Name: Deploying Cowork across the Enterprise — with PayPal
Uploaded: 2026-04-16T18:32:11.297Z
Duration: 58 min 8 s
Description: Deploying Cowork across the Enterprise — with PayPal

Transcript for "Deploying Cowork across the Enterprise — with PayPal": Hello, everyone, and, welcome to our webinar. So really excited to walk you all through how you can deploy Cowork across the enterprise today. We're gonna get started and give everyone, you know, maybe another thirty seconds or a minute to pile in, and then we will kick off. Super excited for the agenda that we have today. Cool. I think based on the number of people that we have in, we should just get started. So welcome, everyone. Just as a little bit of notes of housekeeping, so as everyone always usually asks, a recording of this session will be distributed via email within twenty four hours. There is a Q&A panel. It should be, at least on my screen. It's in the top right corner. Please put all of your questions in the Q&A. We will be going through them, and we will have time at the end of the webinar, after our lovely fireside chat, to answer some of them. And then, you know, any questions that are submitted, hopefully, you know, potentially, we can get you a response if we don't get to them. And then last but not least, we would love to have your feedback. So please rate this webinar, when it's done with the survey. We do take a look at that very seriously, and we wanna make sure that you're getting the most out of it. Awesome. So in terms of the agenda that we'll walk through today, we'll walk through some introductions. My name is John Lopus. As you can see on the screen, I'll do a more formal introduction in a second with some of our other speakers. And we're gonna walk through some new feature updates. Based on our role based access control, a lot of other kind of enterprise features, I'm gonna do a really brief demo as well of the actual Claude desktop application, walk through some of those features. Then we have a conversation with PayPal. We're joined by Jarred Keneally, who's a wonderful customer of ours, and then we have Q&A as well. That being said, before we get started, I think we did have a poll question about, you know, where everyone is in their journey. So I think that should pop up on the screen in a moment. Great. So we just asked a poll. So where are you in your Claude cohort journey? Are you just exploring? Are you piloting with a small group, you know, expanding to multiple teams? We're just really curious where all of you are. So I'll let you all take that poll real quick. Awesome. So I think most people have now, replied. Okay. Cool. We got a bunch of different options here. So seems like the plurality are either just exploring or piloting. So for those that are just exploring, welcome. And then we do have about 20% of us that are expanding across multiple teams and rolling out more broadly. So I think that everyone, no matter where you are in your journey, will get something out of this. So that is fun, and thank you all for participating. A lot of folks in here. Awesome. So without further ado, just wanted to quickly introduce the folks that are on the call. So my name is John Lopus. I am one of the applied AI leads here at Anthropic. I focus on a lot of our strategic enterprise customers, and I'm super excited to be here. And then I'm also joined by Amber Yin, who you'll see on the screen as well. Amber Yin is one of our customer success leads that works on a lot of strategic enterprise enterprise deployments, and she'll be having a fireside chat with Jarred Keneally. Jared is a senior director of product management for AI technology at PayPal, and about thirty minutes of this webinar is just gonna be about their experience of using Cowork and working with Anthropic. Awesome. So to get to the actual, you know, meat and potatoes, we have recently launched a lot of cool stuff with Cowork. So first and foremost, Cowork is now generally available on all paid plans working with Anthropic across Mac OS and Windows. It is through the desktop application, which I'll I get a lot of questions about at times as well. Next, we also rolled out our role based access control. So this role based access control has group limits. It has spend limits. It has per tool connector permissions, and we'll do a deep dive on this as well. Next, we also have rolled out a lot of analytics about how users in each of your enterprises are using Cowork itself. So what are the sessions like? How many active users do you have? What skills are being used? And all of this is getting pushed into an admin dashboard as well as an API to to be pushed into your own services. And I really wanna emphasize how important these analytics metrics are because it gives you all really good insight into understanding how, folks are using specific skills, maybe what skills you need to build, and it can really help you identify power users. And then last but not least, a little more of a technical consideration is our open telemetry enhancements and improvements. So this really increases the observability that you have within Cowork, whether it be tool calls, file changes, approvals, everything and anything around, I would say, observability as well as monitoring and just compliance. So, you know, that one gets all the Infosec people in the room jazzed. Awesome. So let's move into role based access control. Great. So I think role based access control, is fairly par for the course across enterprise applications, but, you know, we're trying to make this as up to date and modern as possible for AI tooling. So the main thing that you can do with our role based access control is create groups with differentiated limits that each get the right access. And all of this role based access control is also directly tied into your SSO, SCIM, IDP system that you have. So whether you're using Google or Okta or Entra, which many enterprises are, this is all connected directly into that role. So you also have the ability to set monthly spend caps at the group level, and you don't have to go user by user, which can be a really big headache when you're deploying this across the enterprise. So, again, I think that the main benefits here are group level automation, and then there is flexibility on the individual. So if you have some power users that are, you know, churning out documents and really going above and beyond, you can give them special, limits and privileges. And then last but not least, this is all through a single pane of glass in the admin UI so that you have a one stop shop to make these changes. There are three main concepts in our role based access control. The first being groups. So this is really tied into, you know, your organization's team or departments. Who are you? What team are you on? Are you a marketer? Are you an operations? Are you a product manager? And then a role, this is really a bundle of permissions that defines what actions they can take. So you can kind of have a many to many relationship between groups and roles. So you could be a marketer and your role is, you know, just ability to use Cowork. You can't access Claude Code. You can't do x, y, or z. You also have the ability to be, you know, a product manager, and maybe they can access Cowork plus Claude Code. Right? So you have a lot of different permutations there. And then last but not least is capabilities. Like, what can Claude do? Can it invoke skills? Can it read or write? You know, can it access certain models? I think that's a big thing that a lot of people are looking at right now, especially from a cost expenditure point of view. You may only want certain teams to be able to access Opus, which just a little plug, we did release Opus 4.7 earlier today. So if you're a user Anthropic, please go check that out and give us feedback. Great. So the initial set of capabilities, that you can see, and I won't just kind of read them all off, is you can filter on if users can use Claude Code, if we can use Claude Code in fast mode. Fast mode is a super important one to get access to because it does use significantly more input and output tokens. Claude Code security, which is one of our cybersecurity adjacent products where we're actually doing code review for vulnerabilities within your system. Cowork, memory, web search. Web search is a really big one that we see. Maybe you only want your users to be accessing and referencing documents that are internal as opposed to looking things up on the web. Next, we'll talk about briefly how group spend controls work. So, again, there are individual user limits. Those can also be bulk assigned to groups. And users that are go between multiple groups, we look at the minimum. Last but not least, you also have the ability to set individual limits that take over the precedence, and we'll show all of this in the demo. Cool. So without further ado, let me share my screen really quickly. Awesome. And you all should be seeing my screen. It should just be a Claude window. Hopefully, you all are relatively familiar with this. One thing I did wanna quickly show out, we do have this as a a preview. A lot of people give us feedback all the time that essentially says, hey. You're Anthropic. Why can't I use Claude to set up Claude? We heard you, and we have this. So if you wanted to, you could say, you know, show me your groups, and that could run. We'll just let that run-in the background. So in this case, I just wanna go in and look at our members. If you look at our members, you can see we have groups and roles here. And in this, you have the ability to look at any and everyone, that is a part of your organization. When I click in on groups, you can see here I've set up a couple different ones. We do have a handy tooltip up here that has a setup guide. But if I wanted to look in, you could see here the sales team only has access to Cowork or they have full access. When you create a group, you do need to assign them a role. So in this case, if I go in, you can see product managers have full access, whereas marketing and sales or Coworker only is kind of a fairly, standard thing that we see. Within custom roles, this is where you specifically can set the capabilities. So in this case, you'll see that Cowork only has the ability to, you know, turn on code execution and file creation, look at memory and web search. But then you also have the ability to turn on and off access to products. So when I go in here, you'll also see access if I can turn on Cowork. and here's where you would turn on your open telemetry. If you are a nontechnical administrator of Claude, I would recommend you work with your customer success manager or the applied AI team, which is a team that I'm on to help get this set up. It's relatively straightforward, but this is how you're gonna get access to all of that observability and all of that monitoring. And then the same thing would happen here if we go back to the members in terms of actual usage and billing. So when I come in here and I click on usage, you're gonna see I have all of my users, including myself, how much they've spent, and then you can specifically set limits. So in this case, I have the sales team, and they're spending a thousand dollars a month, whereas everyone who's using chat plus Claude Claude code is, limited unlimited. So if I click here, you can say, I want the product managers to set a dollar amount. We'll say, you know, I'm feeling generous today, so the product managers get $2,000. Great. So now anyone who is a part of this group that we've established before now automatically gets that $2,000 limit. However and this will wrap up the demo. If I wanted specifically to, you know, give myself a little bit of a treat and set a $5,000 limit, I could do so. And now John's limit overpowers the fact that he's a product manager. Actually, gonna set it unlimited in case I I wanna go crazy later. That's mostly what we wanted to show in terms of the demo. So I'm gonna go back to the slides. And then from there, I'm gonna hand it over to Amber and Jarred for a fireside chat. But before I do that, I do wanna quickly talk about our enterprise control, and apologies. I thought we already were on the, fireside chat piece. So just kind of reiterating what we talked about with the open telemetry, it's really built around observability. So you can get the user prompts, MCP and tool use as well. And then I think one of the more important pieces is analytics. When you're justifying an enterprise rollout, you really need to think about utilization. So how many daily active users do we have? Weekly active users, monthly? What is the per user activity? What models and features are they're doing? And I think that this really helps build an internal business case for Claude itself. So when I think about and when I work with a lot of customers, you know, everyone wants all these AI tools. Everyone wants to use Claude, at least in my experience. But at the end of the day, you gotta justify that internally to your teams, and this really helps you do that. Last but not least is those admin controls. So this is really kind of what we showed in the demo there is we really wanna make it as granular as possible on what users can access. Because as we roll into this kind of new agentic future, it's not only what users are doing, but it's what the agent is doing on that user's behalf. And you gotta be able to track that for compliance purposes as well as just kind of general governance. Awesome. And last but not least, I just wanted to give a couple high levels on how we think about deploying Claude across the organization. So, really, when we think about the path of the first win, we find that it's really helpful to pick your champions. So champions is often kind of a a sales lingo or terminology, but I really think it's appropriate here. So if you are in a large organization, think about 10 to 15 people in different functions. One of your growth marketing team or your outbound and enterprise sales team and find one workflow that's costing them a lot of time. Maybe it's deck generation. Maybe it's, you know, a finance team that has to do, p and l every single month. Then find out and build playbooks or skills that take that off their plate. When you layer in your team's skills and your connectors, they're gonna become hooked, And then you can figure out where those champions get traction, what's working, what's not, and just iterate that across the other functions that you have so that all the marketing teams are using it or all the finance teams are using it. Awesome. I think, you know, I'm getting the light when I'm on stage, so it seems like, my time is up. So I'm gonna hand it over to Amber and Jarred. I'll still be around, and I'll be around for the Q&A as well. And, yeah, I'll hand it over to Amber and Jarred now. Awesome. Thank you so much, John, and thanks everyone for joining today. As John mentioned, I'm part of the customer success team here at Anthropic. That means I get to partner with enterprise customers like PayPal on bringing Claude to life. Today, I'm super excited to have Jarred with me, part of the PayPal team who leads AI technology. Jarred, welcome. Welcome. Welcome. Thanks, Amber. It's great to be here. Good to see you again. Good to see you too. Alright. Well, to to kick off the conversation, you know, today, we're gonna talk a lot about your experience, helping to roll out Cowork to your enterprise. We'd love to just kick off and and learn a little bit, you know, share with the group a little bit more around your role and and how you think about AI at PayPal. Sure. So right now, I'm the senior director of product for the AI technology team at PayPal. And so what our team owns is the entire AL and ML platform that all of our PayPal engineers build on. So that includes the agentic infrastructure, all of our Gen AI tools, but it also includes our governance, and then any of the tooling that helps, you know, manage everything that the engineers and knowledge workers are actually using, so they can safely use these capabilities in their day to day work. So when I'm talking about AI AI at PayPal, I'm not really talking about it from the outside. We're the team that's actually building it. Absolutely. So you are certainly an expert in the AI space. You are no stranger to, you know, what it takes to roll this out at scale for an organization. Tell us a little bit more around just, like, your journey. You know, you you've been working in the space, for much longer than maybe most people, you know, have. So tell us a little bit more around, you know, how you've been thinking about this from the start. Sure. So let's go back about three years or so, and we first rolled out AI coding tools and AI coding assist tools for all of our developers. And I'd say that's kind of the early chapter. It's pretty useful, but it was pretty contained to just code generation. And, we've evolved our our strategy since then. But about a year ago, maybe nine months ago now, PayPal actually had its very first, PayPal wide internal AI summit. And this was for all employees, and Anthropic was there. And it was when we launched Claude enterprise across the entire organization. And that was the probably, the the the first time where, you know, we weren't really talking about just engineering. Right? So what really stood out, we we now have leaders, HR, finance, legal. They're standing on stage in front of everybody at PayPal, and they're talking about how AI is changing how they're working. Right? And so, you know, that that was a moment for me. So, like, I knew, you know, in order to to transform PayPal at scale with AI, it was not gonna be a mandate. Like, absolutely, it was not going to be a mandate. You know? And it and it really starts with the people across your organization that figure out which tools and and really understand how to use those tools to make their their jobs better. And quite honestly, Claude Enterprise gave us the foundation to do that, and at PayPal scale as well. I mean, with the the security controls, the data controls, you know, PayPal is a regulated financial institution. That's that that's absolutely things that we need. Yeah. No. I was gonna say I I I remember I can't believe it's actually only been about eight to nine months, Good. since we did that for rollout. It feels like it's been longer, but wow has, you know, wow, have you guys built a lot of great things? And, I know today, we're talking specifically about Cowork. which is kind of the taking that to the next level. So, yeah, why don't we focus the conversation now a little bit around, you know, what kinda stood out? You know, eight months later, suddenly, we've got this incredible new product in research preview cohort. What what stood out to you? What kind of made this different than just chat, for for knowledge workers? So I think for me, you know, we we had rolled out Claude code to all of our engineers, and, you know, there's a there's a cult following, I would say, of from the engineering side of things. And, you know, what stood out about Cowork was immediately, you know, we could bring that same type of capability to the rest of PayPal. Right? So whether you're a knowledge worker, whether you're on the legal team, or you're, you know, in finance, we could bring that same capability, to all of them, and we wouldn't have to sacrifice our our governance. We wouldn't have to sacrifice our, you know, security. And, you know, as a financial institution, we had to maintain all of that. So that's not a small problem to solve, to be honest. And we're in a highly regulated space, and so none of this, you know, is really is nice to have. So, the the the ability to roll this out with broad adoption, and be able to actually deliver on that governance was was one of the big things that stood out to me immediately, and I said we we really need to move forward with this. Yeah. What what, I know that you guys are always kind of on the on the leading edge of, you know, anytime we have new capabilities, testing it, being a part of those, early access opportunities. In this case, what about Cowork kinda stood out and made it feel like it was different than other AI technology? So I think I mean, I think I think Cowork just felt like a an unlock. Right? And the it it wasn't just about chat. Like, when you're when you're using chat, you're kind of imagining, like, what it is you want the tool to do. I think if if you've used Cowork, you immediately can see, you know, the plugins and you can see the use cases and workflows for those specific functions, whether it's legal or HR or finance and so on. So, you know, if you're an HR business partner drafting a policy, you know, that was in a spreadsheet somewhere or that that workflow is in someone's head. But with Cowork. you can now structure that, and then you can accelerate it, and you can hold everybody to a a consistent standard in the way that they do that type of work. So, you know, same principle for all of those different functions and just a different population, but huge addressable surface area. So that's that's a bit of, you know, what really felt different, you know, from this. It just didn't require you to imagine the use case and, you know, typically, the use cases are like, I wanna generate a PowerPoint, which is fine. But, I mean, really, the transformation comes when, you know, you can use these tools to completely become more efficient and become more effective, and you can standardize the way that people are doing it across the organization with the controls that Claude provides to us. Totally. So, like, appealing to definitely the enterprise, kind of security requirements that you had, but then also we were seeing a lot of demand early on, like, when we had first rolled out with the early access program. There's also just a ton of folks out. Just wanted to be a part of it, starting to spread word-of-mouth. Yeah. I I think that's that is absolutely one of the other things that was definitely different about this. I mean, typically, you know, I work in the central technology organization, so everybody who's asking about things are within that org. They're data scientists and engineers. But this this was a bit different. This is employees from every single different business unit, all reaching out and and asking for access. And, you know, we didn't do a lot of announcing on on the early access program, but they found out pretty quickly that it was happening. And so I I heard from them all. So the demand was was pretty significant. So I think, you know, you got something here if everybody is, you know, requesting to get this, and we haven't even come close to rolling it out yet. So I think that's a good signal, that we want. I would say it's a equal parallel to Claude Code. So, just and drop it. You guys have just done a great job on building tools that people absolutely want to use, and it's it's not just a a typical tool. So, that's what I saw immediately. Totally. So we were kinda getting a lot of demand across so many different parts of the org. And naturally, I'm I'm imagining you started to think about, okay. Do we really have the right controls in place? Do we really have all the things needed to make sure that, deploying this at scale was gonna go well. So tell me a little bit more about, like, what was your thought process around how you were going to actually accomplish this? Absolutely. So I I think about it in kind of different layers. But first, we we wanted to test it out with early access, so we had managed controls over all of that. The second is we really had to understand the usage. So all of the capabilities you're providing with hotel and with the APIs really gave us complete visibility into how teams are using it, how they were adopting it. And then, you know, as we thought about scaling this, what we really wanted to do is make sure we enable champions. So we we picked champions from different business units. Like, I'll I'll tell you, like, our AI team absolutely loves working with me, and they have a monthly AI in action, team. And they do different talks, and I I presented what was going on with Cowork, and they loved it. And, you know, they they absolutely are a huge fan of Claude. And, you know, within, you know, a couple hours, they were already integrating their own legal playbooks into the Claude Cowork plug in for for legal. So they were they were demanding access, but they already had champions who were there to to figure, you know, figure out how they were gonna roll this out to the rest of the organization. And, you know, when you think about, you know, the layers that you have to enable for for the scale. First, you know, who can access it? Right? And we have to absolutely control who's accessing it. So only PayPal accounts can access it. We can't allow, you know, you to log in on a PayPal device with your personal account. We can't allow work alone, you know, for any anyway, it has to be managed with single sign on and so on. So, you know, you can't control you can't control what you can't see, and so we absolutely have to see who's using it. You know, the second layer of all this is, you know, what can leave the building, and so we have to have a a layer of data loss prevention. You know, all of our capabilities that are connected MCP services are zero trust. They rely on OAuth. You know, basically, if you had access to it before, you can have access to it here, enforcing all of our data policies, data access policies, and so on. And so that's just all built in. So the the amazing thing here is that this is a horizontal layer, so the users, you know, don't really notice it when everything's working, and that's kind of the the actual point of all this. The I love the the anecdote is Hippo has compliance training that you have to take, and the fun experiment I always say is go take a screenshot of your compliance training and give it to Claude and see what happens. And Claude immediately recognizes that it's, well, our our controls recognize that it's our training and will immediately tell you they can't do it for you. You have to go take it yourself. So that's a perfect example of, you know, all the guardrails are in place, but they're not getting in the way of of what teams are doing. And then, you know, the the last one is, like, how you're connecting Cowork to your enterprise tools. Right? If you if you can't connect to, you know, Office, if you can't connect to Jira and Confluence, then you're not really gonna be able to to realize the the full power of enabling everybody at your con your company. But, that's one of the things that's part of our strategy. So we we have, you know, only certified MCP services to our enterprise capabilities and to our enterprise tools. Those are the ones that are connected in Claude and nothing else. We we register every MCP service, and they're certified by our team and agents. And we're we're probably close to a 100 MCP services now for for our enterprise tools. But that is also, you know, the key to unlocking your your workforce. And, of course, Claude can only act with the permissions you already have in in this model. So, you know, you you can't do it. If you couldn't do it before, you're not gonna be able to do it with Claude in these particular cases. So that's one of the big critical pieces of it. And, you know, quite honestly, we couldn't have done the Cowork rollout without the new role based controls that you're that you've all turned on. So I now have the ability to control Cowork from enterprise chat, you know, granularly. I also have the ability to, you know, control the tiers and the and the usage as well, which was which was not possible, you know, a month ago. So if I wanna turn on plugins or features just for legal, I can do that. But what we've also done here is we've created, you know, custom roles that are buckets of spend. And we've created the the ability to request more, but we really we really wanna be able to manage, you know, the controls. If you're, you know, using Claude and you're seeing amazing results and you get manager approval or proper approval, we have a self-service way that we can you can request the next tier. So now nobody gets blocked, and, you know, you can continue to do what you want, but at least we have some control over, you know, how much spend you have. And then at the same time, we're we're teaching them, you know, ways to be effective and efficient with with the with what they're using. So, essentially, every employee can still have access, but we're managing the cost responsibly. And we're giving, you know, broad access. Yeah. So that's really the whole point of having these granular controls, making making it more efficient, more effective, giving access to everybody, but still being a bit fiscal responsible with it. Naturally. Right? When you have an organization as as large as you do, you kinda have to balance being able to, you know, deploy this incredible technology for for folks to be able to leverage in in in doing better work. But then also being mindful of, you know, how do you scope it so that, you know, spend doesn't go wild, and you're also making sure that you're able to scope the capabilities towards, kind of the different groups that need to use the different connectors or, whatever it may be. Yeah. Absolutely. And with the new admin tools and the analytics API updates, you know, we can see how Cowork is being used across the organization. So we know what the adoption is. We know what the usage is. We can see the skills, and, you know, we we bring that in with all of our other, metrics and signals as well, and that gives us a really good view and visibility into how our organization is using all the different tools that we have. So, you know, it's it's really not just about being compliant. If you know, you actually have to prove that you're compliant, so you have to be able to show visibility. You know? If you need to go to audit, you need to be able to go, you know, show how you're doing that as well. So Love it. Yeah. I I think that is it's such a key dependency. I'm sure for a lot of folks that are in the room today, you know, just making sure that, kind of the controls are in place before you you share this more broadly. But now that you have some of these pieces in place, I know that there's a lot more teams that have had the opportunity to start to use Cowork more broadly. We'd love to hear a little bit more around, you know, what are you seeing, as kind of, like, some of the the teams that are using Cowork? What have you heard from groups across the business? Yeah. Absolutely. And I'll cover two use cases, legal and and product management. But to be clear, this is available for everybody in PayPal, and they absolutely are clamoring to use it, and they're all doing things. But I'll I'll speak to these two just in in the interest of time. So, they're different teams. They have different workflows. They have different compliance requirements. But no matter what, Claude Cowork is bringing efficiency and consistency, and I think that matters. So efficiency alone, fine. You know, you're you're more productive. But at PayPal scale, you know, the consistency is is really what makes it scale. So I'll give you a couple examples. So our legal teams, they handle a significant volume of of contract reviews, whether it's NDAs, you know, vendor onboarding, and and they split some of that work between, you know, internal counsel and outside counsel. So it's not that they're they're good. It's they're it's not that they're, you know, can't get to everything, but the problem is is standardization. And so when you have multiple parties reviewing contracts or you have external or you have internal or you have nontechnical people, you can potentially wind up with different interpretations. Right? So what Cowork allowed us to do is embed our own legal playbook playbook directly into the workflow. So if it's an external council or internal teams, you run the same playbook, you run the same workflow, and everything is held against the same bar. And so, you know, instead of taking, you know, hours to do a contract review, now, you know, it takes just a few minutes for for that to happen. And, one, you can control who has access to it and make sure it's just a legal team. And, two, you have the ability to standardize, you know, how that is done and so that you can be feel that it and and confident that it's consistent every single time. So our one of our attorneys is just absolutely amazed at this, and he immediately started sending me, you know, all the different use cases. And so we've been working one by one through them all, and he's like, I handwrite all of these notes, and I have to create a formal document every single month of all my notes. He's like, can you help me build a workflow in Claude? And so we did that, and, you know, he was just amazed at how much time that is saving him. So it's been amazing for for the legal team. So PMs. So a little bit different kinda complexity. Right? You wanna synthesize customer feedback. You wanna do some competitive research. You need to write requirements. I don't know about everybody else, but at PayPal, we have a policy on what the PRD covers, what's in it, and so on, and needs to be tracked and managed throughout the entirety of the the life cycle. So, you know, that now can be easily standardized. Right? And you don't slow anybody down. So PMs might have been, you know, doing customer research one way or they might have been doing, you know, go to market planning a different way. And at that scale, we really need that consistency, and we need to that agreement to a policy. And so we have an SDLC policy, and now we can, you know, create these workflows, and we can align to that policy and ensure that consistency. So Cowork, you know, raised the bar again with, you know, all of these things, and it integrates with all the tools that our product managers use. So connected to Confluence, connected to Jira, connected to everything that you would need internally. And so now, it just makes you faster at your job, and the the bar is consistent. And, you know, you you fall into, you know, you fall into being consistent, you fall into being, compliant, and you don't have to really worry about it anymore. So, all the PMs on our team and a lot of the the PMs across the org, they don't feel constrained by any of this. They're they're just doing things a lot faster now. So, they just know it's helping them, and it's it's keeping them in mind. So I think, you know, with those two, you know, the efficiency and the standardization across all that is is really what the impact has been. And and even though they're completely different teams and and different workflows, it's it's solving for both of them. Absolutely. I imagine, the the workflows are totally different, and yet, you know, you can start to see that Cowork has helped to unlock some of these pieces for, you know, anywhere from finance to to also legal to also, you know, product teams. I know you've shared a number of other groups across PayPal. We also started to kind of, like, build out their own skills, and it sounds like, you know, your team has been doing a lot of great work also kind of standardizing now into, like, plugins, dedicated to different functions. And so, it's it's essentially just taking the work that folks are already doing, being able to standardize it, and making it more consistent, but then also, of course, like, making sure it's being done so in a safe and secure manner. Absolutely. I personally, I I do my monthly OKR reviews with a scheduled Cowork task now, and it's amazing. Yes. Scheduled task has been a game changer too. Suddenly, you can just deploy these agents and, you know, not have to think about it. So it's also been a game changer for for for me as well. Cool. Well, I wanna make sure we also have a little bit of time at the end, just to open the floor for some questions for the group. But, thanks for sharing, Jarred, a little bit around kind of the the journey so far. You know, you talked to us about, you know, how PayPal thinks about AI, how you've thoughtfully, really rolled this out at scale, how you've been thinking about, you know, leveraging some of these different enterprise, features that we've had since we've gone GA. You've also shared a little bit around kind of the early experience, across, you know, the impact that you're seeing with Cowork across business. To close things out before we switch over to Q&A, would love to just, you know, hear a little bit more around what advice you might give to other enterprise teams or other people, that are kinda just exploring. It sounds like the group today is, you know, either early in their journey or considering what it would look like to roll out cohort to their enterprise. Would love to hear from you on, you know, if there's any advice that you might give to to folks on this call. Yeah. Absolutely. The two most important things, and I I can't stress this enough, but these are the two things that you really should think about before rolling this out. And I think first is starting with the early access model. So we absolutely did not mandate adoption. So that was a conscious choice. We we always agree we would let people vote with their feet if it worked for them. And what we found was people are absolutely excited to try it out. And it didn't matter whether it was a HR or legal or engineering. As soon as we put it in their hands and, you know, help them solve real real workflows and get real feedback, it started to solve real problems. So this was controlled, and it wasn't a you know, we weren't sanitizing the use cases. We were letting them do whatever they want to, but they build out the proof points early on. And that just made things more credible when you rolled this out. So, you know, somebody's skeptical about this, you can go directly to that team, and you can actually see what they're doing and see the time they're saving. Right? And I think that's a very different conversation than, you know, somebody seeing a vendor demo or reading all of the the news about it. So find your excited people first. Let them let them build the case for you. So that was absolutely critical, and, you know, Amber and team were were amazing. They helped us with that. They're they're embedded with us. They're in our Slack channels, and they're always available for, you know, support and and for discussing how things work. So they've been a great partner. I think the other one is a little bit harder to codify, but I think it's the the mental model. And, you know, most people came into Cowork and they didn't know exactly what it is, and they were kind of like, is this just, you know, another chat? Is this such another chatbot? And, you know, if you're coming to it thinking, like, I'm just gonna ask it a question and it's gonna help me answer it, you you know, you're not gonna quite understand what Cowork is capable of. It's definitely something different. So when you think about this mental model, you have to realize that it has access to your workflow. You wanna give it access to your tools, you know, to your documents. Right? And you're not chatting with it. You're delegating to it. So think of it more as like a coworker working alongside of you. It's not a search engine. You're not asking you questions. Right? And so your mindset shift would really be from, I'm not asking a question to a chatbot. I'm actually delegating work to a coworker. And then when you think about it that way, I think you'll really start to think about unlocking that value. Right? And and sometimes it's not obvious until you see it, but you absolutely will will start to feel like, well, I can I can now delegate pretty much all of my OKR reviews to Cowork. I don't have to have my team do it or I don't have to do it anymore. So, those are my, you know, big things here. Just, start with, you know, people who are generally excited about it. You know, it's it's not about the, you know, the mandates. It's it's that's definitely not gonna work. Start with people who are generally excited. Right? And then, you know, start to think about that mind shift mindset shift of how people can start to use Cowork that's very different than probably their typical chat experience. So, I think those are, you know, the two critical things that I would pass on to everybody. Super helpful for the group. I think just, you know, anytime you're thinking about, you know, using AI, even when, you know, LLMs were first introduced, it is just a total mindset shift. And then now Cowork is taking us to the next level, and I love the way that you describe it of, like, you know, think about the work that you might delegate because that's ultimately what Claude Cowork is doing here is it's really taking apart work and and doing very complex, multistep things. So I love the way that you're thinking about this, and I think getting it into the hands of the users and the people that ultimately, will be using it is the best way to kind of, you know, solidify what will work and and kinda scale that out. Love that. Cool. Thanks for for sharing your words of wisdom. I see right now I've got a lot of questions, Absolutely. in my question bank right now. People have a lot of questions for you, Jarred. So, if you're ready, I'm gonna pivot over to a little bit of Q&A from the group. Let's do it. Okay. Amazing. Alright. So first question I have is, around more okay. Curious to hear more on how PayPal implemented this technology safely. What guardrails or security tools were put in place when implementing? Sure. So for starters, PayPal has a a secure AI, responsible AI committee, and so all new tools go through that. So for starters, compliance, legal, privacy, everybody reviews the new tools that come in. So we're well aware of that. And when we decide on whether or not we will enable it or not, you know, we need the kind of approvals of that. And so that's always changing. Sometimes it's onerous, but we make sure, you know, we're covering our bases when it comes to all of that angle. And what we didn't wanna do is turn this into a tool by tool discussion. What we really wanted to do is look at what horizontal capabilities can we put in place. So I mentioned it briefly, and I I won't go into the specific meaning of the technology. But when you are you know, we have a a data loss prevention layer and kinda like monitoring prompt layer. So, you know, we have to make sure that if you're uploading files, they're getting scanned, and we classify all the data that's in there. And we're not going to allow anything other than our class three, class four, class five data, no PII data, no sensitive data. None of that can be uploaded. And same with the MCV connections. It'll it'll block access to any data that's been labeled there. So right off the bat, you know, we're implementing our policy. We don't want people, you know, trying to use it to make hiring decisions. We don't want it to use use to make promotion decisions. So all of that type of, filtering and kind of monitoring happens as a as a horizontal capability, and we can provide those solutions that are approved. So this layer of of security kinda gives us the ability to, you know, stop data loss to make sure, you know, data that's being uploaded is is proper and is it's within the right classifications. And then if you wanna do something, you know, like use it for, you know, some more, you know, restricted use cases, then we have a more of an approved tool. Like, we don't want everybody, you know, using it to do promote decisions. Right? So we have a guy that helps you with that that's compliant. So we can redirect everybody to use that instead of letting everybody, you know, on demand go make those types of decisions. So we have that layer that's built in. So it doesn't matter what tool we're talking about, and it doesn't matter whether you're using, you know, Claude Code in the CLI or whether you're using Claude enterprise. It's that same layer that's scanning and and monitoring everything for that type of usage and so on. So that's that's ultimately how we did that. So it requires, you know, enabling that across the board. And it also took a lot of work with Anthropic as well. So to be able to be able to enable single sign on into the mobile apps and the desktop apps and so on, you know, there there was work that needed to be done so that we could make sure that we could do that and also so we could prevent our users on PayPal devices from logging into their personal accounts as well. And that's how you prevent data loss, and that's how you prevent, you know, data getting off of machines and so on. So we needed to work closely with with Anthropic to to make all of that happen. So those are a couple examples of that, that we needed to continue to monitor, and and that's how we were able to do that. And, you know, we were able to manage and monitor and change our policies and controls, you know, over time as as needed. So Totally makes sense. So very, very thoughtful approach around, you know, all the different levels of, you know, controls and and and just making sure that those standards are across, you know, every surface that folks are touching at PayPal. Alright. The next question I have for you is, can you speak to more about how you standardize workflows or prevent hundreds of ways to do the same thing? What did you leverage with enterprise controls, and how do you educate others on what exists that they can leverage? Yeah. And so to to be completely honest, like, we're still figuring that out, and we only we've only just launched in the last, few days as well. But in, you know, in the early cohorts and with the new RBAC controls, you're actually able to have the plugins that are in Cowork, and you can assign them to roles. And so as as an example, you can create a group that's just for legal, and you can make sure that they can you either get that installed or it's automatically installed or they can install it. And then if you wanna modify those, we can manage them in source control in GitHub. So we help them with that, you know, so they control what those skill sets are and what those skills are. We can push them into that plug in, or we can make them available to the enterprise. And, you know, that's how, you know, we push out those different skills for those particular workflows and use cases. Now can you force that to happen? Not entirely. So it does take a little bit of training. You know, people can, you know, do things slightly differently. But you can see who's using the skill. You can see the types of usage. You can see, you know, what MCP services they're connecting to as well. So we do have that level of visibility, you know, across the board. So, you know, it's it's easier to give them a way to do it and let them let them be compliant than it is for them to make up a way to do it otherwise, which, you know, takes more time and more work and ultimately will not be compliant. So that's how we're we're currently managing it at the moment. So it's not a perfect system, but that's how, you know, these champion teams are working right now. They're they're the ones that are implementing on the playbooks, and they're the ones that are, you know, championing to the rest of the organization how these things get done. So, we'll we'll see how that goes, and we'll, you know, continue to improve that over time. Totally makes sense. And I know that we're still you know, we we've done done this, deployment. There's still, you know, more more to be done around, like, enablement and and best practices. And as we start to build out more and more skills and plug ins, ensuring that, all those best practices are in place. The next question I have for you is just around kind of a look back. Thinking back to just, you know, how how you all approach this, is there anything that you wish you had known going into this that, you know, may maybe you realized and learned after the fact? That's a good one. I would say I mean, honestly, the the the biggest piece that I wish we had, I wish we had the the RBAC controls and and kind of the different model earlier on. And and I know that's not necessarily us, so it's it's kind of that as well. But I think, I call it Claude Gate. But when Anthropic was was in the news and when our new CEO joined, the the demand for access was just through the roof and, you know, we were hitting our license caps. So we were rapidly trying to move to the usage based models, implement the controls, put everybody into the tiered usage buckets, and so on. I I know I can't go back in time and do that again, but that would have made things so much easier. So not having to manage licenses and not have to, you know, reclaim licenses and do all that would have made things so much easier, from day one. So, I know not a great answer, but that start with that. That would have been that that would have made things so much easier to roll out at scale for sure. Totally. Well, the good news for everyone here is that now we do have, Claude, Claude Cowork in GA. We have our back controls. So, hey, Val. You guys were just ahead of the curve and and, you know, beta testing with us and and kind of experimenting before we had launched widely. But really, you know, excited now that we have this available, and and I hear you on that being kind of, like, a critical dependency for you to feel confident in in scaling this out more widely. Absolutely. We we experience the pains of the rest of you don't have to. Perfect. Alright. Well, I know that. Thank you so much, Jarred, for for kinda just taking, us through your experience, PayPal's journey through launching Cowork. I know that the group has just additional questions just around kind of general capabilities for John, so I'm gonna hand it back over to him. But thanks so much for your time. Thanks just for being an incredible partner of ours. But, yeah, this is a wonderful conversation, and and I hope that the the audience here today learned a lot from Jarred. Thank you so much, and thanks for having me. Cool. Alright. Amazing. Over to you, John. Thank you so much, Amber. Thank you so much, Jarred. I learned a lot from that conversation. I think just it's always helpful to to hear it from the, the horse's mouth. So I think there's two questions that I I wanted to answer that are somewhat related. One's from Jin Yong, and one is from, Ashley Gritzman, which is, are you planning to have enterprise the team use. So just to clarify the nomenclature for everyone, a plug in is an anthropic nomenclature terminology to describe a package of skills connectors, access MCP servers. So you could have, like, a financial services plug in. You could have a marketing plug in. The general answer is right now, custom roles are really only designed to cover spend models, workspace access. MCPs, tools, plugins, skills are all controlled through something that we call server managed settings. That being said, on our near term road map is to roll out this in kind of a a single pane of glass. So I I wouldn't say and I can't commit to, like, hey. I think you all know how fast we ship, so it could be sooner than maybe I'm leading on, but that is something that we're planning to address is just generally having the ability to kinda control anything and everything within the Anthropic platform, across the enterprise because we know how important that is. Awesome. Let's see. Let me look. Oh, here's another good one. So this one's from Josh Aziz, which is, is there a role that allows for extending spend approvals but not having access to other admin controls? As an example, someone who owns spend budget in their org. Fantastic question. As you can imagine, this is one that we get a lot because different BUs have different budgets, different teams. So we are planning to create sub administrator roles soon. So as you all know, there's primary owner or maybe you don't know, so maybe I'll I'll reiterate it. There's a primary owner, their owner, and their administrators. We're planning to roll out almost, I would describe it as, like, team or role based admins. So you could have an admin that only works on your SSO and your security settings. You could have a financial admin that controls that. And then in that same scope of work, we're going to extend that so that you could, like, be a VP of marketing and you can control your team. And that all comes from the SSO and the SCIM as well. So concise answer to your question, John. Yes. Can this one is from Anatoli. I don't want to say your last name because I think I will mispronounce it. Can access lock users to a specific model? Not everyone had to use all the models, for example. Yes. That is rolling out shortly. So right now, the main control that we have is turning on, fast or not fast mode, especially for Claude Code, which I think is a little analogous to this. But, yes, we are rolling out the ability so that everyone who is in a certain role only can have access to Sonnet or certain people can use Opus just because, I think, you know, very frankly with you all, a lot of people use Opus somewhat unnecessarily. I think that there are a lot of things that people underestimate Sonnet's ability to do around complex tasks. Let's see. Sorry. I'm just scrolling through. I believe can we see this live in the UI? We've already covered that. I think the other main one that got a lot of upvotes was can a small team access the same chats data that's output HTMLs? Essentially, you're asking about collaboration. Naturally, this is where we think I I can't commit, and this is just John or John speaking here. Can't commit to us doing this in the short term, but I think most enterprises and employees have gotten to the point in which collaborative tools are the norm, whether it be Google Docs, whether it be Microsoft, like, live online shared. So the next step in our journey is that kind of what we call multiplayer mode where you can create sub teams that can all live and and work in sessions. The difficulty is around the architecture because of how Cowork is deployed, on a local machine as opposed to on virtual machines. I know I'm getting a little into the technical weeds. But, yes, it is something that is very top of mind for us in how to make it more collaborative, generally speaking. Awesome. I think I answered most of the upvoted questions. I think there's, there's, like, kind of one just generally speaking around, like, team plan versus enterprise versus pro. A lot of what we've discussed is really all about our enterprise plans. So the admin analytics is available on the team plan, but a lot of kind of what we had referred to so far, you know, we're rolling out the, enterprise first. The other thing I will say, though, is the open telemetry that we referred to, that's also available on the team plan. So if you are a small group, you are a small, team, you we don't wanna leave you out to dry, and you do have some of these capabilities as well. Let's see. Sorry. A lot of questions when I are coming in the most recent one. Yeah. There there are design architecture documentation around the security framework. If you go to trust.anthropic.com, that is our trust center. So I'm we have nothing to hide in terms of how the security and architecture works. You just simply need to put in your email, and then we approve you to have access to that. Naturally, you need to, have a NDA signed with us, to be able to access that information because it's all gated, and, you know, our enterprise customers have access to it. But it has a full explanation and architecture of how we deploy in all those machines. And it's, I think quite a novel architecture seeing as, you know, PayPal and other large enterprises have adopted it that are completely secure. Awesome. There are a lot of questions piling in that I don't think I can answer. Thank you for the congrats on the Opus 4.7 launch. Maybe the last one that I can just add in is, like, can we import existing RBAC policies? You you should be able to. Yes. I think that, like, I would recommend reading the documentation and contacting your team, but we have designed it so that it works seamlessly with all of the Entra and Okta and whatnot. And with about thirty seconds left, I just wanna say that we do have a self serve enterprise plan. So if you're on team or you're on pro, anything like that, you do have the ability to upgrade. That demo environment that I showed you was on the enterprise plan as well. And now we have opened up a survey, so I really appreciate everyone for joining. I hope you found this informative. Again, just as a reminder, we will be sharing recording of this. We'll be sharing these materials, and I hope you found it helpful. And I hope you all have a wonderful day. Thank you.