Design. Now the beauty of Claude Design is that it doesn't just produce static slides. It builds interactive dashboards and prototypes. So within minutes, as you can see, Sarah has a polished deck with live sliders where she can adjust and flex her assumptions like entry leverage and exit multiple. And she can watch the returns move in real time as well. So this is really a conversation with our client, not just a deck presentation. The last step, let's get it over to the client. Within Claude for Outlook, Claude drafts the email directly. Sarah edits the language with Claude, iterates it until it's client ready, and finally is able to send it with full confidence. Now let's recap what we just saw. The pitch builder agent lived inside the tools that bankers and all of us use every single day. As of this week, cloud is now generally available across Excel, PowerPoint, and Word, and we're also launching cloud for Outlook for the very first time. That means that the same cloud now works across your entire cohesive workflow. And then the analysis you just saw is also grounded in what we think is the most comprehensive data ecosystem we've ever assembled. As of today, we're live with eight new data partnerships, including many friends on the call, GuidePoint, Dun and Bradstreet, Verisk, and more. Now let's look at faster processes. We're also launching five new agent templates across some of the heaviest operational workflows within Phineas, like valuation review and accounting. Let's look at what one of these workflows looks like inside cloud managed agents. Meet Maya. Maya is a senior fund accountant at Arcos Fund Services with a dozen buy side clients, each of them requiring portfolio valuation reviews. So quarter end has always been a nightmare for Maya. With cloud managed agents, she can own the quarter end with much more confidence. She kicks off the valuation review agent. That agent spawns one sub agent per portfolio company, each one ingesting the company's financials and running the valuation analysis in parallel. Now Maya watches all of them execute live inside of our console, ingesting, valuing, and reviewing. She opens up one of these packages, double checks that the EBITDA bridge correctly matches the financials, and approves. And finally, she hands off the work to the statement auditor agent for a final pass. What you just saw was a fleet of cloud managed agents working together. Managed agents are prebuilt, configurable agent harnesses that run directly inside of our own managed infrastructure. They're really meant for long running asynchronous work. They have full observability in our console and are customizable to your specific processes. Now we've heard that processes like these take your teams weeks to months to deploy. We're hoping that by publishing these cookbooks, you can immediately customize and build into your code base. And all of these agents work in Cloud co work as well, so you get to choose where they run and how they run. Right. Now I'm gonna pass it off to Owen to show you how exactly we deploy these capabilities into production. Thanks, Mitch. So what I want to do is take a little bit of a look, under the covers at what Nick just talked about and show you, first off, what is a skill? How does that surface to your users in in co work? Secondly, how can you build a skill so that you can create one of these yourselves that meet your own process needs? Then moving on to the new exciting stuff that that was announced, I'm going to show the cloud within, within Excel, within PowerPoint, within Word. And then I'm going to show you how to use some of the new, agents we've just announced and try to use them within core work and also how to adapt them yourselves within your own environment in order to create your own, specialized agents using what we've provided as a template. Let me go and share my screen here. I'm going to share my view of Claude. There we go. I will, heads up, be switching over and back between sharing various different, applications as you go through this. So, please bear with me as I as I toggle between each of these in the, in the webinar. But the first step then I wanna show, you know, what is a skill? How can you use skills to make each of your workers more effective within their use of cloud and doing things that, you know, they wouldn't have been possible to dream of doing by themselves. But also to do it in a way that is consistent across your organization so that you have a consistent processes for doing things like earning analysis here in the example I'm about to show. So let me kick this off here. I'm going to say within, teleco work, perform an earning analysis on Apple. This is, obviously, a very simple prompt. This is not giving very much information to Cloud about what it needs to do, you know, what exactly is in learning analysis, how can I do this on Apple? What we're seeing here as Cloud is thinking, it has identified a skill that is relevant to this. So it has identified the Silverburn earnings analysis skill. In this demo, I'm a part of a Silverburn Capital, a fictional, organization entirely. But let's have a look at what what's in this skill. So while Claude is doing its work there, let me pull this across so you can see a little bit better on the screen, hopefully. This is what what what that earnings analysis skill is. So you can see here at the top providing a description of the skill. This might be really important. We'll come back to this in just a moment. A description of what the skill can do, and then you can see everything that this skill needs to accomplish. So exactly what we're trying to do, create a professional earnings update report, giving characteristics of the output we expect, specifying when to use this skill. So whenever anybody in, a cloud, a co worker, cloud environment asks to create an earnings update, analyze, post earnings report, any of these will trigger the use of this skill by cloud. You can see here how I triggered it for an earning analysis on Apple. A very simple way to to to trigger. One thing I will ask you to note off here, because I'm gonna come back to it later on, is do not use if the user requests an initiation report. So I'm gonna show how to install a different skill that will handle the initiation later on when you get into the, the the newly announced skills that we have created specifically for FSI. But you can see here the skill is is, you know, very, very long. This is not something that a user would write themselves, type themselves, but something that they can create and share so that everyone across your organization is using creating these earnings analysis reports according to a standard set of of guidelines and rules. They're gonna create standard outputs. I mentioned that this description here at the top is very important. This is what Cloud uses to know when to use this skill or what this skill can do for it. And this is important, gonna dive a little bit techy here, for for context management. So if you think about the context window that you want to manage in in any, execution against cloud, you know, it's very large, obviously, now with with a topic providing 1,000,000 token, context windows. But even so, managing the context, making sure you're not loading too much into context is a key skill key ability. And this is what we're doing with, progressive disclosure within cloud. So the, entire skill itself is not loaded into context until cloud determines that it needs it. So I'm gonna pause there. The description of the skill is made available to Claude, so Claude will always know in context when it can use this skill. But the entirety of the skill, all these pages and pages of of skill description, instructions, output formats, this is not going to be in context until Claude decides that it needs them. That's a a very useful tool. You're gonna be using this all the time to manage context within your your organization. So there's our skill. Let's get back to to see what's going on here. So So you can see that this skill that I triggered, in my co work session has asked me for some additional information. What quarter should I analyze for this earnings report? Let's do, let's do q four twenty twenty five just for fun. What's my current rating on Apple? I'm gonna say it's a buy. I don't have a current price target or don't know what my price target is. I'm gonna focus on let's look at the China exposure and iPhone units specifically. And then I can't upload an earnings model. So from my, desktop, I'm going to upload an earnings model for this, and then I'm going to continue. And Claude is going to use this additional information as context. So Claude was smart enough to know that with that very simple prompt I gave it to trigger this analysis, it didn't have enough information. So it came back and asked me for the additional information that it thought would be useful in fulfilling this request to create this, earnings analysis for Apple. We can look here and see clouds thinking in in progress. So you can see it's reading the files that I gave it. It's uploading tools. It's determining how it's going to best create this, this earnings report. Now this is going to take some little time to run through all of this. You You can see the progress with that up here in the upper right showing where it's progressing. Just to move things along for the sake of the demo, I'm going to jump over to this previous run that I gave it. And we can see here this is the earnings tech or the earnings report. This one. We can see here I gave it this the same same prompt for earning the highest on Apple. It ran through the entire skill, did everything that it needs to do to, connect to the data, make sure it has the right context, create the use the the Microsoft doc skill to be able to create a Microsoft Word document, validated this, and then created the, earnings update that I wanted that I asked it for. So it gives me here a summary of what it is that it it discovered in this earnings analysis, in, fiscal two quarter two twenty twenty six. Apple beat expectations, revenue of $111,000,000 beat consensus. This is a summary. Then it also gives me the Microsoft Word doc, the complete earnings report that I requested. I'm going to briefly stop sharing cloud here and share that, Microsoft Word doc so we can see the output of the skill. So this is the, the document that cloud generated for me. As you can see, it's a well formed word document contains, at executive summary level, the same information that Claude directly replied to me in context of the co work. This is for Apple. The rating is still buy. Maintain that rating, giving the price target that it it determined based on the information I provided to it, even though I didn't say directly in in chat what it was, and and told me about the, the results that Apple got for that quarter. We can see as I scroll through here, this is a a full on report. We can see it it focused in their separate sections on the iPhone deep dive that I asked it for when I specified how I want what I wanted it to focus on. I said iPhone in China. So here we see the deep dive into into iPhone considerations and then also a deep dive into the China effect and what, what impact the Chinese market is having on on Apple. K. I'm going to stop sharing here again. Stop sharing and go back to sharing Claude. Share. So this is what was created from this guy. This is what what skills are doing for you. And what they're doing is providing shared capabilities that anyone in your organization can use and reuse. So I can use this same skill for any other companies in my portfolio, get the same kind of analysis. I know exactly what to expect. I know what kind of information to provide it with. And that is, this skill can be shared across the entire organization so that, as as, everyone within my group is performing these earnings analysis, it's all going to be consistent and performed in a in a very similar way. Let's let's pause here and take a look at how we actually go about creating these skills. Okay. So, we we saw in, in the request of the skill, I just prompted it and got the skill, found this so I found the skill, and, showed and used the correct skill. Let's have a look here. I have to go to customize what skills are available to me. So I can see here in the customize section what skills I have, what connectors I have, what plugins I have. So here in the skills, I can see I have a number of personal skills that I'm using. I can see some some organization wide skills and some built in skills as well that the cloud has provided for me. I can also see all of the connectors that I'm using and all the ones that I have available. So here in this section here, we see the connectors that I'm using that I've connected to. You see I have the Microsoft three sixty five connector, and, importantly, I connected to S and P Global. This is where I'm getting the real time information about Apple, in this case, that I used to generate this report. So this is giving me, these connectors are giving me the ability not just to act within these surfaces, like you saw pulling up the dock there, but also to retrieve data that I may not normally have access to. I in addition to the the skills and connectors that I have, I also have plugins. And the best way to think of a plugin is really a a packaging of skills. So I mentioned that this, earnings analysis skill that that you saw me use could obviously be useful to other people within my organization. If I package that skill up as a plug in, I can share that with others in my organization. So, really, plug ins are our packaging delivery mechanism for skills, connectors, commands, tools, whatever you need in order to accomplish these reusable tasks. This Silver Earnings plugin is actually the one where I got the earnings analysis skill that that I used and that I showed to you just there. So somebody else at Silver Earn Capital created this skill, and shared it with me so that I could I could, use it within my personal use of cloud. Let's have a look at at how exactly that was done. So I'm going to go back here. You can use Cloud co work to create skills for you. Let me quickly grab my prompt here that I want to use. So rather than retyping live in a demo, I'm going to go and paste in this prompt. So what I'm what I want to show now is how to create one of the skills like I like I just used. And the best way to do this is not to create it yourself, but to have cloud created on your behalf. So I'm going to specify here to cloud the kind of skill that I want. So I wanna create a new skill. That doc was was great and fantastic. How about creating a a PowerPoint deck that shows similar kind of information in in deck form? So I wanna create a new skill here. We call it silver earnings deck that generates a PowerPoint earnings deck and beat this analysis. I want the skill to be based on an Excel output from the beat miss report and generate a five to seven slide Silverone Capital branded earnings deck. So we want to make sure that it follows the brand guidelines for Silver and Capital so that the deck looks like something that we could produce. And the deck we're gonna generate here as output is going to be, a full on Microsoft PowerPoint deck. It's going to be fully editable. I can interact with it. I can use it as I would any other Microsoft deck. Okay. So I I trigger the, the request to cloud to create a skill. Now your cloud core comes with a a skill building skill. So in fact, what this is doing when I enter this prompt, it is, triggering Claude to go out and create and access the skill creator skill in order to create this skill on my behalf. A lot of use of the word skill there. The the, skill creator skill is something that that allows Claude to, create new things for me. So this guy is gonna create a new skill that first I can use internally myself, and later on, I can share with, with the rest of the organization so that they they too can use this this same skill. We can see here what how it's working out. Let me go to the open the skill creator skill here so we can have a look at it. I'm going to kinda squinch this to the side again. Screen space is a little bit tight. We can see here in the in the skill creator skill, what this is doing, the description, creating new skills, modifying and improve existing skills, that measure skill performance. So these are all the capabilities that this skill creator skill has. Again, this is being exposed to Claude through progressive disclosure so that the entire context, the ability to create a skill is not something that's going to be loaded into every context window that we're using with cloud. But roller, whenever it needs it, whenever it needs to create a skill, it will go out and pull this into context. So this entire description of how to go about creating a skill, its job, how to communicate with the user, the the standards that it has for for creating skills, writing the the markdown files. All of this is packaged up as a skill that Cloud will use whenever it thinks it needs to create a new skill or, in fact, improve an existing skill, which is also part of this this skill creator skill, ability. Okay. And, again, the the context management is key here. You don't wanna be pulling this entire big long description of how to create a skill into context every time, but only when it's needed. I close that down. Okay. Okay. So creating a skill is, again, going to take some amount of time. Let me once again jump over to, some work I I did this morning and show how that works. So what the what the skill is doing again, we can see the thinking that cloud is going through here as it works through the creation of this skill. It's going to explore the work workspace, read the relevant skills. It's going to because I specified, check for the silver and brand materials, so that it makes sure that the skill that it creates is going to use those materials. When it, when we package this up as a as a plug in, we can include the, the instructions to make sure that this is going to be, Silverman branded. Okay. It has all of the the, Silverman information that it needs. It's going to then write the skill as a skill markdown file and do its own passes against this to make sure that the skill is is is working and is performing the output that you would expect. So once, the skill has been created for me then, I can save the skill. This here, this is a sample output that I generated in order to test that the skill that I just created for me actually works. This is just a a sample output. What I've done is I've gone ahead and saved this silver earnings deck skill. So now when I go back to look at this, go back into customize, skills, I now have this silver earnings deck skill in my personal skills, folder. So now whenever I want to create a silver earned earnings deck, I can, in fact, use this skill to, to generate that deck for me. You can see that this is immediately available to me. So if I go back here, that's knocked something off my list. Let's do that. Silver and earnings deck skill. And what I can see here is that, all of the skills in that description, it will tell Claude what kinds of prompts in general are going to trigger the use of this skill. We can also, explicitly invoke this skill using using these slash commands. I'm going to, upload the, upload the the beat miss report that I wanted to use. Let's do, q four twenty twenty five beat miss, and then execute this deck building skill against this Excel spreadsheet. So it's gonna go ahead now and, use that skill that I just created in order to create this, this, earnings deck from the from the, data in the spreadsheet. This is Microsoft Excel, Microsoft, PowerPoint. So as we go through and and build this output, let me, chief again a little bit, and I'm going to stop sharing cloud. Okay. What's my current target? Again, it it knows that it needs a bit more context, so it's going to ask me for some additional context to this. Let me just use the defaults for now as it goes ahead to then, build the the deck that it wants. Like I said, let me stop sharing here briefly, and I'm going to share the output that I actually, let me share the input first. So let me share the the beat miss. So this is the, the spreadsheet input that I used to create the output. So we can see this is the the raw data. Stop sharing. Instead, go to the actual output that it generated. Where's my PowerPoint? Here's my PowerPoint. Share. So this is the deck that it created. So it created for me, the slide deck version of, the earnings analysis based on the the same data, that same beat miss report that I have in spreadsheet form. Okay. So we saw what skills are, how to use them. We saw how to build a skill. Let's have a quick look at what was recently announced, as GA, the ability to use cloud within your Microsoft Office environment. So we can see here within, within PowerPoint, I have the cloud plug in enabled so I can interact with cloud. So let me go back. Stop sharing this. This is more fun to start from the from the spreadsheet view. Let me go back into Excel, and you can see that, Excel, likewise, has Claude started up in here. Let me do something seemingly fairly simple. Let me actually, update iPhone revenue on row 11 for actual to be, I need to make sure I'm good enough, 50.29. Okay. So, yes, obviously, this is something that you could do just by editing the cell. What we're gonna see here is that it's not just going to edit the cell. It's actually going to look at what the implications are within the broader spreadsheet of making this change. So, yes, it wants to set the cell range. I wanna make sure that it's not doing anything under prove. So what Cloud noticed here is that if we, update the, the actual iPhone revenue from 49 to $50.5.0.2, that actually, changes the miss to a beat in terms of the expectations. So even though this, beat was not this cell here, was not a formula cell based on the other two, Claude was smart enough to know that it should update this information in order to, make sure that the the sheet was consistent. And there wasn't just one cell that impacted, but it knew, what else was relevant there. So this is the Cloud intelligence acting within your spreadsheet. How about this? How about we can update the corresponding PowerPoint to match? As we saw earlier, I used my deck building skill to, create a PowerPoint deck based on this beat miss analysis. Now that I've changed beat miss analysis here, the the PowerPoint deck is out of sync. What I can do within Claude in Excel is tell it to update the PowerPoint deck. I quickly stop sharing here and go over and have a look at what's going on then in PowerPoint. We can see that in PowerPoint, it has received the, the the cloud is cloud is the same cloud. So cloud is basically talking to itself. The cloud in Excel is talking to the cloud in PowerPoint and instructing it to update the slides appropriately. So now you're keeping your, your, SlideWare in sync with the raw data just by having cloud connect the two together and make sure that, that that that any changes you make to one can, in fact, be propagated to the other. This, again, just released the ability to have, just generally made generally available the ability to have have cloud within your Microsoft Office environment. K. I'm gonna stop sharing. This is going to keep on updating updating my deck here for me. In the interest of time, let me stop this. Let me go back and share window. I want to share. Actually, let me share a Chrome tab. Share this. Okay. So, the the other big thing announced, even the good thing announced, is the, standard, agents for FSI specifically. So this is a a suite of agents that we have released that are going to be specific for financial services. These are available for you to use, directly, or or you can use them as templates to create your own skills, your own plugins, based on on what's available here. So what's in this repo for, Ontopics Financial Services, a public repo, you can see the plugins that are available, and you can see the cookbooks, essentially instructions for, how to, use these as managed agents within your environment. So that last piece that Nick showed, showing all of this this swarm of agents running, those can be run as managed agents, on, anthropic infrastructure. So none of this replaces or makes obsolete the messaging API, the, the agent SDK, but it adds additional capabilities on top of that if you want where the, you know, infrastructure management component is all taken away. And these can run, on within within Anthropic as as managed agents. So you don't need to worry about how to set up and and define your your infrastructure and your environment for managing them. So this is all that's available. Let me stop sharing it again, go quickly back to, to share my cloud window back to cloud. Okay. Back here. Here I am. Let's look at talk a little bit about how to use those, just published FSI, agents. So, all of these are available as plug ins downloadable from, from from that Git repo. So the first thing you're gonna wanna do is create a plug in. We're gonna add a marketplace to connect myself to the the repo where all of these are stored. So within cloud, I'm going to connect to this, sync to that repo. Oops. That, is not the fully copied URL. Let me just get the full URL there with the git part of GitHub. Same to that. Now, clearly, I'm I'm doing this as a an individual user within this organization. You probably wouldn't want to do this for real in production. You probably your admins to, only have the ability to go out and connect to these different marketplaces. But just for demo purposes, I'm going to assume that I have the, authority to go and connect to this financial services marketplace for for FSI skills that, Entropic has created. Let me so okay. So I've I've connected to the marketplace. What I can do now is under personal plug ins, I can now browse the plug ins and see all of the plug ins that Entropic has made available within that within that marketplace. Go to personal, financial services. These are all of the the plugins, all of the agents that we have built and made available for you to use. Now you can you can use these. You can install these, into your into your organization for use. Again, the earnings of yours now installed and ready to use. And you probably wouldn't want every user to be able to do this, but doing this at the admin level so you can, share these, plugins, these skills, these capabilities to all of your users. That's probably something that that you will want to do. And you can, as I said, use these as is, or if you wish, use them, for, as as templates to create your own. So one of the, new plugins that I added doing this is the equity research plugin, and this has, among a list of skills. I said plugins are essentially bundles, packages of skills meant for distribution. So within this equity research plugin, there are a number of skills that are part of it. And we can see here this includes that initiate skill. So if you think back to the start of this demo, I pointed out that the earnings analysis skill explicitly said that it was not intended for for, generating initiation reports. This skill here in equity research is intended for just that. So now that I have installed this into my cloud, I can now go back and start a new task, and I can say, initiate Apple. And now this cloud is going to find that new skill. I'm going to use that skill, instead of the the standard earnings analysis skill to perform this, initiation report. Initiation report is obviously gonna be massively long and a lot longer than than the, simple, you know, 12 page earnings analysis. So I'm not gonna run run this entirely through. But you can see how you can use the the skills, the the plug ins from the marketplace to install new skills into your organization and make those immediately available for your use. K. The final little bit I want to show is the managed agent perspective of that. So I said in addition to, all of the, the plugins and co work, you can also use these as templates to, not the oh, where's your Chrome tab? You can use these to to build your own managed agents, and run them on on traffic infrastructure as managed agents, on your behalf. These cookbooks, in the same Git repo, have all of the instructions you need to create a managed agent. This is a a data analyst agent. In this example, we can see here that it is giving you full instructions. It's giving you the code that you need to execute to create and deploy this agent. And in fact, if I stop sharing here and quickly move over to my Versus code. You can see that these are also available, in as a Jupyter Notebooks that you can execute directly against and have this execute your code to, create this managed agent and deploy it and and make it available within your org. These, are entirely editable. So as I said, you can use these as is, or you can use them as templates to create your own equivalents of these, agents specific to to your organization. In this case, the managed agent is gonna perform some sales data analysis for me. The final step in doing this is to stream the run, and I'm not going to do this live right now. But as we come back and share a Chrome tab, You can see the, output of that, sales analysis skill that was generated, once I deployed this managed agent into my environment. So now, anytime that any user wants to perform this kind of sales analysis, it's going to be, again, consistent. It's going to be, managed by your organization so that it behaves in the in the same way, And it's going to be, only, called as needed, and Claude itself is intelligent enough to know when to use this capability, when to when to call this, perhaps as a as a sub agent, perhaps, as a as a discovered skill within coworker. It's all the same underlying agent, the same underlying capabilities that's being deployed across these different formats and these different workspaces. Okay. Let me stop sharing again. I believe that I have managed to keep mostly on time in terms of the the demo expectations. Team, are there are there questions that we could address? And if if anybody does have questions, please, in the, put them into the q and a section, and we'll address them as as we can. We see that there are a lot of wonderful questions. So, Nick, maybe if you could come back on stage as well. But yeah. So we'll go ahead and share the questions that were the most upvoted. There are also some answers to those, are already in the chat as well. But I'll link to the top are some really great questions around, you know, whether or not Claude would use personally identifiable information and security. Owen or Nick, would either of you want to comment on that? Also wanna share that we do have a security webinar, taking place next week, so we'll share that as another follow-up for all of the folks who are on the call today. Nick, did you have an official line from from product on on PHA? Yeah. For sure. So we integrate with all of the relevant systems that you should have in place for data loss prevention. So, any of the enterprise use cases for PII and PHI would be addressed through those integrations. And we have a full suite of documentation on our precise security architecture available in our trust center, which I believe we have linked as well. So feel free to take a look at those. Awesome. And a few questions that came in around Outlook as well as other, types of add ins across the Microsoft three sixty five, portfolio such as Teams, noted that Outlook should be landing very shortly this week, and Teams, there's no specific plans around that yet. But we also got some good questions around, cloud managed agents, what specifically are those, and when to use those instead of coworkers. So, Nick or Owen, if either of you would want to elaborate more on that as well. Yeah. For sure. I shared a link to the documentation on cloud managed agents. They're really meant to be used by developers building their own asynchronous, more autonomous agents using our API. And the cookbooks that we published a few days ago are really meant to be a starting point for you to customize and tailor to your specific processes. And this is obviously very different from our cloud.ai chat and co work, which is really meant to be used by non developers to interact directly with cloud intelligence. We need to enhance your experience for specific workflows through that is through the plugins that we've also published on our marketplace. Yeah. Exactly that, Nick. Thanks. The the the way to think of managed agents is as a a step up in capability from the messaging API and the agent SDK. So it's an additional layer on top of that that your developers would use, rather than your, your standard knowledge workers. Great. See a question here as well from Ross around, how does the model adjust for bias in its analysis? So, for example, if Owen noted in his prompt that, his rating was a buy, how do you know that the model is, controlling its response so that it's not biased towards a buy rating? Yeah. So Claudius is quite objective, on average, and we've done a lot of safety and alignment training to make sure that Claude remains quite neutral in most of its responses. I do think the way to really ground Claude in how you perceive some of these, workflows to work is to connect it to, one, the right sources of data so that Claude has the latest and the most fresh and accurate information to work with. And two is to also build a set of skills and instructions to direct Claude in specific workflows. So for example, in earnings analysis, you probably wanted to be a quite a critical, critical analyzer of the information to provide you with some some feedback and pushback when it comes to your own hypotheses. Right? So it's important for you all to think about what is the role of Claude inside my workflows and how can I get Claude to be the most thoughtful and objective thought partner in that process? Great. We also had a few questions along the themes of, do I need to have separate subscriptions to the other sort of market data providers such as FactSet, S and P, what have you, in order to access that data within cloud? Yes. So we have an open MCP connector ecosystem where you do have to leverage your existing licenses and relationships with underlying providers in order to access. So how that how this really works, and Owen showed us some of this inside of our product, is that you're going to connect the MCP servers, and each of those servers is going to redirect you to the authentication screen where you would enter your credentials of the other systems, and that's how you enable clause access to that data. Great. There's also a few different ways to interact with, you know, Claude as well as Microsoft three sixty five. So there were some questions as well around, you know, what are the differences between using Claude with, you know, Microsoft connectors versus, you know, add ins, within the Microsoft apps versus Microsoft co or Copilot, which is, you know, on the road map, for the Microsoft team as well and thinking about, like, the nuances between some of these different surfaces. So Microsoft is one of the important partners of Anthropic, and the cloud models power a number of the Copilot services as well. Within our own product ecosystem, you can think of the connectors as connectors being MCP servers as a way to directly interact with all of your Microsoft data within our chat, co work, and cloud co inter interfaces. So think of those interfaces as a single pane of glass where cloud is able to reach into the Microsoft data as well as everything else you have connected, SMP, fax, and others, all in one place. So that's obviously very powerful for for cloud to really have broader context. Now assuming that you want to go a lot deeper into one specific piece of analysis or workflow, this is why we've built cloud into those applications themselves so that anytime you have Excel, PowerPoint, Word, or Outlook open, you want to do iterations with cloud directly, you wanna create directly within those surfaces, cloud is available for you as well. And the beauty of these surfaces is that the same cloud, so the same plugins, skills, and connectors carry over to both surfaces as well. So you will feel a very seamless interaction throughout your entire Yep. Exactly that. It's largely about context. When you're using cloud within PowerPoint, Excel, Word, workflow. in in the way that I've benefited here with, you know, a a slide a slide deck or a spreadsheet or a doc open, that's the context that cloud is is fully aware of. And it also, of course, has the broader context awareness as well. You saw how I was in, in a spreadsheet in in Excel, and it was aware that there was a a connected PowerPoint. So I was able to to understand that broader context. But when you prompt it, as I did, change the value in this cell, it knows exactly what what spreadsheet you're talking about, what cell you're talking about. It it can it can find that within that context. So it's about managing and and controlling the the context available to Claude so it's better able to do his job. Great. There's also a question around, you know, are we able to share skills within just specific teams, you know, knowing that you can share skills across the whole organization but wanting to, sort of narrow that to specific individuals. Yes. You can. So within cloud dot ai and cloud co work, we have a concept called role based access control, RBAC. And RBAC essentially allows you to create user groups. And user groups can be teams, departments, marketing, you know, products, finance, sales. And within each of these user groups, you're able to also segment skills and MCP server access directly as well. So and this is really the main way for you to not only share skills, but also make sure that you find your green control in terms of the information that your your teams have access to. Okay. Great. I think there were also a few folks on the line who noticed that at different points, Owen used, SONNET versus OPUS. And so any sort of, context around when to use, various models would be great as well. Sure. And the the hierarchy of models that we have available, I I hope it is fairly well understood at this point. Opus being the longest thinking, deepest thinking, and Haiku being the the fastest, will take turnaround responses more quickly, and Sonnet being kind of a a a middle ground there between the two. So, it really depends on the the task you're trying to accomplish and the kind of output that you want. If you're looking for, you know, fast responses to relatively simple questions, I I don't wanna I don't wanna dis Haiku. It's still it's it's it's pretty smart. But that that that would be the model that you would use. If you want to do that kind of deeper analysis, the the the deeper thinking as it decides how to respond, what kinds of things it should create, what, what the relations are between what it was creating now and and everything else we created before, then then Opus is going to be, a a deeper thinking model, going to give you more reason responses. You you saw the the the trace there, the the thinking trace as I went through the demo, how it was explaining, every step that it was moving through, explaining its its use of tools, explaining why it was doing what it was doing. You're gonna get a lot more of that with with Opus. And Sonnet is kind of a a happy medium between the two. It's good for, ordinary day to day tasks. The, you know, kinda a useful rule of thumb that that I like to use is, you know, start with Sonos, see see how it how it, what results it gives you. It seems to give you the kind of results that you want. If they're, you know, if it's not fast enough, you want something faster, maybe move to Haiku. If it is something that is, you you feel that you could have used a deeper analysis on this question you asked us, then try moving up to Opus and and seeing what results you get there. Great. So I think that puts us at, just about the end of the time. But we did want to share a few of the resources that are available to all of you. We'll also be sharing all of these via email as well. And we just wanted to thank all of you for taking the time to join us today. Again, please give us feedback. We love to continue to hear how we can make these sessions more valuable for you. And, let us know, you know, how we can continue to share more resources to help you as well, as you embark on your journey to take advantage of more and more of these capabilities. So with that, our team just wanted to say a big thank you, and hope you all have a wonderful rest of your day. Take care.