Most teams are using Claude like a smarter Google. Type a question, get an answer, close the tab. Every person starts from scratch every session. Different outputs, inconsistent tone, no shared context.
That works fine for individuals. It doesn’t work at a team level.
On 15th April we ran a free 20-minute session walking through how we actually set Claude up at The Coders Guild – not a tour of features, but the thinking behind it and the process we used to get the whole team working consistently from the same place. The recording is below.
The problem we were trying to solve
We’d been chasing this idea for a couple of years before it clicked. A central place where AI knew our business – our voice, our processes, our policies – so we weren’t re-explaining everything every single time. We tried a few things. Guru for a while. ChatGPT. A Notion setup with RAG retrieval. None of it stuck.
What changed was a combination of committing to Claude as our main tool and the introduction of skills – which made the kind of setup we’d been trying to build suddenly feel achievable. So we stopped running experiments and started building something properly.
Stop treating it like a tool
The shift that made everything else work was this: we stopped treating Claude like a tool and started treating it more like a new team member who needs a proper induction.
A new person joining your team needs onboarding. They need to know how you work, what you stand for, what’s in scope, what’s off the table. Claude is the same. The setup is the work. But once it’s done, you’re in refine mode – and the gains compound from there.
How we structured it: Ask Your Org
The starting point was what Claude calls the organisation-level knowledge base – we think of it as our intranet bot. Into that we loaded all of our policy documents, process references, and internal guides that we’d built up over the past couple of years. Expense policy, sickness policy, pension options, operational processes – all of it queryable in a single place.
Anyone on the team can now ask a question and get an answer grounded in our actual documents, not generic advice. It reduces errors, reduces the back-and-forth, and means nobody has to remember which folder something lives in.
One thing worth saying: don’t feel two years behind because we spent two years building those documents. You can start the other way – have a conversation with Claude about what’s missing, let it ask you questions, and build the documents together as you go. The knowledge base and the documents can develop in parallel.
Projects: organising by department and function
On top of the organisation-wide setup, we have a set of projects – more focused spaces organised by department and function. Operations. Delivery. Content and Campaigns. Business Development. Each one has its own knowledge base tailored to what that team actually needs.
The content project, for example, has our value proposition, customer personas, tone of voice guidelines, and the LinkedIn growth playbook we’re working to. So when anyone in that project is drafting a post, writing an email, or preparing a webinar, the responses they get are already aligned to what we’re trying to do as a business – without them having to brief it from scratch every time.
You don’t have to organise projects the way we have. You could create one per client, per campaign, per product. The structure should match how your business actually works.
Talk to us about your team's AI Training
Not sure which route is right? Book a free discovery call and we’ll point you in the right direction
Francesca Roberts
Head of Commercial
Skills: how you teach Claude to work your way
If projects are where you share context and information, skills are how you teach Claude to do tasks the way your organisation does them.
We’ve built skills for things like our SEO and AEO writing approach, our scheme of work format, the compliance framework we operate in, brand messaging, and a commercial legal helper for when we need to review contracts. There’s also a presentation builder skill built around how I personally think about and structure a talk – so when I’m pulling that into a project, I’m getting responses shaped around my actual approach, not a generic outline.
The skill creator tool that Anthropic released recently made this significantly easier. You can now have a conversation with Claude about how you do something, answer its questions, and it will draft the skill file for you. You refine it, push back on the bits that aren’t right, and iterate until it’s accurate. Then it goes into the shared organisation space where everyone can use it.
The result: Claude applies your standards consistently, without anyone having to explain them every time.
Where to start
If you watched this session and want to do something with it today, the suggestion is simple: open Claude, create a project, and start a conversation about what it should contain. Bring in the people who’ll use it. Take the transcript of that conversation and feed it back into Claude to help you refine the structure.
Don’t try to build everything at once. Start with one department or one use case, get it working properly, then expand. As you go, ask: is this something that could become a skill? Is there a process here that needs its own project?
The setup is the work. But it’s a one-time investment that pays back every week after.
00:10 Rob: Good afternoon and welcome to another quick bytes session. Today Crispin’s going to be showing you how we use our Claude at The Coders Guild as our go-to AI hub. We’ll leave a little bit of time for a Q&A at the end so please get your questions in via the LinkedIn comment section this week. But I’ll hand over to Crispin now to give you a run through our Claude setup.
00:35 Crispin Read: Thanks, Rob. So today I’m going to show you how we turned Claude into a proper team hub. So not just a chatbot that people are dipping in and out of but something actually knows how our business works and how our team works, including specifically the bit that most people skip that makes everything else fall into place. So demo should be about 20 minutes. As Rob said, please drop questions into the comments as we go. Francesca and Alex from our team are also in the comments so if there’s stuff that they can answer directly they will. And I’m not going to monitor the comments cuz I’ve too many I’ve got too many moving parts already. But yeah, we will Rob will try and get some comments together for for a Q&A at the end.
01:30 Crispin Read: So where to start? Where to start? This is actually kind of the thing that we would have been dreaming about and trying lots of different ways of developing for for a couple of years. Now it wasn’t in Claude specifically, but like the idea behind like this central place where the AI can um where AI helps us do our work, doesn’t replace um people, but helps us all work more efficiently together. You know, a place with that knows our business, knows our voice, our processes. So we’re not explaining everything or re-explaining stuff to each other every single time. We tried a few things before. We um we were doing a thing, we had a um a thing running in Guru for a time. Um we did some stuff with ChatGPT. We um started creating our own like um we had a thing going in Notion as well, and we were going having like RAG retrieval, um which is like um kind of like a search but we were going and grabbing the specific bits of documents that that are that are relevant.
02:43 Crispin Read: So we kind of, you know, we were always kind of like moving around trying to get something useful together, um, and we kind of knew what we wanted, we just couldn’t make it stick. Then earlier this year a few things started happening. Um, we’d, I’d already moved on to, uh, onto Claude at the end of or like midway through last year kind of, um, as my dominant LLM of choice. Uh, and I think like as that, um, as sort of like the system behind how Claude works for, um, for people and for teams as as an application, as that started developing more like towards the end of last year and the beginning of this year with, um, with skills in particular, um, that kind of became really really appealing to the sorts of things that we were trying to do. Um, so then we we moved over to to moving making that project kind of work, that team project work with Claude rather than the other things that we had trial. So what I’m going to what I’m going to show you today is what we built that works for us, but really in more general terms what I’m talking about is the principles behind what we did as a team to make that effective.
04:02 Crispin Read: And and this I think that’s really important because like if you want AI to work consistently across the team, not just for individual silo people who are who are doing really well at at something, if you want to you know you want to get the team benefits out of it, we you kind of need to stop treating it like a tool and start treating it more like a a team member who needs proper induction, needs onboarding. And and I guess that’s that’s where we’re at. So we’ve got um the the Claude interface here. Let me just check that I’ve got the yeah I have got the right one. So I’ve got the um I’ve got the desktop version here, um which has which is attached to um co-work, uh it’s also got Claude Code as well. Um it’s a very similar interface to the to the web one, if you’re more used to the to the web version it’s it’s always the same. You know, we have our recents and stuff down here, we have our main uh interface here.
05:05 Crispin Read: Um the the things that uh that are kind of different from ChatGPT I guess, you know, we’ve got these like connectors here that will give us some prompts about oh how can we start working with Google Drive that I’ve connected or from the calendar or from uh from Gmail. Um I’m not going to go into those. I think that Claude will take you through that when you um if you if you want to just if you want to ask it because I’m trying to stay focused on the team aspect today of of what we do. We’ve got some other webinars coming up around uh getting started with co-work and also uh getting started with uh Claude Code and um as I always as I was saying in uh one of my chat groups recently, like when you when you when you get Claude Code up and running and then you can start to move into the command line interface it’s really empowering. You feel you feel really clever, you feel really uh bleeding edge. So um more webinars about that. If you want to be notified of them let us know in the comments um um we’ll be able to ping you the links or you know like follow subscribe uh as the as the motto goes.
06:09 Crispin Read: Um so let me just grab my find my notes. Too many windows open. Too many windows open. Uh Right, so um here is our out-of-the-box thing. The The This um Ask Your Org, I must be in the wrong Oh, here we go, Ask Your Org. They’ve just changed it. The The update came through and that threw me because that used to say Ask The Coders Guild. But here’s kind of where I want to start um the story. So this is the thing that we were trying to that we’ve been trying to create for ages and that when we got Claude for the team, it just came out of the box. So this is like our like a chat-based interface for our intranet. Um and so here you can see like loads and loads of files. What we’ve got in in in this, we’ve got the custom instructions, you know, similar to to ChatGPT or similar to like a personal um instance of uh of Claude, you’re giving your custom instructions that are going to tell um Claude, give Claude context about how you want it to respond, the sorts of things you do, um and all these all the things that I show you today around the the files in the knowledge base, the custom instructions, the memory, um and skills, what they do is build context around what you’re asking for. And then what that is doing is basically refining what sort of response you’re going to get. So you’re putting all of these things into every um every prompt that you’re putting into the chat, do you see what I mean? And so the Ask The Coders Guild um intranet bot as as it is. So we’ve put into here into these files, into the um here you can just add files in here, it’s very easy upload from your device or you can just do a uh you can paste in stuff, you can add stuff from Drive. And what we’ve added in here is all the policy documents that we have, internal and external, all of the reference documents that we’ve been creating, all of the process documents that we’ve been creating over the over the past couple of years. Um and because we were always trying to do this thing, we like came into this marching you know running really really fast because we already had these documents.
08:41 Crispin Read: Um now what this means is that anyone can come and ask a question of this particular um chatbot, and it knows our policies. It knows our expense policies. It knows our sickness policy or um the options for um for pensions. And all of these things are now contained in this one place so we’re not giving generic um advice to people, we’re not telling people to go and look inside certain documents. They can query these documents, and that’s what the custom instructions do is is make put these documents down as the as the gospel, and reduce hallucinations and reduce um opportunities for error. So this is the thing that was our like real starting point of like, oh great, now we’ve got this. And the the team part of it, so there’s there’s like an amazing advantage right there. So, oh, and one thing I should say is like, um don’t feel like you’re 2 years behind cuz we just spent 2 years building up this library of process and reference documents. You can start to go in and build this and walk through what’s needed, and then and enter into this sort of like question answers conversational um relationship where you’re asking, what else could we put in here? What’s the sort of uh what what do you what’s missing from your knowledge base? Oh, I could do with a um I could do with an understanding about this. Let’s create that document together. Ask me some questions, you’ll get that document created and then it can go uh into the knowledge base.
10:13 Crispin Read: So similar, when we’re looking at projects. So projects are then more specific. Um so we’ve got a bunch of projects here. So we organize ourselves uh at The Coders Guild, we’re organizing ourselves basically by in three departments. So we have um operations, that’s like our operational activity, then we have delivery, which is um, you know, delivery of training and services and uh and that sort of stuff, and then we have Comms, Biz Dev, um sales as a third. So we but we have more than more than um three projects, so they’re kind of departmental in like Operations and Systems is our is our ops one, delivery and learning is the delivery one, um then we’ve got um, for example, Content and Campaigns and Business Development, they’re separate because they have um different things in the knowledge base. So if I pull up a if I have a look quick example of our Content and Campaigns, so the knowledge that I added to to that particular project was about our value proposition, um customer customer customer personas, uh value mapping, um segmentation, um we’ve got stuff about our tone of voice, um things that we’re things we’re working in there about um optimizing our LinkedIn profiles, um the LinkedIn growth playbook that we’re that we’re operating on. So that gives the context for this project immediately about um The Coders Guild, what we are trying to do as a business, what we’re trying to do in terms of the content that we create, and uh and how that’s all related into into into other things. So all of us are now referring to the same documentation. So when we are um looking for help writing emails or um or for this one, there’s more like creating content on uh on LinkedIn or preparing for for this webinar, for example, immediately the responses that we’re getting out of this project are aligned to our um business objectives and the way that The Coders Guild works.
12:35 Crispin Read: The memory part up here is good as well. So once we start working with this, the memory starts off as blank when you start a project, but then it’s filling stuff in as you are giving more and more context, it’s picking up on the on the repeated things and then that memory gets updated uh as you go. The knowledge base here as well, that is also that’s live. So if you want to um delete something or or change something, you just delete it, change it and then the next chat goes from that new knowledge base straight away, it’s not bringing in that that legacy stuff. So you can see now with the with the projects we’ve got our departmental stuff. I got some um personal stuff in there as well, so there’s the with the stuff I’m writing some books at the moment, so there’s some stuff helping me um write and edit and prepare um stuff I’m doing with the books. Um there’s like a little baking project there that all that oh, that would be a cool webinar as well, so I’ve got the in the baking project I have a thing that’s going Claude to Notion direct with loads of different ratios and updates on on recipes and stuff, which is um which is really good that I can just have like a pal as I’m um as I’m baking and I can just do a voice note and it’s updating um recipes and uh and ratios and stuff in the background. But that’s not what this one is about, this is uh this is about the team stuff.
14:00 Crispin Read: Um I’ve got a a demo um project here cuz I would I would say that whilst I said, you know, that the biz dev um project and the content project are different, they’ve got different knowledge in there, they’ve got different um parameters, um you could create a project for a particular customer. You could create a project for a particular project that you were doing. You know, you group them how it’s useful for um for your company, for your team, for your business. Um so in the getting started with the projects, I think that your best place to start is to have a meeting with uh with people. What we did, we started like a pilot group of uh Francesca and me and Rob. We had a meeting about, oh, how would we do this setup? What would what do we think would be useful? We took the transcript of that meeting and we put that into Claude and said look, this is what we’re thinking of doing, ask me some questions about how to refine this. And we entered into like a a one-to-one, in like a back-and-forth with Claude to help us refine how how those projects look. So you can do that follow that same process and you would get different projects with different titles and different um knowledge modules, of knowledge modules, different knowledge bases and and so on for the that match your business.
15:40 Crispin Read: Um now the other thing that I would like to to talk about now is the is our skills. So if I come to customize here where the skills live, just close up my So we’ve got personal skills and we’ve got organizational skills. So we’ve been building these out and refining them over time. And the skills um part of Claude I think is what makes it super super useful. So, um if you think like a project is your collection of how you are collectively sharing um information and data and chats about a particular thing, an area, um then the skills is how you do tasks associated with that. So if you think about it like you’ve got your projects and your skills, you can call skills into any project. So um if I talk about some of the skills that we’ve put out here. So presentation builder, so this is a very apt one, this is one that I’ve been working on um quite recently and like, um and this is a one that’s personal to me because it is how I build and deliver presentations. So to get a back and forth, um I ask myself like, who are these question, the questions about who the audience is, what do I want them to get out of it, what’s the um what’s the action they can do right now, do you know what I mean? So this is how I try to think about how I’m doing presentations, so I get by pulling in that presentation skill into the content project, I can then have a one-to-two with Claude that is around these parameters that help me build a presentation that is accurate to my voice and the business objectives of The Coders Guild.
17:26 Crispin Read: Um we’ve got other ones here about like, you know, with um things about our SEO and uh AEO writing, um an interesting one here for a scheme of work, we have a very specific way we like to build our schemes of work for courses, this is a completely internal thing but, um you know, this is something that we worked on and refined over over time. But now Claude knows this. So for Claude to draft up schemes of work that match The Coders Guild’s way of creating these documents that work for everyone and everyone always understands it, that’s a that’s a sped-up process now. Um we have stuff in here about the compliance framework that we operate in, we’ve got like an operations reference, we’ve got our brand messaging, um got a little uh commercial legal um helper here for if we need to um engage in a in a legal way with uh with people, it gives us advice, not proper legal advice I would say, but you know it helps um frame our arguments coherently in uh with those sorts of things.
18:40 Crispin Read: So this is by no means like an exhaustive list, this is as far as we have got so far with uh with skills. Um they they can also describe like how to create documents uh and things like that. And what I’d like to do now if I go back to the whoops back to the project area where I’ve got this demo, what I’d like to quickly run through is this example here. So I set up this demo um project that is for a fictitious uh consulting firm, and then I went through this chat to show you how I would build out a skill. So, um imagining that I’m having a conversation in the in the operations department of uh or the operations project, and I think, oh, this is something that could be repeated, like onboarding or or this one is like a proposal creator, so oh, how do we how do we write proposals? Ask me some questions, and then so I got some questions in here, then I answered those questions, it had some more questions and I answered some more questions, and then that like it is you can see up here we have um invoked the skill creator guide which is uh which is kind of new from um from Claude, so that understands how to create skills. And then it’s given me this markdown file, um now so based on just those two conversations that that we have had, I’ve built out this skill now. So then what I would be doing is I’ve been reading this and thinking, oh no actually that’s wrong, oh we don’t want to do that, oh it would be better if this, so then I would engage in the in in dialogue now back with this chat, and then re- and then get that skill up um to the standard that I want it. Then we’ve got this the save skill here which will save it to my personal skills if it was a personal one. If I wanted to um add it to the organization as as a skill, I would download it, oh you can add it to project, um you can download it and then what I have been doing, this might not be the right way to do it go because seeing as uh I just updated Claude, um silly, 5 minutes before the the webinar started, um but what I have been doing is downloading that skill file which Claude has its own um like .skill um file, and then uploading that as a shared resource into the into the shared domain.
21:12 Crispin Read: Um what am I at? 22 minutes in, okay. Um so I think that basically I think that is that’s all the things I really wanted to cover in this session. I like I said, like I’m happy to go more into uh skill building, which I’ve got a um a short piece of content to talk more um about that later on. But the co-work aspects of Claude we’re finding um really powerful, and indeed uh the Claude Code stuff like not we’re not using Claude Code so much to write code, but in order to um have more structure, more control, um more granularity around the skills and projects, we’ve got some stuff running in code and I’ll be I’ll be sharing those in uh in future webinars. Um I guess closing notes is um I think like everything that I’ve just shown you comes back to the same thing, Claude is only really or any tool like this is only as useful as the context that you are giving it. Uh the setup is the work, um but then once you’ve done the the first bit then you’re into refine mode, uh and your your gains are compounding. So if I if I want something to come out of uh you watching this webinar today, open Claude, create a project, um and get stuck into a dialogue with with Claude about what you should do next, and and and be clear about involving the other people in your team. And then Claude will help you build out that project to to be something super useful, and help and we always be thinking, where is this a skill? Is this a new project? Um and you know and ask for ask for that help. Um yeah. That brings around to kind of to the end. Um so I’m hopefully there’s uh some questions for me to answer in uh in the comments. Um anything else to say? Um I can’t remember. Rob, give me a question to answer instead.
23:15 Rob: I’ll chuck some questions at you. Uh we’ve got quite a few, but you may have kind of been been handling them on the fly. Um uh first of all, what’s the difference between co-work Claude and Claude Code? Just for those that aren’t monitoring the chat cuz Francesca did reply to this in the chat.
23:31 Crispin Read: Okay, so so basically what I showed you there is Claude, the Claude AI, the the interface to talk to the large language model, so that’s where you’ve got projects and you’ve got skills, um and those types of things. So co-work is something that you need to um use on a desktop computer, so it’s not in the browser, on a desktop computer, then you are giving Claude co-work access to your email, your email, your folder structure, you can get it to do stuff in um on your computer, so that’s where you start to get more um like agentic um flows going with with things that can be repeated. Then code, um that I think the co- Claude Code that is a that is another webinar, that’s not something I can answer but um in general, Claude Code is about writing code, but what started happening at end of last year is people working with Claude Code to do non-code stuff, and basically that’s what developed into Claude co-work was the things people were doing with code to write little bits of code that do little agentic things, and now co-work is essentially an a user interface on top of that so all of the code stuff is hidden. Hope that makes sense.
24:52 Rob: Yeah, that was great. Um um how do we handle versioning of of skills, um I guess that’s another one. Well, so here’s where we are going into, as I mentioned like the stuff that we’re venturing into with with co-work and code, so version control of the skill files in GitHub. So as so then holding all of those things. So that is a Claude Code version of the things that I just showed you there, so you would have a repo that had all of those skills and um you know access to the different bits of knowledge, and they there’s your version control.
25:32 Rob: Um uh am I right in thinking uh you can use ask Claude to help you create the skills? Yes, we’ve you covered that.
25:38 Crispin Read: Oh, yeah, absolutely. And like you know, that is the uh and that’s that’s quite recent, it’s only um a number of weeks ago that they released that skill creator skill, and that was a a massive jump forward in in how easy it is to to to create skills. And I think just a really good example of like how how this stuff is evolving. The things that held us back before were the effort in time and the expense in time and money to learn a new platform or a new um piece of software or a new language, and now that’s the easy bit, you know, this is you know sort of maybe a little bit more difficult than Instagram but, you still got to understand that you got to do the the work as well, you know, you’ve got to understand the you’ve got to give it the a good context. If you give it uh a half-assed context, you’re not going to get great results, and that and it’s about putting that work together.
26:44 Rob: And I think one thing I’d like to add is that because of the way we uh have a history in the in the business of owning process documents and uh policies, and with skills, those are going to be updated by the admins and also it’s an iterative approach so you can ask um at the end of a chat, uh what elements of this can be used to improve the skill, what can be used to improve the project, so um uh uh I that would be uh owners of the the skills and the documents and uh updating as you go uh to make the best use of it. Um right, sorry, I’ve lost track of the questions. Um does everyone in the team have access to the same Claude account or does does everyone have separate accounts?
27:32 Crispin Read: So, uh well, at the moment as it as this training, I’ve got some chats in my old personal pro account that aren’t that are that are still needed so I have two. However, everybody so we have a team license and then and then everyone has access to that. So they have access to different things, they like so only I can see my personal projects, like the book one and the baking one are just mine. Um and then all of those projects are then um you’re giving access to them of people who can use them, people who can edit them, uh and it’s and it’s very very simple, you know like that that’s the level, it isn’t um layers and layers of uh of admin and uh and privileges.
28:19 Rob: Um we’ve got uh, how do you edit the .md file? Um that’s, you get Claude to do that as far as I’m aware.
28:27 Crispin Read: Yeah. I I, generally it’s a voice note for me of like when I say, can you update this skill file, what I want you to do is this, we should be doing more like this, blah blah blah, show me the result, and then and then I’ll and then I’ll give feedback to that, but you can just you can just edit it in place if you if you want to.
28:46 Rob: Uh for those in team b- team businesses, does anyone have any recommendations? I think this is probably for the wider chat, but have we got any recommendations for useful skills um either that we’ve built or that we’ve um brought in?
29:00 Crispin Read: Uh I think if you can look at um Anthropic’s GitHub, they have a bunch of skills that are there that you can look at, but you um Claude, even the um the web interface has access to those skills, so you can just say are there any skills um for this sort of thing and, you know, you can you can, the more value I think it comes in when you give it your proper context, so you can start from um, you know, the generic um Anthropic SEO skill but then you want to add to that with your specific um you know take on it.
29:40 Rob: Okay, we’re we’re a little bit over, but um uh does Claude automatically reference skills uh and the files you’ve uploaded to the project, or do you need to prompt every time?
29:51 Crispin Read: Uh depends, it’s pretty good at when you ask it something, so if I if I’m saying I’m going to do this, it’s going to, oh I’m just going to read the Crispin’s voice skill, so it it’s pretty good at that, but then if you get a funny result, then that and often I will go and say, do you need to look at that skill? Oh, yes, of course, I’ll go and pull that skill in. So it’s it’s kind of uh a bit of both.
30:12 Rob: I think the fur- the longer the chat goes on as well, the the longer the context window you might need to refresh. Um I think we’ll uh we’ll have to to call it time there, um the only other thing we’ve got is the upcoming webinars. I know Francesca has been adding um you to the to the commenters to the to the list of of invites for the upcoming uh webinars, and we’ll uh do a deep dive into co-work this time next month so make sure you sign up. This is our um Linktree link which will give you um the links to uh all of the events coming up, and uh if you can sign up via LinkedIn or Eventbrite then we can make sure that we’re reaching out to you directly to give you any additional information that you might need. And uh a little bit over over-due, but um thanks very much, Crispin, that was really good. Um and uh if you have any questions um please keep them coming in into LinkedIn, uh if you’re rewatching at a later date, um and please get in touch with the team, Francesca and Alex are available anytime you need them.
31:18 Crispin Read: Brilliant. Thanks a lot.
31:20 Rob: Cheers.
