00:11 Rob: Good afternoon everyone and welcome to another installment of our Claude webinar series. Today we’re talking about Claude Cowork. We’ll walk you through how to set it up, show you what we’re actually using it for day-to-day as an SME, and then show you a live demo with some processes and tools in action. And we’ll leave a little bit of time at the end for a Q&A, so please get your questions into the LinkedIn chat comments. And hopefully Francesca and Alex will be there to help you while we run through the webinar. But without further ado, Crispin, over to you.

00:54 Crispin: Thanks very much. But Rob’s staying this time. Usually Rob disappears at this point to leave me on stage on my own. But Rob’s staying this time. If you don’t know me, hello, I’m Crispin. I’m the CEO of The Coders Guild, a digital skills training provider. Obviously in this day and age, digital skills training providers are doing a lot with AI. Part of our mission is about getting skills training delivered to SMEs, in particular funded training through the government. So a lot of my work is based around that, accessing government funding to upskill your teams with AI. Now, I have done a few of these webinars showcasing what we are doing as an SME ourselves and the enormous benefits that we’re getting out of these tools in terms of productivity and efficiency for the team. This week, however, Rob is sticking with us as we go through because he has some unique, or a much better viewpoint for the subject area today. When we talk about Claude as a team, Claude is, I think, our favorite AI tool at the moment. It is like our Swiss Army knife, if you like, and it runs through a lot of what we do as a team, as a business. So I did want to start what we’re talking about today with just making sure that we’re all aligned on what we’re talking about and the different things. So Claude has kind of these three flavors. They have the main Chat interface that everyone is hopefully super familiar with, much the same as any other large language model. You chat to it, you set context. If you want to know things about like projects and skills and those sorts of things, go look at our webinar from last month where we covered setting up Claude as a team environment. There’s a lot of good stuff in there. However, when we talk about Claude Code, Claude Cowork, and Claude Chat, there’s a few differences. Now, really, the way that I was using Claude Code last year was actually not really writing code with it, but I was using Claude Code to connect to different MCPs, to connect to different other applications and to do stuff, to do agentic stuff, to make it do repeatable tasks. So Cowork is essentially a user interface on top of that. Anthropics, when they released that, kind of nailed the team, SME, business interface for AI. Cowork is really what kind of cemented it for us. So this is why this is much better to have Rob with me on the webinar today because as my transition to using Cowork was from Claude Code, Rob’s was from Chat to Cowork, which I think is much more applicable to how most people are going to be coming to this interface, to this tool. And everything that you will see today, Rob has built as a demo for the purpose of this webinar. Also, the run-through is a project that Rob’s actually done. Just before we get to the demo, though, I do want to kind of talk about the sorts of things that we like to show you. So the image here isn’t actually our live stuff, that’s from the demo that Rob set up. Obviously, we’re not going to be exposing live client or employee data. But Rob did manage to set up like a demo in about 45 minutes, was it, Rob?

05:12 Rob: Yeah, yeah, 45 minutes. A few different projects, prompts, artifacts, schedules. So built things in Monday for me and yeah, really quick. I was quite surprised.

05:28 Crispin: That’s good. And that’s, you know, basically just repeating, we’ve gone through like an evolved process, back and forth, that Rob will demo to get to a place. But then repeating that was super, super fast, right? I think that’s a key point to learn. Once you’re learning these tools and how they’re clicking together, to repeat that is faster. So talking about Cowork, what’s Cowork useful for above and beyond the Chat? We are all taking transcripts from meetings, but what do we do with them? Generally, oh, they might have some actions attached to them, we might share them round afterwards. But the things we can do with Claude Cowork is like, after this particular sort of meeting, get the action points that are here, send them to someone on Slack or as an email or create items in a specific board or something. You know, it’s about like when this happens, start doing this thing. Another great one for Cowork is onboarding. When you have role-specific onboarding, if people come into our organization as, say, for example, as a subject matter expert trainer, so generally the people who train for us, with us, who lead the training for us, they are industry experts who are not trainers. So there’s an onboarding process around that, and so that’s like a specific thing that we could do with Cowork. And also as you can see this artifact here on the screen now, creating dashboards and stuff that have multiple sources and one view, that is an enormous efficiency gain. Not just for creating the thing in the first place, but then how that actually gets us as a team to rally around the figures and the things that tell us about our efficiency and tell us what to do. I do want to chat as well about what to skip. Now, I see a lot of, I get a lot of AI stuff in my social feeds, as you might expect. I think it’s like 50% AI stuff and then the other 50% is people doing baking in French or something like that. Now, a key point here I think, you don’t want to be copying other people. These tools are really flexible and ready to work with you. So if you think about like what a content provider, what an influencer is putting out as their content is things that will get clicked, things that will get liked, things that will get shared. So they’re not actually in the game of giving you business efficiency. They’re not actually in that game. Actually, what they’re trying to do is make more popularity for themselves. So the things that make good content aren’t necessarily the things that provide business efficiency for you. And that’s what we’re always coming back to is productivity and efficiency with these tools. These tools give us an amazing opportunity for doing more with our time. Some other things to, as kind of a word of warning, when you start with Cowork, you want to or even in the Chat interface as well, now, using all of the connectors that you can straight away is a no-no as far as I’m concerned. I think like that would become overwhelming and a risk straight away. So you don’t want to be connecting everything that you can at once, connect it as needed and then think about why you’re connecting what. My last point, and I think, you know, that I’m guilty of this, and I know, Rob, you’re like me. I get some nods there. It’s really tempting when you get tools like this to start automating everything. But you’ve got to have a word with yourself about what it is that you’re automating. Are you spending two hours automating something that actually is only a two-minute job once a week or once a month? Would you agree with that, Rob?

09:56 Rob: I would, yeah. But there is one caveat, though, for people very new to the process, I would say it’s okay while you’re learning to try on something that you’re familiar with the process so that you can learn how to get Claude to take what you’re trying to tell it and give you the output that you want. So a little bit further on in your journey, you don’t want to be automating things that take too long. And what I’ve found is Claude’s really good at saying, “You don’t need to automate this.” And if you kind of put that in your instructions to tell me not to automate things that don’t need automating, then you’ll be setting yourself up for success.

10:45 Crispin: That was the bit I was going to jump in on, Rob, because I remember having this conversation a while ago was let’s put that in the custom instructions then for both of us of like, do you really, now, do you really need to be automating this or are you just having a play? So let’s get on to where do we actually start?

11:06 Rob: Well, I’ll take this one because I’m more familiar with the setup process from the business perspective. And for me, the key understanding was that Cowork is your tool that lives on your computer. And that tool needs somewhere to store its brain, and that needs to be in a folder. And it can’t be a temporary folder, it has to be somewhere as if it were a sub-item in your work. So I would recommend putting it in your documents on your computer. It gives Claude a known space to operate within so that you can store your memory documents there, and you can put in that folder your templates and your reference documents and things that you may have uploaded to Claude Chat in the past. But in the folder is where you can expect the outputs to drop into, whether it’s your slide decks or your .md files. And once you’ve set it up, it is quite cumbersome to change it, so I would recommend having a little think about where the best place to put it for you is. Secondly, look at your connectors that you might be using in Chat. Look at what you’re using every day, where you spend most of your time. That’s probably going to be your communications and your diary platform, as well as your project management. We use Google, Monday, and Slack, and those are the most commonly used connectors that I’ve got set up in my Cowork space. One of the important things I think is to make sure that you are aware of how to use Cowork safely. Anthropic has some great resources on their website. In Claude Help, there are some good pages on how to make sure you’re going to set up and work with Cowork as safely as possible, making sure that you’re protecting your data, your privacy on your computer, as well as your business and your customer data. But I think the key is to start slowly. It’s okay to start with nothing, build up from there, take your time, and use Claude to help you design what you need. You can get it to interview you to explain a process and then keep it in the planning stage. You don’t have to build anything until you’re absolutely ready. And the process that I’ve used of asking Claude to look at an existing process, feedback, and then iterate and learn has really helped me build some really efficient, time-saving setups at The Coders Guild.

14:10 Crispin: Thanks, Rob. Do you know, like when you were talking there, one of the things that struck me was about choosing the folder, because I have seen, and I think I’ve seen this piece of content from probably half a dozen creators of, “I’ve got Claude Cowork and the first thing I did was get it to like file all my downloads.” And I was like, what, for instance, what an absolute waste of time. File your downloads into the different file types. Why is that? Downloads is like a super temporary folder, right? So they’ve either chosen like the top root folder which is, as you just said, not a great idea, you’re giving loads of access willy-nilly to everything on the system. And also, going back to our previous point of the previous slide, it’s like that’s not a useful thing to do, right? Even though it’s only a, you know, a half-second command, but what’s the, what’s the point? What’s the, what’s the point of that? If you’ve got so many things in your download folder that you need to, you need to get them organized by file type, you need to just trash everything in your downloads folder, surely.

15:18 Rob: Exactly, yeah. And it’s quite a lot of content churning out these, I wouldn’t even say they’re efficiency wins. It’s trying to get any old thing that they can advise people to do, but for us, it’s not, that’s not an optimization for us as a business, as an SME operationally. If you can’t look after your downloads, then you’ve got a bit more work to do.

15:43 Crispin: Yeah. And like you said, I think the, you know, the help files that are on Anthropic’s own site, the documentation around getting started with Cowork are great, and they’ve got really good, you know, they’re basically saying the same things as us here for, you know, choosing the right folder and being aware of that, of the security. But going back to what you were saying earlier, Rob, about, you know, turning the things on one by one as needed. And then, and you’re going to get, you’ll see in the demo like the different, when you look at what your, when you’ve got a connector, you can ask it, it can do stuff whenever you want or it can ask every time or it can be denied. There’s different security settings for just about every action that you could possibly do, and it’s really worth spending the time doing that and thinking about each use case of like, now I’m only doing this. And I think like for me, I’m letting it have stuff when there’s a, when there’s good reason for it. So I’m setting things, as a general rule, I’m setting things to ask every time until I’m annoyed that it’s asking every time. But I would go that way first, just so it’s reminding me what I’m giving access to.

17:08 Rob: I might be wrong, but also it can affect your token usage, I think, if you’ve got loads of connectors open and you’re not specific with your prompts. It can go off and check loads of stuff that you’re not aware about and you can use up your allowance really quickly.

17:25 Crispin: Yeah. Um, so onto, onto our demo. And what, do you know, one of the things that I love about this demo and particularly the thing you did here, Rob, is like, this is, this is us really eating our own dog food, as Julian, our nerd, would say, of like throughout all of the training that we do, like with all of the apprenticeships and not just the AI ones, we’re always trying to get them to learn by doing something useful in the business and also starting with something that you know well and using that as your learning. And that is such a powerful way. If you’ve got to figure out what it is that you need to do, that learning cycle is super empowering. It’s really, really fast. It’s basically the fastest way to learn stuff is figuring something out on a project. Like, say for example, you know, when you see a tutorial, you do a tutorial on YouTube or whatever, and you follow that through as a developer, as a data person, as someone who’s working in AI, you’re following someone’s example and it’s very systematic: “and then do this, and then do that.” Whereas actually you can, if you’re doing your own project, you know the creators who’ve got good tutorials. When you hit a block, you go and find a tutorial that gets you past that block, and as soon as you’re past that block, stop the tutorial, back to your own project. That is a super fast way of learning how to do complicated stuff. So, I don’t mean to talk only then, Rob, but what I want to ask you is where did you start with the thing? Where did you start?

19:07 Rob: Okay, so I found a bottleneck. This was a problem where we had multiple individuals using different platforms: spreadsheets, PDFs, emails, looking up things manually in Monday. It was a time sink for the team and multiple different people. So I found the problem. From there, I went into Chat, actually, because if you watched the last webinar you’ll have seen how we have projects set up in Chat for the specific business uses that we have that have all of our documentation. So I went into the Ops chat, I dropped in the context as to what the problem was, and asked Claude to give me an overview, three options depending on time and difficulty, and then I was going to take that project off into into Cowork.

20:17 Crispin: So you went into Chat first off, you didn’t start the project in Cowork, you started it in Chat. And did you, so when you’re saying you’re setting the context, did you do that as a, did you type it in or was that a voice note?

20:31 Rob: No, I’m quite lazy. I use voice notes. And also, I find that if you can dump a load into the prompt, it makes for a more efficient chat. So I prefer to use voice notes, although if people want to type, that’s fine. But I feel that if you can get more in that first prompt, you can get to the outcome faster. And our processes and the projects that we have in Chat are a much better way of understanding the business and getting to the outcome faster. And then once I’d got that prompt and the breakdown of the tasks and the connectors that we needed, I could drop it into a Cowork project and provide specific links to the Monday boards that we needed where the data was held. I could add the process documents into the folder, and then very quickly within five or six back-and-forths with Claude, we’d got the process set up. Claude actually built me a Monday board. I can actually show you our demo.

21:51 Crispin: Yeah, do you want to, do you want to go into the demo? I think one of the things I just wanted, while you’re doing a screen swap, the thing that I wanted to bring up was just to put some clarity around if you didn’t see the last webinar around our team setup. We have departmental projects, and that sets the context around that chat. So we’re already giving, so before Rob would start the chat with the operations and systems project, that already knows all of the, you know, it knows that we use Monday for a lot of stuff, it knows that we’re in Google Workspace, it knows about our business operations, it knows how we do process documents, it’s probably got a process document for the process that we’re looking to refine. So all of that means that Rob was able to do a really short voice note without pages and pages of prompting and adding loads of things to that chat. Both methods would work, but if you go back to the previous webinar, you can start to set up your various departmental projects, and that makes that a lot faster. Sorry, Rob, back to you.

23:01 Rob: No, that’s all right. So this is our kind of, we obviously couldn’t show you business-specific stuff, as Crispin said, so I created a fake demo commission report using a prompt in a project, and I think there were 11 back-and-forths. It was mostly about making sure it wasn’t using our business information. So I got Claude to create a commission tracker board in Monday. None of this information I put in, it created all of the boards, all of the fake learners and the fake dates. And then we were able to create the commission report which, ultimately, the output of that is a scheduled task. So we can look at the commission report. I probably won’t be able to run it live, but the actual report is set to run on the first of the month. And so by the time I’ve powered up my laptop and am ready to work on the first of the month, the report is already in place and the stakeholders can then review it in Monday. Oh, is it going to be fast enough? So at the moment, it’s looking up on the demo board what we kind of provided for it. And the output that we get internally is it drops the draft into my email, so it’s ready to go over to finance. And we’ve cut out the spreadsheets and the PDFs and the emails. And in Monday, we have the stakeholders signing off a status column. So all they need to do, the human in the loop, is double-checking the output of the report, and then once that’s all automated and I get the notification that the stakeholders have signed off, then I can send the email that’s already sat in my draft. Okay, it’s not happy with it. I can show, here’s one I made earlier.

25:22 Crispin: That’s fine, Rob. And do you know, the other, the points that I’d like to make in this, you know, if you, it’s not just about the time saved for you in operations getting that stuff, it’s also if you think about the context switching that everyone in that process is going to have to do to stop doing what they were doing, start thinking about their input into this process, and then stop thinking about that and go back to another thing. So just to cut all of that out because it’s not things that we should be worrying about, you know, we can do it so quickly. So yeah, not just about the time saved, but it’s also, when you think about efficiency, efficiency just isn’t time. Efficiency is also in like the the cognitive load that we have as people doing our job. And the more stuff we do, you know, when we’re looking for opportunities to create efficiency and to create or increase productivity, that’s one of the things that we’re looking for. It’s not just time and motion study, let’s automate this so that we don’t have to do that. You know, reducing the cognitive load for things like that that we shouldn’t be worrying about is really, really beneficial, not just to the efficiency of business but also like the well-being of everyone, right? Like, we’re reducing cognitive load, we’re hopefully reducing stress as well.

26:48 Rob: Yeah, and I’m cognizant of time, so I don’t think we’re going to have time to look at the session feedback reporting. But that is not just a time saver, that’s a 10x. It’s a 10x on what we’re able to do. It’s pulling things that were separately reported on, and if we can bring it all together and enrich the service that we already have. But we…

27:16 Crispin: I tell you what, Rob, I think we should just do, we should just do another webinar just on that session feedback, because what that is is a much more evolved process in terms of all of the things that go on in that Cowork setup. So I think that’s a webinar by itself, I’m sure. So let us know in the comments if you’re interested in a kind of a repeat of this but with a more complicated demo. Maybe we just go straight into the demo and go the full half an hour on that Cowork process, because I do, think that’s my favorite thing that we’ve got running with Claude Cowork because it is not, that’s not just a rehash of what we were doing with people, we’ve used AI there to absolutely elevate what we are able to do as a team.

28:11 Rob: And it is Cowork specific as well because it’s on a schedule. It’s not something you can do in Chat.

28:24 Crispin: Yeah. So, we need to, we need to like shuffle along, don’t we, and try and get to some of the comments. Rob, do you want to go and check to see if there’s any comments, any questions for us in LinkedIn? I’ll try and do a bit of a close-up. The way that this is working for us and the way that we’re seeing just the as a general thing of what are we doing is working with large language models and getting them to help us figure out what to do. So, bouncing things off of them. There’s a degree of setup, as we were saying with setting the context inside different projects or whatever. But once that is going, once you’ve got those custom instructions where we’re getting Claude to look out for us, you know, “Crispin and Rob, should you really be automating that or should you just do it?” like, you know, those sorts of things. What I will do, what I’ll do is I’ll create like a start a prompt for getting started with Claude Cowork. And let me know if you want that in the comments, and I will, I’ll DM you a starter prompt that you can put into Chat that will help you walk through a process that will help you in Claude Cowork and do the learning at the same time. How’s that?

29:48 Rob: I think, yeah, that’s decent.

30:51 Crispin: And of course, you know, we do, this is our job to train people with this. We do training that is aligned to business goals. If you want the government to pay for it, we have the AI and Automation Level 4 apprenticeship, what I wrote. And you can access funding through the Growth and Skills Levy to have people on your team trained in this sort of stuff and apply it to your business. It’s over a 12-month period, it is non-invasive into people’s time. It’s a really good way of having a supported, trained AI adoption process mapped out for you. Okay. Rob, do you have any questions?

30:35 Rob: We do. I’ve found a nifty tool on StreamYard where we can show them on the screen. “I run multiple businesses, do I need multiple Claude accounts?”

30:48 Crispin: Oh, that’s a good question, Kyri. It depends, I think. I used to have multiple Claude accounts. I had a personal one, and then we had our team one. I do have another business as well, Specs, which is a digital partner, digital advisory firm. Probably we will have a different one for Specs, I think, as needed. The team environment is the powerful bit, I would say. Like, Cowork is going to be Cowork on your machine, so that’s maybe your personal one. Yeah, depends, I guess. Like, as if you would have a different Google Workspace or a different Slack team, I guess. So, you don’t need it, but the need will present itself to you, I think. Like if you had one main set and you’re like, “Oh, I don’t want to to share this stuff.” But the sharing, the, you know, what we’ve got in the Claude team, that has different projects that are shareable to different people, a different team, so you could have, you could manage access to projects like that. And then if you’ve got like one master admin account, depends, I guess, on how closely related the businesses are.

32:10 Rob: Okay. “Are .md markdown files the most efficient for Claude to read rather than text, doc, or PDF?”

32:20 Crispin: Well, that’s a good question. I don’t know about most efficient to read. I feel like I really like markdown because I’ve been using it for a long time, you know, being an ex-developer. Lots of our stuff, in fact, the first version of the apprentice curriculum that we did back in like 2012, 2013 was markdown files on GitHub. Markdown is super efficient. That’s how skills are basically written. Markdown is good because you can get Claude to more easily update things in markdown. So I don’t know whether it’s the most efficient for Claude to read, but I think the most efficient for managing, I do really like markdown. You can get a markdown file and then translate it into HTML, for example, or translate it to a docx or Google Doc quite easily. So I do think it is a very portable format, and I expect we will all be all over markdown from this point on. I think it is, maybe it’s this half of the decade’s QR code, things to come into resurgence from 20 years ago and now be important.

33:43 Rob: I think David’s got a kind of a pair of questions. I think from the demo, “Is this email only a draft?” Yes. I think at this point Claude Cowork doesn’t send on your behalf. You’d have to bring in an automation tool like n8n or Zapier. I think he mentions in his second comment. So, yes.

34:06 Crispin: I’m fussy about that as well with the with all the stuff with Claude. I’m like, you’re not sending my emails, I’m sending my emails. But you can try…

34:15 Rob: I totally agree. And I think for for this use case, I want to have the final view before I send it anyway. So, yes, I could have run it through n8n and not have to touch it, but I just want to double-check. I think the human in the loop is something that we constantly talk about in our courses, and it’s still really important.

34:36 Crispin: I think as well, you know, there’s a point here, isn’t there, Rob, of like AI and automation as a thing that is that’s definitely connected. So if something can be automated, when AI is in there, I want to check it with that human in the loop, human in the lead aspect like we’re saying. But if it’s an automation, if it is just when this happens, send an email to someone, that’s just an automation, there’s no AI involved there. So for example, when, you know, if someone applies for one of our government-funded courses and they don’t live in the UK or are allowed to access public funding, we don’t need AI to make that decision. They tick a box that says, they don’t tick the box that says “I can access this funding” or, you know, we will understand they’re ineligible, that’s just an automation, it’s not AI.

35:21 Rob: Yeah. Ian Harding, “Any experience in integrating Cowork with HubSpot?”

35:26 Crispin: Um, no, because I don’t, we don’t use HubSpot, so, no. But I would, but I guess the thing is like going back to like make it work for you, you know, you could start doing that, certainly. Think of your use case. What’s your most, like with Rob’s advice, what’s your most simple use case first and only allow the the permissions that will enable you to run and test that use case, and then what’s your next bit and your next bit? So I wouldn’t, I wouldn’t think it was any more or less of a big deal than any other of the of the integrations and connectors that you can do.

36:07 Rob: Yeah, I think for me one of the highlights has been just how good the connection is with Monday. It’s our tool of choice in the business and it really, it kind of amplifies what you can do with the tool to to make things much easier to track and follow. So, yeah, I’m sure HubSpot, I know there’s an integration, and I’m sure it’s probably a very a very good setup, but you have to experiment with it a nice and cautious way because you are using other people’s data. I think we’re we’re out of public questions now, apologies if I’ve not seen any. There was just one that I wanted to flag that we had prior to. “Are there any connectors you wish existed?”

37:08 Crispin: Yes. Oh, it HTML code does. So, um, and I was talking actually on the Autumnus Clark podcast about this recently. There is an announced connection between Anthropic and Xero, which is going to make a lot of the financial reporting and forecasting, like I’ve always been, I’ve been for years going round trying to get the live reporting that I want to see in terms of forecasting matching to what’s being reconciled in Xero. So that, I’m just waiting for that MCP when that comes out, that’s like, right, I’m just going, I’m going dark for 2 days to get this done. And then that would be achievement unlocked for me. So yeah, very excited about the about the Xero and Anthropic connection.

37:59 Rob: Okay, I think that’s enough. We’ve we’ve gone way over, apologies for any waffling that may have been done. We’ll have to do another demo webinar on Cowork. We’ve got our AI for, Claude for Leaders coming up next month and we probably need to touch on Claude Code at some point. But thanks very much everyone, it’s been great to to finally get a bit more screen time in one of these webinars. And thanks, Crispin, for all your insight, really appreciate it.

38:31 Crispin: Awesome, and thanks for thanks for joining, Rob, instead of instead of hiding in the background making everything work. And yeah, any more questions, put them in the comments by all means, or any other content that you would like to see, let us know in the comments and we’ll do our best.

38:47 Rob: Cheers.