Webinars
Build Your Own Sidekick with Copilot: AI Agents and Playbooks for Legal Teams
49 views
What if you could build your own legal AI sidekick – without coding and without a developer?
In this hands-on webinar on June 18, Arunendu Mazumder will guide you step by step through how to build a practical Copilot agent for your legal work.
The session is designed for legal and compliance professionals who want to work more confidently and concretely with AI in their day-to-day work.
During the webinar, you will:
Learn how to build your own legal Copilot agent live, step by step
Work with relevant scenarios and use cases
Learn how to give AI agents clear instructions
Explore how to connect agents to legal knowledge, such as policies, templates, and FAQs
See how to create practical AI support for legal work without coding
Get a checklist to help you recreate and extend your setup in your own environment
This isn’t a session about large-scale process automation. It’s a practical no-code instruction for building useful AI support for recurring legal tasks.
The webinar is led by Arunendu Mazumder, AI and Legal Innovation Lead at Synch law firm, who works at the intersection of law, technology, and learning design.
Save your seat now.
Format
Duration: 45 minutes (including Q&A)
Language: English
View transcript
RANDjsadro Joukminen O, Hubbarim R Brands decine하. Svendiellaайтесь connected by motorhead RANDji A.H. GS'A.R. RANDji A.H. about things like technical issues. Even though we're 2026 and we're past Corona, sound is still an issue. But with no further ado, thank you everybody for joining us. And once again, sorry for the delay. Before we get started, we just kindly want to ask if you can hear me. And now, Aru, can you try? Hello. Can you guys hear me as well? We hope you can because I couldn't in the beginning, but now we got it to work. So thank you so much, Aru. And thank you all for joining us today. Absolutely. It's going to be fun. Exactly. And the thing is, we got a guy that I met in Stockholm, what, a month ago on today because he blew me away when we were in Stockholm. So you are in for a treat. For those who have not joined an openly community event like this before, the way it works, as you can already see now, we like thumbs up. We like questions. We like emojis. So if there is anything, you have it right there on the right-hand side. You can ask your questions. You can applaud. You can put a thumbs down. Hopefully you won't. But it's just to say, feel free to engage and feel free to ask questions. Absolutely. It's all about learning. And you're going to learn from Aru, who will present in a second. But just also so you know, the event is recorded. We're going to share a few slides. And Aru is going to teach us how we can build our own sidekick with Copilot. Before we jump into it, Aru, for those who aren't able to actually know you like I do, can you maybe share a little bit about you and your background? Hello, everyone. Thanks for joining. I can see people are still joining. So a quick introduction of me. I'm Ardandu Mazumdar. I'm from Bangladesh, originally from Bangladesh. But I have been living in the Nordic for the last eight years. I have a legal background. I studied law. And now I work with AI. You can call me an AI geek, but I'm not there yet. I'm trying to be a bit more educated on AI. But I have been working in the legal industry for the last three years. I'm working at a law firm at SYNC in Stockholm. It's a tech law firm. And I'm their AI adoption specialist. And also you can call me a legal engineer. So my main work is to help lawyers to build agent automation and also focus on AI literacy and AI ethics. So you can hear we got a guy on who actually is able to build the sidekick for us and know what legal needs. So what we are going to be talking around today is, as you know, building your own sidekick with Copilot. And Aru, the floor is yours. Okay. So I will just before going to the Copilot Studio, I just want to do a little bit of information. What exactly is in Copilot Studio, at least the part we're going to touch upon today. Let me share my screen. Now you need to just go into preset mode and it should be good to go. Now we're on questions. I think we're at the latest stage. So you need to go up a few pages and then we're there. Sorry. It's the least of today's technical problems. So you don't need to apologize. Okay. So to start quickly, I would just want to go through one thing. Like this is a very information and knowledge sharing session. So whatever AI tools you use, Copilot, Cloud, ChatGPT, you should always do your own due diligence to make sure that you are allowed to use your client data or work information. If not, it's always easy. It's also good to use it for even ideation, strategy, create workflow. So far in the legal community, Copilot is highly, heavily used since it's owned by Microsoft and a big, like a lot of law firms use in Stockholm or also I think in the legal community that they use Microsoft to their go-to-go communication. And since my Copilot is part of the 365 package, it's safe to use. And of course, it has a enterprise level security as well. Two things that I have found that lawyers or anyone who use Copilot has this question. What is this work mode and what is this web mode in Copilot? So work mode is basically connected to your Outlook, your SharePoint, your Teams, your calendar. So basically when you turn on the work mode, it can go through all your emails, all your calendar, your Teams message. So it can help you to say, summarize a meeting, summarize the emails that you have with the client for three months or also like, you know, for scraping information that you need. Web mode is mostly for external research. Like say, if you want to do a more deep research, you want to like look for, okay, what is happening in the AI Act? What are the new updates coming in? What is Omnibus? So when you turn on the web mode, that does that. But good thing about when you have an enterprise level security, even if you turn on the web mode, it actually doesn't share your work data or take out the work data outside the system because that's how Copilot is providing you the security, enterprise level security. What the modes mean in practice? So when you do, when you turn on the work mode, so you know that the system is not actually getting into your work information and then take it out from the system. So there is a trust. And in the work, in the web mode, you are allowed to do any kind of research beyond your work, but then make sure that you turn on the web mode, not the work mode. Workflow rhythm. Teams often start in work mode, like say, if you want to do, especially when you are in the open chat version, try to always use the work mode. But if you have to go to web mode, it's always, I would always recommend to turn a new chat and do that separately. So it's basically based on the needs and usage of the user. Core capabilities for legal teams. So in Copilot, they already have pre-built agent. And few of the agent are called one called researcher. That is basically if you want to have an in -depth research of something you're doing, such as like you have a client who has been your client for like seven months or a year and that you need the data. You need to basically scrape or you need to analyze what conversation you have been doing last seven months or a year and you want to create sort of like a bullet point to present it to your management or to your boss or to your team member. So researcher is good for that. And also you can also go to do in-deep research outside the work mode as well. Analyst is basically for if you want to, if you have like a lot of, I think people, lawyers who work in M&A or they are like, you know, creating term sheet, they need, they work with a lot of numbers and analysts can, could be a good, useful agent to create like a dashboard or compare data between two different type of datas. I will go through the agent in a few minutes after I explain this. And the regular agent that is like your personal agent that you create for yourself. Okay. I will, instead of going through all this, I'm going to go how to build this. If you have any questions, please can ask as well. And I will read them out if anybody has questions while you're working. But this is where the beauty of actually joining an online live session. And you can ask the questions and you will be able to see live Arou playing around in co-pilot. Now I'm going to show the, basically the co-pilot. And while Arou is getting the screen up, I can just let you know that both Robert and Inger has been asking about the recordings and the slides will be shared after. My point who is the community manager will be the one who will help us do that. And so now I'm going to, yes, we can now see, hey, what can I help you with today? I'll take it away. Yep. So this is the co-pilot studio. So as I mentioned before, there are two versions, work mode and web mode. So if you, if I say click work mode, it can actually say go for, tell me my next week meeting schedule. And while it's thinking Arou, Adriana had a question for you. Have you investigated the newly released co-pilot, co-work for the use, especially in relation to processing data? I would be honest that I haven't done that yet, but that's in my list to do it. And once I have done that, I will definitely share in the, in openly community. Cool. So you can see that it gives me a sort of like, you know, what type of meeting I have. And then it gives me also, you can click and check who is this person. And then, yeah, so you get the details of it if you work in the work mode. But like, say, if I say go to web mode, okay, again, work mode. If I say, what is the new update on Omnibus? This is the information that has been already, it knows. But if you want to do more in depth, then you go to web mode. Give me more details on the web mode. Give me the compliance requirements of Article 6 of AI Act. And if you have already enterprise level plan, you can see there is a sort of like a shield button that you can click and you can see that this is showing that you have an enterprise data protection. And it says that enterprise data protection applies to this chat. So your data is safe. When I talked about the agent, like different type of agent, I said about researcher and analyst. So if this is a pre-built agent that you can actually use in already in Copilot that you don't need to build. So they have already some templates that could help you to do research for. Okay, someone questioned, if you have a hybrid task, provide legal advice in response to work, you may want to use the work mode or combine the output of both models. I would definitely depends on if it's a specific client. First to understand the work, I will do use the work mode. But then if you do need to do like a research beyond the work, then definitely web mode, I would suggest. But it's always if you already have, if you already have the research response, if you already have the data, then I will suggest you to use researcher agent where you can add your documents or also your knowledge base to do the work in terms of giving the legal advice. So if you see here, there is an option called sources. There you can actually select. If you want to, when you do the agent to, when you use the researcher agent, you can always select if you want to use your work mode, which is a Microsoft 365. That means this is connected, connecting to your chats, your teams, your emails, your meetings and everything. But just say, OK, fine. I don't want to do I don't want to use my Microsoft 365. I just want to do web. Then you just select web. But make sure that you turn this on and off based on your requirements. Like I say now I want to do only my emails. And here I can say, could you check my emails with Michelle? And what we have decided. And while it's thinking, Aru, Brian had another question. What do you do, I presume, with the information from clients that are confidential or documents that contain information related to GDPR? How do you document compliance? I think that's a good way. Thank you for the question. It's a great, interesting question. And I think the best answer is to create your own personal agent. And that will help you to do more in-depth work, but without even leaving the space. And that's actually secure for you. And I will show how to create your own agent in a few minutes once it's done. Well, it's actually happening. We can actually go to other agent to do it. Here, analyst. When I say that you have a lot of data, you want to have a visual, you want to compare the data, or you want to present your data to your management or your team or your manager. Here, it's actually quite good. It works as a power BI. So, like, say, I added a document. This is basically a sort of fake document that I created with an AI maturity level in a company called X. So, where it will identify what's the maturity level for this company and what they need to do. What's the next steps? Could you create me a Power BI presentation using this data? So, it's doing some coding. Show me the visual dashboard. Yes. And so, while it's thinking, Aru, can I ask you a question? Many are talking about Claude. And so, today it's around Copilot. Where do you see the differences between Claude and Copilot? I think it depends on users' needs and, like, the policies they have in their companies. Because I think a lot of in-house legal teams are using Claude because it gives much better answers. And you can connect to your processes, not only just emails or your calendar, but you can connect different MCP. You can create skills. You can create projects. That is actually useful for an in-house legal team, definitely. But also, I would say anyone who wants to use Claude for their work. But in terms of Copilot, I think it's based on company to company. Because some companies require only Copilot because they have the enterprise-level protection. And also, in Claude, you can get enterprise-level protection if you have more than five users or if you want to go for the enterprise-level plan. So, what I'm trying to say is basically that it depends on if your company or if the team you are working with, whether they want to use Copilot or they want to use Copilot. Either way, it does the work. You just need to know how to use it for. Let's see. So, you can see that's, like, a Power BI dashboard that creates. So, in the researcher, I was doing this. Yeah. And you can see the steps it did. You can actually also check the steps it has been doing. What is it answering for? And it still is working. While it's working, because it takes a little bit of time in researcher, because it does, it goes to a very in-depth, in-depth of your conversation, in-depth of the information that you were looking for. I will create a, I will show you how to create your own agent. So, the easiest way to create an agent, either you can talk to the agent, talk to, you can write it here. Oh, I want to, I want to, I want to, I want to NDA agent. I want to create an NDA that will, that will have, that will have a structure, that will have a template that I use it for, that people, a lot of lawyers, they have their own standard template for NDA. And they can use it as a knowledge base, and then can run against any kind of template, any kind of NDA that they need to review. Before going to that, I saw that notification that research is done. So, I can show you. Yep. It couldn't find it, because there is actually no conversation, there is no person called Michelle. I just gave it a, like a go, to show you how it works. Let me, let me create the agent here. Agent builder. So, like say, can you create me an agent, which will, which will run NDA review. Against my standard template. Aru, when we're talking about using Copilot, what is, and now you're building the NDA review agent, and it's popping up here. How much do you need to test, like how much due diligence do you need to kind of like do with your NDA agent? Does it just work out of the, like out of the gates, or do you recommend like? So, in this case, I think, if I understood the question correctly, that how, what's the guard drills you put, right? How it works. So, I think the best way to say it is that you, when you set the instruction here, you set the guard rails as well at the same time. And I would recommend to every three months or every few weeks, you go and check if the guard rail is still working or not when you're working. Because you need to know that this is a system gets better by using it on a day-to-day basis. So, the more you use it, the more it gets the sort of the capabilities, the strengths to do better work. So, that means like, if it does start doing the better work, that means the guard rails that I'm putting, it might be failing to do it. So, what you need to do, every couple of weeks, you need to go again and check the guard rails and you need to put the guard rails again. Like, make sure that it actually stays in the guard rails. So, what happens here, I have added, it's like, either you can actually set your own instruction. You can always change instruction as you go, or you can accept the instruction that that copilot is giving to you. But what I'm going to do now, I have created sort of like a system, playbook rules. So, what is the NBA playback rules where it's sort of, it's like, I create a real book where there are deviation rules. So, that means like, which clauses are unacceptable, which clauses are acceptable. And if there are some clauses that can be accepted, but needs to flag out issues. So, this is sort of like, you can create your own NBA playbook rule. I can share it in the chat, what kind of playbook rules I have done. So, you can have an idea that how it looks like. And I can say, take this playbook rule and update the instruction. Set dos and don'ts. And also add, also add the deviation rules. Another thing, in the agent section where you create the agent, you can always add, if you want to create an agent, that is, that needs to go out in the website. Because now we are creating, that is in the work mode, but not the web mode agent. Because if you want to have an agent that also needs to check outsource, like any kind of sources verification, then you can also select this here. Like, yeah, here, either you can either enter your name or drop file. Either you can select this, like say, it only comes to emails. You can select my emails. So, my emails means like any kind of NDA comes in to my emails. It will check, it will go through my emails and see if there is an NDA review that I need to do. So, each time, if you want an NDA review to be done through your email, that you can set up here. It's up to you. It's up to the users how they want their agent to act. And if your agent, if you want your agent to, say, to do legal analysis as well, then you can always add sources like, say, AI Act regulation or GDPR or supply chain agreements or something. Otherwise, you can also add your standard template as a knowledge base here. And that will always be here and that the agent will always act upon on this. So, it knows that every time it needs to do the review, review any NDA, it needs to go run through this as well. So, you can also give them capabilities, creating a image or Excel file or doc or a PowerPoint presentation that you can do. Aru, you've gotten a question from Emily. And the question is, unlike the idea of having an NDA review agent, are there some other examples of use cases for an AI agent or in-house legal teams? What are the most commonly used for? I think you can also create a triage, intake triage. Like, you know, you can connect your email. And if there is a request comes in from a client, you can run the agent in your Outlook because Outlook has your co-pilot agent. So, you can just select the agent that you created, like a triage agent, and it will run against that request. So, that can also be possible. Or like, say, you need to update a specific client on a monthly basis. Because maybe that client is a client for a privacy sector. Like, say, DPA. You need to do a DPA or a DPIA or a ROPA update. So, those type of clients you need to let me to show. So, first, I will finish this. So, you know that you can actually create the agent here. You can also change the instruction as it goes. And then you just click create. And then you can also share this agent to either in your specific persons in your organization or anyone who you want to use it. That's how you can share it. But let's try the agent first. How can I help you? So, I will just upload an MDA. So, it flags out the issues as well. The good thing is that because you set the boundaries, you set the guardrails. So, it doesn't take the information from somewhere else. You can see that the knowledge base I added and the playbook, the knowledge base I've added, it's actually referring to the rules from that. So, yeah. Okay. Another agent that I was, Emily was asking what else can done. So, like, say, could you create an agent which will help me to inform two of my clients? So, like, say, the two clients are. And good thing about if you use this version of the agent builder, it actually does not understand what type of features, what type of work mode or web mode that it needs to turn on or not. But then, of course, you need to do double check when you're creating it. So see if it's automatically turn on the web mode or turn off the work mode or what type of capabilities it's giving it to the agent. Adriana, yes, you can always do that. As I said, it's up to the use cases you have and that if you have a standard template for software vendors reviews or the structure you follow to do it, that you can also set it up here. So this is just an example to create agents, but you can create any type of agent that is useful for you. And I will try to send a step-by-step process through openly and they can share maybe at some point. Yes. So here, as I said, you have two different clients. You don't need to give the, since it's going to be work in web mode because it needs to check whatever update comes in. So I didn't add the client details. I just add the details like what type of sector it is, what type of data they work with, and so on. I can maybe mention I have my own little agent. I don't use a co-pilot, so I use another tool. But every morning when I drive to work, it is reading out what I have of unread emails. It is taking me through my calendar and it's always nicely very British when it's communicating. And every morning it says, hello, Stein. And I'm like, that's not my name. Can you please remember that it's Steena? And then it reads out also what I have of unread messages in my kind of like Teams type of system. And that means that when I get into the office, I actually have a pretty good idea of what I need to respond to. I actually sit and chat with it because I'm driving my car. So I can't, of course, type. I would never do because when you're driving, you need to drive. But I'm talking to it and it's kind of like I have my secretary back from the law firm days. I'm sitting and I'm giving it instructions. So it's drafting my messages. And so when I get into the office, I have a lot of drafts just pending that is written in my exact tone of voice because I'm dictating it. And so that actually has made a lot of difference for me. Firstly, I don't feel I'm wasting my time when I'm sitting in the car. I also get into the office with a clear overview, prioritized ideas of what I need to do, but also lots of drafts ready for me to just simply press. And also that you can create since you work with a lot of time and you need to do billable hours. You can always create an agent that is like every day can give you, OK, how much, what type of client you work with today? Do you have you missed any any any time recording that you need to do? So it can also give you update or give you update every day after six or five. And that could actually you just what you do. You just press one button and give you the details of your day. And then you can see what time how much time you have to record. So you made the billable hours or even like if you missed out something or if there is some kind of update that you're supposed to do last week, but you haven't done it. So, yeah, I said that this is a regulatory update agent. And as the last six months, transition for future life and why it matters, impact on client A. I have one small hack. And that is when you're running late to meetings, you know, like I'm never late. Well, hence I am quite often late, not on purpose. And I wasn't late today for the webinar, I promise. But having these short kind of like summaries sent to me before every single meeting, just so that I go into the meeting prepared. It's like a five line item saying, hey, Stina, and it can go into your Slack or however you work. Hey, Stina, remember you have a meeting at 11. It is around negotiating a DPA. Their main point is to stop process issues. And it is also around limitation of liability, just as an example. So just like, oh, yeah, OK, that was what this was about. Great. So it can also help you just feel more on top. Absolutely. And also, I think you can create an agent like the work for your negotiation with the contract negotiation. You could sort of like create those criteria to add the agent like, say, a bad cop or good cop and help you say, OK, this is a contract that I need to review or negotiate. And you can always just run it with the agent and see which type of key points that the agent needs to focus on for a contract negotiation, which type of situation that can come up when doing negotiation. So you feel prepared. So we have two minutes late. So if you can take us through the last steps of your regulatory kind of agent running in the back end and then just one minute for wrap up. Yeah. So this is like what I did, like, you know, based on like two different clients. I just give the background and then what type of regulatory update that I do need to update my clients on monthly basis or every six months or even myself just to work around so I'm not missing out things. So this is sort of the as I said, as I mentioned before, that you need to make the instruction in a way that it feels like you are working. It's you that you are working with. It's not the agent. It's like you are the one that's creating these things. So the instruction is mostly your the heart, the important part of to make the agent useful is that you need to make the instruction very good as and very put a lot of guardrails and also make sure that the agent is actually talking to you, not someone else. And then this is the agent that is said, it's a it's a web based and not a work based. So you can always also check the sources that is giving. And you can always find your agents under the agent here, the agent you created. Yeah. Yes. That's I think that's all for today. That was like high flying into the realm of co -pilot and the creation of your sidekick. What is most important when it comes to AI? I think it is around experimenting. And I'm not talking, you know, the risky experimentation stuff. It is just like Arvoo showed us for the last 45 minutes. It's about playing around and getting to use the models. We need to do that for different reasons. One is keeping up and knowing what's out there. Another thing is we need to guardrail. We need to protect and we need to guide. And if we don't understand the technology, we haven't used it. It's almost impossible for us to succeed. And so that's... Yeah. Sorry, you're interrupting. No, I just wanted to say thank you for that. Thank you. Thank you, Sin, for giving me the... Yeah. I mean, love sharing. It's like... And knowledge sharing is the best you can do. And now at this with these people, like all the technology comes in, all the new models comes in. And I think people are still having a struggle to start. And just like when you see that, oh, this is different type of capabilities comes in, but I don't even know where to start. So I think the best thing you said that you need to start, you need to start today. Try it every day, small by small, and you will get there. Yeah. And we will be having more of these kind of like online sessions where we will learn different models. The next one is Claude, my all-time favorite. I just have to say, and I know that there are some that don't necessarily agree with me, but me and Claude are friends. We might also sometimes be disagreeing, but we are friends. So that is the next one we're going to go and show you what we can do. And if you guys have any more questions or something, you will find us in LinkedIn. Reach out, you know, anything. And on that note, everybody, have a wonderful day. And Aru, thank you so much for sharing. Thank you, Sin. Bye-bye. Bye.