Name: Claude for Legal teams
Uploaded: 2026-04-21T18:32:03.081Z
Duration: 56 min 48 s
Description: Claude for Legal teams

Transcript for "Claude for Legal teams": Okay. Awesome. Let's go ahead and get started. Thank you all for answering the polls. I'm excited to kind of walk through all of the results as well. I'm Nancy from our marketing team here at Anthropic. Thank you so much all for joining us today. We're excited to have you all since, you know, since we launched the legal plug in, a few months ago. We've really just seen so such overwhelming demand here at Anthropic around how to use co work for legal. So this webinar really is aimed at teaching you some best practices directly from our own legal and applied AI teams. So without further ado, let's go ahead and dive in here. So, first, the top question that we almost always get with these webinars is, will this session be recorded? And, yes, it will be. And we will make sure to send this to everyone at the end of the session. Second of all, this is very much intended to be an interactive session. So please do enter your questions into the chat window, and we will answer as many as we can. You'll also see an upvoting function, so you can upvote the questions that you would most like to see answered. And finally, give us feedback. We run a lot of these, webinars, and we read every comment. We're always working to make these sessions as valuable as we can. So now I'm excited to turn it over to, my wonderful colleagues here, Mark Pike and Maggie Russo. So I'll let them introduce themselves. Awesome. Thank you so much. Hi, everybody. I'm Mark Pike. I'm a member of Anthropic's legal team and the product lead for Claude for the legal industry. I've been here at Anthropic for two and a half years, which in AI time definitely feels like a decade's worth of change. I'm joined by my colleague Maggie Russo, who will be driving most of what you see live. Maggie, why don't you go ahead and introduce yourself as well? Thanks, Mark. Hi, everyone. My name is Maggie Russo. I'm on the applied AI team here at Anthropic. And our team kind of sits across product, engineering, research, and, of course, our go to market team. And we work with our customers to make sure they're getting the most out of Claude. And one of the things that I've been able to do, while at Anthropic is work really closely with our customers in the legal industry. So I'm super excited to share some of those learnings today, and walk through a live demo a little bit later on. Thanks, Maggie. So quick thing before we dive in here. I'm excited to say that over 20,000 of you registered for this webinar, That is absolutely insane to me. I've never seen a number like that for a a legal session in my career. It tells me the question here has really changed. I think, like, a year ago, lawyers were asking whether to even use AI, and now everybody's asking how to use it well on real work. I actually pulled people on LinkedIn last week. Feel free to find me on LinkedIn and and follow along. But I I asked people on LinkedIn to comment what they were building with Claude already, and I got back a snapshot of what the community has been figuring figuring out to to build with Claude. Solo practitioners building their own contract and matter management systems from scratch. In house legal ops teams open sourcing their outside counsel management playbooks on GitHub so peers can fork them and build on top of each other. Trial teams are building purpose built back ends for large litigation matters. In research builders, we're wiring millions of statutes and regular regulator decisions across dozens of jurisdictions into Claude via MCP. You'll see the same building blocks that they're using, these skills and plug ins and connectors. They're gonna show up in everything that Maggie and I walked through today. There's actually one story that stopped me in my tracks. A lawyer named or or sorry, a paralegal named Andrew reached out and emailed me this week. He was part of a a four person pro bono team, two lawyers, two paralegals up against an AmLaw 200 hundred firm in an elder abuse case, a pro bono case. Andrew built a tool on our API that sat at the counsel's table during this trial. It was pulling in cross examination angles, sometimes before opposing counsel could even finish asking a question. And they were able to walk out with a large jury verdict for their client. These tools that we're building here in Anthropic are ties in access to the best legal support, not just for big law firms and in house legal teams, but for many who need that support. So I'd I'd encourage you to think of what you could build today and and meet this moment. So, with that, I wanna do a hands on exercise here. I want you to really feel the range of this tool right away. I think we're gonna actually drop this exact prompt into the chat so you can cut and paste it, into your own Claude. It connects to your email and calendar, your Slack or your Teams, and can create a briefing of your day. Who you're meeting with, what's outstanding, what to read before you walk into a meeting. This is the kind of a great chief of staff I think all of us want to have, who who might drop a brief on our desk at 8AM. Like, who doesn't want their own personal personalized chief of staff? I'll also confess, I took the same prompt and I asked it to write me a daily horoscope that reads from my actual calendar. And with Claude's help, I reverse engineered my kids' toy sticker printer. It's this little Bluetooth printer. And, I asked it to give me a horoscope of the day, and it says, you're gonna crush your webinar today. So it's awesome. Sort of a silly fun example, but, it raises a serious point here. The same setup, calendar, email, a scheduled task is what runs your morning briefing. It can also be what helps you with an MSA review against the playbook. It can help you with NDA triage, regulatory assessments, wired in once, and it's wired for all these use cases. And this this really is a working session. It's not just a demo. These are gonna be real documents, real workflows, real questions. So let's continue to get into it here. We're gonna show you a different way of working. You can delegate real legal tasks to Claude. It's gonna read the same matter files your team does, run the workflows that you run every week, and carry context across Word, PowerPoint, Excel, your email, so nothing gets lost in the handoff. And it's not theoretical. Our own legal team at Anthropic uses Claude every day for this type of work, contract review, research, and the work that used to eat up hours. We call it antfooding. We refer to ourselves as ants here at Anthropic, so we like to antfood what we're building. Many of you are interested in technology but aren't builders yourselves. The the first time I saw Claude take action on something for me, not just to answer a question, but actually do the work, it really changed how I thought about my own job. My engineering colleagues had that moment with Claude Code last year, and I'm feeling empowered now with co work. And that's what we're hoping, you learn today. So how does Claude know my legal work? I'd like to present these four pillars. One is connecting to your live data via model context protocol or MCPs. These are the the USBCs of of Claude, of AI. It's an open protocol that helps you connect things from your your matter management system, things like I manage your net documents, pulling in stuff from CLMs, Drive, Outlook, the full Microsoft stack. Claude reads the same matter files your team does. It's not a static upload that goes stale the moment a new draft lands. It's live across these systems where your work actually lives. The second is that it has legal skills. The workflows your team runs every week are codified, NDA review, contract redlining, privilege privilege log drafting, matter intake, clause library checks, checklist, precedent search, deal point analysis. Claude doesn't just start from a blank page on the work you do hundreds of times a year. It it pulls from that corpus of knowledge that you've created within your department and can help empower new employees who maybe didn't even know to look in these places. And again, it understands these playbooks. Claude reads the agreement structure the way a lawyer does. It tracks defined terms across exhibits and schedules. It explains in plain English what a clause actually does and flags exactly where the risk sits, not just doing a keyword search or an index. It actually has comprehension of how the document holds together. And then it carries that context across apps. The red line you ran in Word can become a summary slide in a PowerPoint. It can then become an email that you draft and send to a client. One thread, no re explaining to the AI how these things work. No copy paste handoffs between these tools. The work moves with you. And I like to ground things here in in calling out that anthropic's been at the the frontier, for the past eighteen months. Our models have been the best in the world at coding. And I think if you had told me, a few years ago when I started anthropic that the best model in the world for coding would also be good for legal work, I would have been shocked. But it turns out you don't need to fine tune models to give Claude the engineer a legal degree. Instead, you just need to give these tools access, the same tools that the lawyers use every day to get their work done, and that's what helps it become a a great teammate within the legal context. Alright. I'm gonna take things back a couple of years. How many of you remember your first experience interacting with generative AI? Probably 2022 with Chat GPT or in 2023 using Claude as a more of a, like, chat interface. You ask a question, you get an answer. You were getting summaries. Maybe you're writing first drafts. There were probably a bit of hallucinations as well. But then something happened. Last year, it was only happening for developers. We launched Claude Code, and they got access developers got access to something more powerful. AI that could actually do things, not just answer questions, but build things, work agentically, ship new products and features with incredible speed. Claude Code changed what's possible for engineers. And now we're beginning to bring other functions, into the fold here beyond engineering. Cowork does that same work for knowledge workers. It's a desktop agent for users to handle complex multi step tasks using Claude, and Claude helps get it done. It's the same engine as Claude code. No terminal required. Whenever I look back at terminal, I I still think of the movie The Matrix. These these screens with all this code running across it. Cowork is in the four corners of an application, that is in a familiar interface for those of us who are are used to working with things like Microsoft Word and, you know, Word documents. It's still early. We're iterating rapidly, but but we've already seen so much interest in some very cool use cases by many of you, which makes us confident transformation is gonna continue to happen. The transformation happening inside of Anthropic Legal, looks a bit like this. We're we're using Claude to do contract review, red lines in twenty minutes instead of hours. Our litigation team is using it to do transcript search or expert prep. We're doing privacy impact assessments. Things that used to take hours, can now happen in twenty, thirty minutes. Our IP team is using it to do patent prioritization, to surface patentable ideas, from product briefs and GitHub repositories. They're using Claude agents to scan those things and then move with the type of speed that a hyper growth company like ours really needs. But where to get started? Well, I recently analyzed 742 Jira tickets, which is an oddly specific number, so you know I'm not making it up right now. That kind of analysis would never would would not have been possible for someone like me who doesn't know how to do data analysis very well. I used the plug in. I invoked that plug in in Claude and then had it read the repository of all those recent client intake matters that, that we've received using Jira. Claude did all this in a co work session. The output the, the output here it is. The output helped me identify some trends. I noticed that NDA triage this is, by the way, synthesized data. I'm not gonna leak our secret sauce of our our actual work. But, it identified that we do way too much NDA triage, and we were spending a lot of time with busy work on that. Open source software questions were really taking up a lot of time for our IP team. We saw these patterns emerge. We we identified where that busy work was, and then the next step was to ask Claude, okay. How can we build automations to actually help us get better work done and move with the type of speed we need to? And so I asked Claude, help me build a plug in that actually helps us automate a lot of this work. That's how I helped create the legal plug in here at Anthropic, and it really made an impact on the industry. And I've been excited to see others in the community build on that and build their own skills and plug ins that you can integrate with these tools and systems. Now I've shared a lot here. I I wanna pass it over to Maggie who's gonna provide more of a live demo showing these legal use cases and really get hands on keyboard so you can see this working. So over to you, Maggie. Awesome. Thank you so much, Mark. Alright. Let me pull up my screen. And while I'm doing that, I did see a question in the chat come up, regarding the legal plug in and and folks, asking where they can actually install this. So because I'm gonna be using a couple different skills and things from the plug in, I'll just walk through this right now. I'll I'll show it again in more depth and kind of talk about, how you can actually customize using Claude a little bit later in the demo. But just so that folks can follow along, if you go into Cowork and customize here, either in your personal plug ins or org plug ins, you should be able to see under anthropic and partners the legal plug in here. And this will give a little bit of an overview of the skills that it has and the connectors, and then you can just go ahead and hit this install button, and then that will add the plug in for you. We will also send out these directions after the call, I believe, where the information is linked, in our resources. But, again, that's customized. Go to plugins, and then just search for the Wego plugin under anthropic and partners, and you should have that there. Alright. So going through a couple, different demos today, I I wanted to kinda show you what an average day in the life of a member of the anthropic legal team, might look like. And, Mark, you're the expert here, so please, keep me honest and chime in with your anecdotal evidence, but wanted to walk through a couple of different flows. So we're gonna start with a really common problem and task, which is prepping for a meeting that you maybe didn't have, quite enough time to prep for. And I know that not everyone has access to Cowork yet, so I wanted to start with an example in chat, and then we'll flip over to Cowork and show more long running task, as well as working with Claude directly in the Microsoft Word add in. Okay. So to start, I'm gonna be working inside of a project. So I have this project pinned here. It's called Outside Counsel Harrington & Web, which is a a fake firm that that we're working with as our outside counsel. And you can think of projects really as shared work spaces. So in this example, my whole legal team would have access to it. I can add my team members there. I can add instructions for how I want to work with Claude within this project as well as, tools to use with it. So I have, my G Suite here connecting my calendar and drive. And I can ask also, add files directly. So I just have a firm overview here of Harrington Webb, kind of who we work with. That's gonna be available to every conversation in the project for everyone in the project. So these instructions, these files, it's basically just allowing my colleagues who might pick this up later today or tomorrow to have the exact same context that I do so that they're not starting from zero. Alright. So let's see how it would help me prep for a meeting. So if I go to my calendar here, it's a light day for me. I've just got one meeting here at 2PM. An outside counsel QBR, I can see the attendees in the agenda here. I've got some files linked, for my internal team to look at, some notes from prior meetings, maybe some, looks like, some spend, information in a CSV file. And then I also have some email threads in Gmail, from from previous correspondence with this firm. So I'm gonna go into chat, and I'm gonna ask it to use a skill. And you'll you'll see that here in a second. Basically, I'm gonna say use the meeting brief skill to prep me for my 2PM QBR, with Harrington & Webb today, pull from my calendar, email, and Google Drive. So you see the first thing that's happening in this chat is that Claude is reading the meeting brief skill. Now this is based off of this is a personal skill that I'm just using for, the demo, but it's based off of the meeting briefing skill that is part of the plugin. So if you have the plugin installed, you'll be able to to use this directly. And you'll notice that I didn't actually use a slash command or anything special to make sure that the skill is being used. I just, used its name. I could also use a description, like, I'm I'm prepping for the meeting. Help me help me create a brief for it, and Claude will know that that is what I'm talking about. And then as Claude is working here, we can see that it's using the tools it has access to via my connectors, which, again, we're gonna go back into the customized tab in a little bit today and kind of talk a bit more about how to set those things up. But we can see that it's used these integrations. It's looked at my calendar events. It searched through my email threads, and it searched my Google Drive files. And now it has everything I need to to prep this briefing. So we can see that it's loaded a pretty comprehensive briefing here, and it's able to give me a participant's table. So it knows from the calendar event who Diane and Marcus are from the Harrington and web team, kind of what their their role on the call is. I've got a rough agenda. And then the really cool part is that it's pulling in open issues that I maybe wouldn't have noticed ahead of time. So one thing I'll call out is, hey. Like, we have a a draft out for for an engagement contract, that has a different rate than what we have in the email thread. So that's flagging that from my review as something that I either wanna talk about on the call or address prior to the meeting today. Looks like we have some action items that we dropped and and kind of forgot to do from our last meeting, which is likely pulling from previous meeting notes. And we can see that Claude is actually, giving me its sources here of where those things came from, as well as some other open items. Right? So, I've got some talking points and things that might come up in the meeting, and I'm walking into this with a much better, much more prepared than I would have been, if I was just trying to read across all of these, you know, disparate notes and emails and things like that, on my own ahead of the meeting. So one thing I can do to take this a step further is say, hey, Claude. Draft this as an internal email to my team. And what Claude's gonna do is, go and add so actually, it's it's I didn't even plan for this, but it's it's using, our system prompt to say, hey. I need to to add a draft, before sending. I'm gonna allow it to go into my Gmail via the Gmail the Gmail MCP, and go ahead and draft an internal team note to the anthropic legal team members, that are gonna be on this meeting with me. And so that way, I'm able to say, hey. This is not only useful for me, but I can also, share this with my team and make sure that everyone has this context as well. You know, I might have also asked Claude to create a calendar invite, for example, maybe an hour before the meeting or thirty minutes before and attach this as a a Google Drive doc or something like that. But if I go into my yeah. Go ahead, Mark. I I was just gonna chime in. A lot of times I meet with customers and and lawyers generally, they're very concerned about AI taking action without their approval. And so I. think that's prompt is a great example of keeping the human in the loop. So for lawyers, you you wanna ensure that you are checking things for hallucinations or making sure it's not speaking, to to things you didn't want it to. And so, I think for a lot of lawyers, that system prompt is a really good solution to something that the people have raised questions about. So I'm I'm glad that popped up. Absolutely. Yeah. Yeah. We'll we'll actually look at that in a little bit more detail. Great call, Mark, in the how to set up the connectors. So you can actually set up for for your own personal preference. Of course, like, your your organization might have some settings already enforced there, but, you know, what you would want Claude to ask for permission for versus what you're okay with it taking, direct action on. So great point. But I allowed it in in that case just once to go in and create this draft. So if I go into my email now, I can see I have this draft email created already, with email addresses for my teammates. And I can go ahead and review this, and then send it if if it looks good. So this is just one pretty basic example, right, of something that I could do that would help save me some time. I'm sure we've all had those moments where, you know, we know we have a meeting coming up and there's things that we still need to review ahead of time. So, just a a really, you know, simple example you could apply today, something you could do today, regardless if you are in chat or Coworker or Claude.ai on the web. So that's chat. So so, you know, this is kind of where chat can start to hit its ceiling. There's a lot of powerful capabilities, but chat is basically reading what I pointed at. Right? So it's it's working within a project or I'm adding files directly to the conversation, and then it's handing me back a deliverable. But it's not necessarily going to, like, open a folder on my laptop, chain multiple tools together to accomplish a specific outcome that I give it. For example, like, taking this doc, that we wrote, and making edits to it, filing it in, like, the the matter folder for a particular case or something like that. Those are three multiple asks across the conversation. So let's move over to Cowork and kinda take a look at this, of how we could do something that's a little bit more long running. So let's flip the roles here a little bit. So we're no longer the anthropic legal team. We're gonna pretend to be, our friends at at Harrington and Webb. So it's the same firm, that we were just going to meet with. They have their own kind of similar problem, and we're gonna pretend here that we are Marcus, who was one of the people on the the call, the associate, and that he has five NDAs sitting in this folder on his local machine that are waiting for review. So if I open up this folder, you can see that I have Cowork pointed at it. This two Cowork NDA triage folder, You can see within the incoming NDAs folder, I have these five NDAs. And I also have a playbook. Right? So this is a representative file of, hey. This is how the organization is, is set up to to triage these. But the in your case, this might be, a living, breathing document that we're connected to via an MCP. It might be something in Google Drive that we're pointing out, or it might even be a skill that your team has actually created. But the important thing about co work is that it's it's kind of like club with hands. Right? It can read folders. It can open Word docs. It can write files, run things in parallel. So what I'm gonna do is is keep it pointed here and basically ask it to triage these NDAs, using using a skill, and look at that, that playbook that we have there. So I'm gonna go ahead and kick this off. And I'm gonna say I have five NDAs in the incoming NDAs folder. Use the triage NDA skill. And you can see this time I'm invoking it with a slash command to review each one. So this is gonna kick off. And we can see the progress bar up here. We might actually, for for a longer running task later, see that build a plan. But because this is using a skill, it has a plan that it knows it needs to execute against based on a workflow we've given that. So with that, and while this is running, let's actually jump back into the customized tab so we can take a closer look at this. So here we have the legal plug in. So we touched on this a little bit earlier, but this has basically been put together in, collaboration with our team. So Mark has played a huge role in this. So thank you, Mark. With coming up with, you know, skills, connections, things that are gonna really help you, augment some tasks, you know, just by installing the plug in and using it with Claude. So, if we go look at our skills yeah. Go for it, Mark. I I was gonna say, I yes. I built the plug in. I the weird thing is I always tell people not to use it out of the box, which is a weird thing to. say. Like, I've this is this thing we launched, don't use it out of the box. It's at its best when it is actually been remixed and incorporates your own company's voice and risk matrices and your playbooks and your fallback language. And so, you know, it's like a you would maybe you wouldn't wear a suit that you bought off the rack. You would you would get it tailored to ensure it fits you well. Same thing with with these playbooks and plug ins and skills. So as you can see under the hood that Maggie's showing you, there's just these plain text files that incorporate things like how you would typically do an an NDA triage. But you should feed it your existing work, and it's Claude's very good at helping you create skills. It's kind of recursive. You can use Claude to create skills that then Claude then uses. So yeah. Yeah. I'm glad you brought that one up. It's a good segue into the skill creator skill. So, as Mark mentioned, this is a way that you can actually talk to Claude about a task that you would like to create as a repeatable workflow. And it will help you do that either directly in line in a conversation, which we'll see in a bit here, or based on maybe a a text file or something else or a word doc that you have created, that kind of already walked through that process. But, yeah, great point, Mark. You you really do wanna kind of customize these things. You want to, add your own flavor, add your own tribal knowledge so that it's more personalized, more more, customized to your organization. And there's an incredible community of skills that are being created on on GitHub and the other third party platforms. So, I I always love reading on LinkedIn. when people drop new ones, and I give them a test spin. So check them out. Yep. Yeah. Yeah. So the the the plugin comes with all of these preloaded. I think I have one or two that are that are outdated here, so ignore those. But this is the skill that we were using, in Cowork currently, and we'll flip back to that task in a second. But within my own personal skills, if I have, something that I've created or I've, you know, downloaded for my organization and and added, some customization to for my team, I can also share those within my team just using this button. So adding people, within my team or, like, a list server or something that I wanna share that skill with. And then as Mark mentioned, having kind of, like, an organizational marketplace, that your admin can control is a great way to distribute, for folks that might actually be managing, the their their organization's skills library. It's a way to kind of share those across the organization so that you don't have, lots of different kind of, disparate skills that aren't available to everyone. So it's really a way to kind of democratize that organizational knowledge and make sure that people are following best practices, as well as enabling people to really speed up, the automation of their work. And if you think about it, if you typed all of this into Claude, that's a huge prompt. We can invoke this with just a slash command or a single sentence, something very, very simple. I'll also talk about connectors while we're here. So I think this was something that I saw some folks in the chat ask me about. So I'll just use the Gmail connector as an example. I have a couple different types of tools within this MCP, so read only tools and write delete tools. And you can see all of these. By by default, we can set to always allow, needs approval, which is what you saw Claude ask me before in Gmail, or or blocked, which means that it's not able to to happen from, work within working with Claude. You can also set them to be custom. Right? So what what the connector comes with out of the box or what your organization kind of ships out of the box for that connector, you might want to be a bit more conservative, so you could switch these to needs approval. I could switch any of these to needs approval instead of always allow. So it's really just about what you feel comfortable with, and you can update these over time. So that's kind of how you can, customize and have some more control over what Cloud is able to do on your behalf first kind of working, with you and with your permission. Alright. So let's jump back into our task here, and let's see what Claude has been able to do, while we were talking through the customize, customization options. So it looks like Claude has been able to write our triage report, which we can open up here, and it has flagged a couple different issues for us to look at. So I'm gonna scroll up to the top here. But we can see that it triaged all five NDAs in our folder, and it used every single one of these NDA documents to basically come up with this summary. Right? So we have five incoming NDAs. They each have a specific classification, and we can see that three of these are classified, as high risk. And if we go a little bit further within the report, they each have, different areas of risk called out. So if I go down to let's see one of the red ones, like NextGen AI Corp. It has a a red classification here. Here's the part the receiving party, which is us. And then I can see the actual detailed results here. So, again, this is all criteria that's set within the skill. The the prompt that I actually gave it within Cowork was was one sentence. Right? Use the skill triage these NDAs. But what I get is a super robust review, of everything within these documents, that I can actually use as a handoff to go into Microsoft Word directly. So, I can scroll through this a little bit more, but, you know, definitely encourage folks to to use this. This is also, I think, pretty applicable to, like, the contract review skill. I don't know, Mark, if you have other kind of examples that are are sort of akin to this. But, Yeah. Yeah. Yeah. yeah, Like, go for it. redlining is is another huge one. Wow. The the q and a, is really popping off. A lot of great questions in there, that we can follow-up on. But, yes. It's quite easy to create these types of skills and playbooks and to ensure that they're only available to, people within your department. And I like to think of it as, like, imagine a new hire joined your legal team. What knowledge would you like that new hire to have about how you do legal work? And that's what Claude is. It's effectively a a Coworker Cowork. So you're giving it all the legal knowledge and access to those. But other it's possible to set that up in a way so that other people at the company aren't accessing those same skills, and you can really ensure that you've got the right information barriers and, set things up in a privileged way on our commercial offerings. I I would emphasize that. There's been a lot of rulings recently. I get a lot of questions at CLEs about this. Our enterprise plan offers, the ability for you to ensure that you're maintaining privilege within the boundaries of using Claude, with those of you on the legal team. So, yeah. And I I've loved creating documents, memos, doing NDA triage, commercial redlining inside of Claude. But I think what you're about to show with Claude for word is so powerful. This is something that we've just released in the past few weeks, and I'm excited to see what other surfaces we can bring that to. But it's Claude meeting you where you are inside of your word processing environment, which for those of us who are lawyers, that's where we spend a lot of our time. So, this has been a a great breakthrough for us and our team and our productivity. So starting over in co work, but then shipping it over to Claude for Work. So, yeah, Maggie, why don't you take it from. there? Yeah. Thanks, Mark. Yeah. I kind of view this. I think I saw some questions earlier about Claude code and, like, using the difference. I almost think of, using the the word adding as kind of similar of using the terminal verse and IDE if folks are familiar with that, like a kind of environment where you can see, diffs or, like, edits to the code in real time. This is kind of like the same idea. Right? So if I I open the Claude, add in here on the right, I can see now that I have my NDA documents. So, Claude didn't make any edits to this yet directly in Cowork. It just produced that that triage report for me. Right? And I think this NextGen AI one was the the worst offender according to the report. So I'm gonna open up and and actually have it go through and and do the, do the red lines directly within Word. So, for folks that haven't used this yet, if you just go to add ins within Word, if you have access, you should see Claude there. And then you can you'll be prompted, I believe, on on your first sign in to just log in to your account. But once you're in here, you'll see that there's some suggested skills that are specific to working within Microsoft Word. I have access to all my other skills that I was using before. So you can work within a very similar way that you would work within chat or within Cowork. But what I'm actually gonna do right now is add this file to it. So if I go back to, my webinar here and go to this folder that we're working in, I have, this triage report file. Right? So I'm gonna upload this and just say based on the report, on the report, start redlining the NDA. And we can check here and make sure that this is being added. Okay. Great. So we have the triage report, and then I can say based on this, start redlining this NDA, and I will go ahead and kick this task off. So what this is gonna do is it's actually gonna go through and read the report. I could also if this was just, you know, in line and in Cowork, copy and paste it and put it in the chat. I could, create this as a Word document and add that directly in Word or work across two different, files. But Claudia's gonna go ahead and kind of review the the work that's already been done here. It it's pulling in what we've already flagged as the most high risk issues, and then it's gonna ask me for permission to go ahead and deliver these as track changes. So I'm gonna give it that approval, and then it's gonna take off and start working here. One thing that's really cool about all of the, add ins for Microsoft Office, and I'm gonna go ahead and allow this, is that they can kind of share context across them. So let's say I wasn't working on an NDA. I was working on, like, a meeting brief, for example. Something I could have done, I I might have mentioned this when we were in chat, is, actually create the meeting briefing as a Word document. I can have that open chatting with Claude about the briefing and the agenda and the the talking points that I wanna cover, and then open up PowerPoint and start to create a deck for that. Right? So that's another great use, of working across the the products that that you're working in every day today. Same thing for Excel. Right? If there's information related to, an analysis of some sort that you need to pull into a briefing, Claude is gonna share that context, which is super useful when flipping back and forth between those different tools. So we can see now that Claude has gone ahead and staged these red lines for us, so we can jump to each one. We can see that it's actually giving us the reasoning for it. So we know that this triage reports reasoning is grounded in this skill. We know that this skill is actually using our playbook, so it's kind of like a a nested sort of, system there for grounding these results and suggestions into what our organization determines is is the way we want to, do this work. And then I can go ahead and either dismiss or apply each one here. Right? I could apply all, but I can basically go through and agree with these red lines or change them, and do this directly within words. So, this is just one example. I I think, Mark, you you mentioned a couple other ones. Right? But working directly within Microsoft Word and kind of having Claude in that sidebar table to chat with it and use it as an assistant, as you're going through these types of tasks. Yeah. I'd I'd mentioned redlining, you know, sales agreements. I'm a product lawyer by training, and so, I love to read through my launch calendar product briefs. And I have a checklist of issues that frequently come up when we're launching certain types of products. And so being able to use that checklist as a skill and leave comments for my product managers and engineers so that they're aware of issues that might be popping up. And then, yeah, the context switching, like, we don't just live in word documents. Oftentimes, we'll do this, and if you're at a law firm, you're creating a pitch for a client, and maybe that's in a PowerPoint. Or maybe you're working on a large m and a matter, and you need to go over to Excel and, keep track of things in a spreadsheet and and even do, some data analysis there. Sort of that, like, entire office agent trilogy, the those three important surfaces that we often work on, and then also incorporating email. And this is so powerful because it can share context, and on top of it, you can update the skills and plugins over time. You can tell Claude, you know what? You keep getting this wrong. Can we actually update the skill so that, going forward, other lawyers on my team know that, this is how we approach these things. It it's really. cool. The other thing I've loved seeing in the community is, like, lawyer I think of many of us coming up in law school, maybe we're on journal, and you had to do blue booking or other formatting exercises. And it it feels like every state and local jurisdiction and courthouse has different formatting needs. I'm I'm not a litigator, but I've heard this feedback a lot about how difficult it is to get word documents to be formatted the right way. There's some incredible community skills that help people format Word documents to look the way you need them to. And this is only just the start. We just launched this product recently. With the pace of innovation that's happening right now and and how quickly we're able to improve these documents, or formatting skills, things are only gonna get better over time. So I'm I'm really enthused about what I'm seeing here. Yeah. Awesome. Thanks so much, Mark. Yeah. Speaking of of creating skills out of these types of flows, I think one thing that's really cool that we haven't addressed yet is actually creating or generating a new skill off of something that you've just, worked with Claude to do. So, for example, you know, I've used a skill, obviously, in this task to perform the sort of workflow of triaging the the NDA documents. But there's a lot of ways that I could extend this from from what I've done. My likely next steps if I were just to use this were to would be to go through in Word, and either approve or or dismiss or maybe modify some of those red lines. After that, I'm probably gonna go and draft an email to the, to the other council and kind of explain what our requested changes are, maybe attach the document, or maybe I'll send it to another person on my team for review first. Right? So there's a lot of steps that come after this. And I think as you get more and more comfortable working with Claude and and more codified as as we say, you can kind of add in more and more of those things to a skill or some other repeated workflow. So for example, I could paste a prompt in like this that says, save what we just did as a skill called daily NDA intake or contract review intake or anything that's gonna be a big part of your day to day. Right? I want to scan the watch folder, which I can point Claude at for new NDAs, and that might be something in your drive, and then say triage it against the playbook using parallel sub agents, produce the risk report as a doc, docx file or draft red lines for anything high risk, and then file the outputs into this, you know, specific file path and draft the emails. So this is going to actually ask me to create a skill file. And I could also just as easily, create this in a in conversation with Claude or use the skill creator skill, to do this. So it's gonna actually go and package this up, and then I'd be able to to run it on demand. The other interesting thing, and I'll actually if I go up here, you can, click turn this into a skill, which will kind of kick off that prompt for you. But the other really cool feature of this is the schedule feature. So if I click this, that's actually gonna kick off the slash schedule. And what that allows me to do is set this up as a, a task that I run on a daily basis. Maggie, you've got a great example of a scheduled. one. Go for it. So, my teammate, Pamela, she joined a few months ago, and she's on our regulatory team. And she was chatting with co work and frequently having to look up, updates in the regulatory space, news articles. And then she would spend about two hours a day synthesizing all of the latest news, for regulatory enforcement and and new policies and things that were happening globally. She, through a conversation with Claude after scheduling lots of these types of exercises, realized she could create a daily newspaper, and host it within our Google sites. So she actually built a regulatory tracker that we all use. And so something that used to take her two hours every day, Claude was able to schedule it, and then she was able to actually create an external resource, hosting it within our Google site so that other people can read it over a cup of coffee in the morning. I I know we talked about that chief of staff earlier as well and having. that a daily brief be ready at 8AM every morning. That's another great example of something that you can just it's repeatable work. I would encourage all of you right now. If you have hands on keyboard, what are things you could do that you feel like are a bit energy draining and repetitive that you do on a daily basis? What could you do to have Claude help you schedule those to help you hit the ground running? And just go ahead and schedule that repeatable work And maybe even publish them outwards into Slack or Teams or post them on a Google site so that your teammates are visiting them as well. That's how I've really been able to uplevel my work in the past few months using these type of cron jobs or or scheduled tasks. Yeah. Yeah. Absolutely, Mark. And I love that, that kind of research agent, basically, that your team, built because it's not something that's just available to that person on your team where they're running it daily for themselves. It's now updating automatically or on whatever kind of scheduled cadence, and and it's available to everyone. So I I think that that follows a kind of a common path of identifying those more complex long running tasks that you're doing manually, automating for them for yourself, and then figuring out how you can kind of scale or multiply your impact there across your organization. So, super cool example. I just wanted to kind of, flip over to show what this would look like if we were to run that more, long running skill with all the different steps. So, I just ran this earlier and basically what it did is instead of, you can see instead of creating just the triage report, it actually, created each individual document in Word with the red lines directly. It actually went into my email, and created a draft for each of the red lines so I can see for each of these companies, that I had, you know, been been, working with within Claude. I have an actual draft email with what we are are proposing that I can then go and attach the file to once I've reviewed it. So it's created a lot more assets than our initial conversation. Right? And these are all, you know, available then for me to pick this up. So if I go into the scheduled, scheduled tasks window here, I can create a new task, and I can call it, you know, daily NDA. I can run this as a skill, or I can just give it the prompt itself for exactly what I want it to do. But, basically, I would pick that folder. Right? Let's say there's a a working folder or somewhere where, you can you know, you expect these, files that you want it to pick up, to live. I can set this on, let's say, a daily basis at 8AM or or 9AM. And so every day when I'm coming into to work or or or sitting down to start my day, I'm not actually having to kick this off and kind of, like, watch the task complete on its own. It's already waiting for me. So, again, this is just like a really kind of I I won't say simple because there there is a lot of complex steps involved here, but, a way that you can set yourself up to be more of a manager than, a doer in some cases. Right? You think of Claude as kind of your your team of agents working for you. So you have this at your disposal to to run work in the way you want it, schedule it, and have it run on a repeatable basis. Okay. I think we've got about ten minutes left. So I'll just end with this that, we kind of covered a couple different services, right, chat, Cowork, working within Word and other, Office products. But, the the main takeaway is that it's the same model underneath all of it. Right? We're using Claude, same family of models available within these products. And that same tribal knowledge that you're creating via skills, plugins, MCPs, is driving those standards. Right? So anywhere that you're able to kind of add in your institutional knowledge, of what how you want Claude to run and work with you, it's able to to propagate that across all the areas that you're working with Claude. So that's kind of giving you that ability to hand that work off, schedule it, have Claude do it for you, and then focus your time on, strategic work and making those judgment calls and all the things that are, are most critical. Okay. That's all I had, Marcela. Let me stop sharing my screen here so we can jump back into the slides. Perfect. Yeah. I think there's just a few key steps for success here. So we talked about setting up those MCP connectors, getting started with the legal plug in. Maggie showed you where to navigate and find that within Claude. And then working across apps like the new word add in. So those are those are the key steps for success there. And we just wanna emphasize that Claude really meets your team where they're already working. So Claude for Excel and PowerPoint, Claude for Word, And then Cowork, view using things like the the desktop application, can really help you with work that spans these different systems, pull things together. Cowork is so powerful. It's using local compute and and sort of working in a sandbox and using the power of your computer, along with Claude to get this type of work done in a safe environment, which is so important for those of us who work in legal, in the legal industry. Let's see. I'd I'd mentioned the Jira analysis I did earlier. We looked at 700 plus tickets from the past few months of legal work. If you, right now, are looking for how to get started within your own legal team, maybe you work in house at a large enterprise, maybe you're in a small firm, maybe you're a large firm, maybe you work in, public service, have Claude analyze your busy work and then provide suggestions. That's how you can really solve that cold start problem. Connect it to your email inbox. What is the energy draining work? Another quick example for me is every Friday, we write, like, a a weekly newsletter to share with our cross functional stakeholders of what the legal team's been up to. I dread that Friday morning reminder. It's like an hour of busy work of compiling. You know, I manage a team of other product lawyers. I have to compile all of their their wins for the week and and synthesize it. And I don't like doing work about the work. And I think that's what Claude is so good at, is taking that busy work. For those of you who remember the the movie Office Space, the TPS reports of the world, Claude is very good at helping understand the progress, grounding itself in existing wins that your department has had over the past year. Share the old weekly newsletters and tell it this is the high bar we keep ourselves to for success in our department. Write me a newsletter for the week that shows the other similar type of wins. You know, don't be sycophantic and just tell me we did good work. Show me what's actually impactful based on what other people are saying and draft that weekly newsletter. So this is how I would suggest that you you use Claude in your work to identify the best next steps for you. We got a few minutes here where we can maybe take some question. I've been skimming the chat all along. Wow. Like, incredible questions coming in. I think we're gonna send out a link to this webinar later. Please share it with your teammates as well. Hit us up on LinkedIn. Reach out to our support team. We've got so many talented people here at Anthropic who are here waiting to help you and your legal department, transform, we can we can help with that. So please reach out. We're here to support you. Nancy, I don't know if you're around and wanna share some of the questions from the chat or maybe some of the poll results. What we got going on? You are on mute, though. Sorry about that. Yes. Okay. We have a lot of wonderful questions. So, we can spend the rest of the time hitting some of the ones that were the most upvoted across the group. So one question that came in was, how is your team specifically dealing with attorney client privilege when using Claude? It's such a good question. I I briefly touched on that earlier. There's been some recent rulings about this. I think the whole industry is really sort of waking up to this moment that lawyers are using AI to get their best work done. And it reminds me of some of the conversations I've read about that were happening in the nineteen nineties around whether or not lawyers could use email or in the early two thousands about using Claude, Claude software, Claude based solutions and SaaS. Now the the Hefner ruling, I think, is one that on LinkedIn a lot of people are discussing. This was a non attorney using a consumer plan who didn't have things set up, and they were working at their own direction. For lawyers who are using Claude on our commercial offerings, our team and enterprise plan, or prosumer and have their privacy settings set up correctly, I'm not your lawyer. You should speak to your own lawyer about privilege, but I think we are quite confident here that you can set things up to be able to use Claude in a privileged manner and maintain privilege. Yeah. So it's a it's a great question. I would love to see policymakers and other CLEs address this more because people are using this, and it's gonna keep happening. I think, again, it helps us provide the best support for our clients. The ABA actually has ethics rules that tell lawyers we need to stay up to date on technology and be aware of how it works. That is our duty to our clients to understand these things and help provide the best client services we can. So, yeah. It's a great question, though. Awesome. Great. We also did have a couple of questions thematically that came up around security. So, would love any sort of general best practices as well as how you keep information, especially sensitive case files, secure and confidential across clients. Yeah. Yeah. Maybe yeah. Go ahead. I'll I'll take a stab at it. And, Mark, feel free to to add, you know, specific to to how your team is working, with Claude. But from just, like, a broad security standpoint, so Cowork specifically is running within an isolated virtual virtual machine, on on, the desktop app. So, those kind of permissions and, you know, what it has access to is, you know, the blast radius is kind of limited with that, architecture From, like, a broad kind of feature standpoint, connectors, so MCPs or connectors, are gonna inherit those source permissions from you. So think of it as, like, you logging logging into these, these different connectors. You're kind of authenticating them. And so nothing that you like, Claude is not gonna be able to access anything that you don't have access to. Right? So for for keeping things, that are meant to stay within one specific team or related to one matter, you know, it's not gonna be able to to access anything that that you can't. Projects are, I think, are another good kind of, like, workspace delineation tool there. So that way you're able to share, you know, resources and work with Claude collaboratively, but scope to a specific team or or client, or what have you. And then RBAC is something that we've rolled out recently within Claude for Enterprise. So enterprise, Claude for Work for Work plans. So, that can kind of help, limit who has access to certain capabilities. We're gonna be rolling out more capabilities and and kind of continuing to enhance those features in the future, so look out for that. And then the last thing that I would say is, like, compliance, auditing, logging, that's all available, to be exported. So it it kind of gives a it it has that record of, who ran what, what was accessed. So it's all visible to to your organization, for kind of auditing purposes. Mark, anything you would add there? Yeah. I would just call out that our commercial plans all offer these type of admin based controls. So you can set. things up so that only approved apps and connectors in the directory that your employees are able to install those. Yeah. I think our compliance a v API is very robust. It can help with the data loss prevention or other ediscovery needs. And I would acknowledge that there is rogue IT, that people are using consumer grade plans and, that our commercial offerings are a great way to ensure that, we're able to meet the needs of your organization. Please visit our trust center, trust.anthropic.com, that has all all this information and more. Yeah. Wonderful. So we also did get a question around how legal work can oftentimes involve very long documents, which can also lead to, you know, context degradation. So curious how Claude's legal team, also addresses those types of situations as well. I I think this is why I love co work so much is that the agentic harness allows things to swarm and make plans and execute and, distill a lot of the mass corpus of knowledge that legal teams often have or encounter in their work and make sense of it all. In chat based or turn based, conversations with AI, yeah, you get that degradation. But when Claude is able to make use of a local computer, store files in the right places, and use these types of, like, file systems, we've really find found that it's able to to keep up, and it doesn't sort of have that, like, context rot. Yeah. Maggie, I don't know if you have anything to add on that one. Yeah. I'll just add that, you know, we have, models with with really large complex context windows, million context windows available now, which I think helps, but also what we saw in the the redlining use case, asking Claude to cite its sources is another big thing. That way, you have that confidence that, it's not, you know, hallucinating information. It's actually grounded in whatever skill or playbook or or other data that you're feeding it, so that you feel, you know, more comfortable using that information going forward rather than just kind of trusting it at face value. Yeah. So many so many lawyers are concerned about ending up on the naughty list of over 2,000, court filings that have hallucinated information in them. And so ensuring that you're using tools and connectors to ground these AI systems in real case law or your playbooks. And then, again, keeping the human in the loop, this is supposed to empower lawyers to get their best thinking done. We acknowledge that it's not always a 100% perfect, and you you still need to to review and make sure this work is is accurate and reflects your clients' needs. So yeah. I think that's a great place to stop because it it's we're so excited about these systems and and what they can do, and we know things are just just going to keep getting better. So thank you so much for your time today. And, again, reach out, use these resources, and let us know what you're, building. Thank you. everyone. You, everyone. Have a good one.