Most businesses using Claude are getting results. Just not results that feel like theirs.
The difference between those two things is not the model you’re on or how well you write prompts. It’s whether Claude has been set up to understand your business — and for most, it hasn’t.
This post covers what that setup involves, why it matters commercially, and where to start. There’s a full session recording below if you prefer to watch, but the post works on its own.
The “I’m not a techy person” argument has run out of road
Early in my career the leaders above me would always be distancing themselves from the work the technical people did. That was the techie people’s job. I’ve heard variations of this my whole working life, and for a while there was some logic to it. The technology involved was genuinely complicated – multiple coding languages, server infrastructure, things that took years to learn properly.
But the tech has been getting easier for 30 years. PowerPoint has been around that long and people still create documents in it without ever learning the basics. Not because it’s hard. Because they decided it wasn’t their job.
This one is different. AI affects every person in your organisation, every process, every decision. You can’t hand it to the technical team and stay above it. That said, you don’t need to know everything. I’ve led excellent engineering teams as someone who was never the best developer in the room. Understanding enough, asking the right questions, trusting the right people – that still works. Opting out entirely does not..
The commercial case
Anthropic published research showing the gap between where AI could be applied across business functions and where it is actually being used. The gap is significant. The research on productivity gains is consistent – between 10 and 55 percent is available when AI tools are properly implemented.
Even at the absolute floor of 10 percent, a business turning over £1.2 million has around £120,000 of recoverable capacity per year. Not headcount reduction. Recovered capacity for the things that actually grow the business.
Most businesses are at one or two percent. The gap between that and ten is not about having access to better tools. It’s about setup.
Context engineering — what it is and why it matters
Context engineering is the practice of loading the right information before a conversation starts, so what comes back is specifically useful rather than generically capable.
When you open a chat with Claude, it reads everything in the Project you’re working in before you type a word. That means if your Project contains well-written documents about your customers, your voice, your team, and how your business works, Claude already has that context. What comes back reflects it. A couple of years ago we were writing pages-long prompts to achieve this manually. What exists now is a cleaner way of doing the same thing.
Claude has three mechanisms for managing this context.
Memory is personal and passive — it builds from your conversations over time, useful but not structured. Skills are intentional documents you write once: how to draft a proposal, how to follow up after a sales call, how your business handles an inbound enquiry. Knowledge is static information Claude needs to make good decisions — who your customers are, how you make money, what your team looks like. Projects organise skills and knowledge so the right context loads for the right piece of work, and so your whole team draws from the same foundation rather than managing five separate versions of the truth.
Webinar Demo – Claude as a VA
The demo I show in the session — a VA dispatch skill that creates tasks, drafts emails, finds meeting times, and updates notes all from a single voice note — is only possible because of that foundation. The skill itself is fairly simple. What makes it work is everything underneath it: the connectors set up correctly, the knowledge and skills already in place, and the team setup done properly. Each of those is a separate session in this series — links at the bottom of this post.
AI Tools for Small Businesses
A practical short course for small business owners and their teams. No technical background needed.
Where to start
The question that does most of the work is: what does Claude need to know about me to be useful? The answer breaks into things specific to you as a person, things specific to your business, and things specific to a particular project.
The mistake most people make is writing long, comprehensive documents. What works is short, specific, and opinionated. A two-page customer profile with a genuine point of view is worth more than a ten-page brand handbook.
In the session I promised starter prompts. Here they are. These are not templates — they are conversation starters designed to get Claude to ask you the right questions rather than produce something generic. The key principle in all of them: do not produce a draft until you have enough real information to make it genuinely useful.
Your voice
I want to build a voice and communication style document to save as a skill in my Claude Project. This will help Claude write in my voice rather than its own default style.
Before you write anything, ask me questions. I want you to understand: how I write differently for different audiences and contexts (LinkedIn posts, emails to clients, internal messages to my team, formal external letters), the phrases and words I actively avoid, my general tone and register, and any specific things I want Claude to always or never do when writing as me.
Do not assume anything. If an answer is vague, push back and ask for a specific example. Keep asking questions until you have enough real, specific information to write something genuinely useful — not a generic style guide. Do not produce a draft until you are confident you understand how I actually communicate.
When you are ready, write the document as a practical reference in plain language. Include: tone and register by context, phrases and words to avoid, and specific examples of good and bad output. One to two pages maximum. Write it so Claude can make real decisions from it, not so it reads like a brand guidelines document.
Your ideal customer
I want to create an Ideal Customer Profile document to save as knowledge in my Claude Project. This should help Claude understand who actually buys from us, so it can make better decisions when helping with sales, marketing, and communications.
Before you write anything, ask me questions. I want you to understand: who our best customers are and what makes them our best customers, what problems they come to us with and how they describe those problems in their own words, what their situation typically looks like before they find us, what makes them choose us over alternatives, and what types of company or person we have worked with that turned out to be a poor fit.
Do not assume anything. If I give a vague answer, ask me to be more specific. If my answer sounds like marketing copy rather than a description of real people, push back. Keep asking until you have enough real information to write something useful. Do not produce a draft until you are confident you understand who we actually serve, not who we wish we served.
Write the document as a practical one to two page reference. Use plain language. Avoid marketing-speak. This should help Claude make decisions, not describe aspirations.
Your customer value proposition
I want to create a Customer Value Proposition document to save as knowledge in my Claude Project. This should help Claude understand why customers choose us, what we uniquely deliver, and how we frame our value for different types of buyer.
Before you write anything, ask me questions. I want you to understand: what we deliver that alternatives do not, how our customers describe the value they get from working with us in their own words rather than ours, what objections or hesitations we most commonly encounter and how we address them honestly, how our positioning differs for different buyer types, and what we would never claim about ourselves.
Do not assume anything. If an answer sounds like a marketing claim rather than a genuine description, push back and ask for something more specific and real. Keep asking until you have enough honest, grounded information to write something Claude can actually use. Do not produce a draft until you are ready to write something specific.
Write the document as a concise, opinionated one to two page reference. Plain language. No marketing copy. Include: who we serve, what we uniquely deliver, why customers choose us, how we frame value for different buyer types, and common objections with honest responses.
How your business makes money
I want to create a commercial model document to save as knowledge in my Claude Project. This should give Claude a clear picture of how our business works financially, so it can help with pricing conversations, proposals, and commercial decisions.
Before you write anything, ask me questions. I want you to understand: what our products or services are and how they are priced, who makes buying decisions and what their typical budget situation looks like, what the sales process normally looks like from first contact to signed agreement, what our most common revenue streams are, and where the business is currently focused for growth.
Do not assume anything. If I give a vague answer, ask for specifics. Keep asking until you have a clear and accurate picture of how we actually operate commercially. Do not produce a draft until you have enough real information to make the document genuinely useful.
Write the document as a short factual reference. No fluff. This is not for external use — it is a working document to help Claude make informed decisions when I am working on anything commercial.
Your team and how you work
I want to create a team and working style document to save as knowledge in my Claude Project. This should help Claude understand who does what, how we communicate, and how I prefer to work, so it can route tasks, draft communications, and support decisions appropriately.
Before you write anything, ask me questions. I want you to understand: who is in the team and what each person’s role and responsibilities are, how we communicate internally including the tools we use and what is expected, how I prefer to receive information and make decisions, what external partners or agencies we work with regularly and in what capacity, and any working patterns or preferences Claude should know about.
Do not assume anything. If something is unclear, ask for clarification. Keep asking until you have enough to write something genuinely useful. Do not produce a draft until you are confident the document reflects how the team actually works.
Write the document as a practical reference. Include names, roles, and context. Keep it specific and factual. One to two pages.
Block out an afternoon next week.
You will feel the difference the same day, and the returns keep building from there.
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
Questions from the session
We had some good questions come through in the Q&A. A few worth sharing here.
How do you share skills across a team?
Skills can be personal, shared within a specific project, shared between specific people, or organisation-wide. In Claude’s customize section those tiers are clearly visible. The organisation-wide layer is particularly powerful — when you have a shared way of doing things, you update it once and everyone draws from the same foundation. At TCG we update skills from meeting transcripts as our processes evolve, so the knowledge base improves as the team does.
How long does it take to set up a VA?
My current setup has been an ongoing project for a couple of years across several different systems — Telegram, Whisper, n8n, a RAG setup in Notion. The current version took about an hour to pull together inside Claude from everything those previous iterations taught me. Starting from scratch now, I think an afternoon is realistic for a small company to get a solid basic setup in place. It is never finished — it evolves as your business does. The best thing you can do is start.
Do you host your skills externally or keep everything inside Claude?
For our day-to-day team setup, everything lives inside Claude. We’re moving toward storing the skills and knowledge files in GitHub so we retain ownership externally and call them in when needed — but that’s a later step. For most businesses starting out, getting set up inside Claude Projects is the right first move. The architecture can get more sophisticated once the foundation is solid.
Any good resources for people starting out?
Anthropic have a strong learning section on their site — link below. I’d also point people toward this webinar series: the team setup session and the connectors session both build directly on what’s covered here. There are good creators on TikTok and YouTube sharing worthwhile practical content — I’ll add recommendations in the comments.
Creators worth following
Matt Wolfe — @mreflow on YouTube, TikTok and Instagram. Runs FutureTools.io, tests AI tools against real business workflows, and is honest about what works and what doesn’t. One of the few people in this space who shares his actual setup rather than just reviewing products.
Sabrina Ramonov — @sabrina_ramonov on TikTok and Instagram, and at sabrina.dev. Sold an AI company, Forbes 30 Under 30, teaches AI practically and for free. Her content is aimed at people who want to build real systems, not just follow trends.
Follow me on LinkedIn
I post regularly on AI adoption for growing businesses — specific tools, what’s actually working, and content from sessions like this one. If you want to stay across it, the easiest place is LinkedIn.
The rest of this series
The VA dispatch skill I demo in the session is only possible because several things are already in place — the connectors are set up, the knowledge and skills files are built, and the team setup has been done. Each of those is a separate session.
Sessions already available on the site:
- Setting up Claude for your whole team
– projects, skills, knowledge, how to organise your team’s shared AI environment - Getting Started with Claude Cowork
– basic set up. What to try, What to skip. What we did.
– Automation work and connecting to multiple tools - How to get your whole team using AI effectively
– people > process > tools
– what’s getting in the way
– our rollout plan
Coming up:
- Connecting Claude to the tools you already use — how to set up Connectors safely, what permissions to give, what to watch out for
- Claude Design – Creating a design system that Claude can use to draft on brand slides, docs, print materials – and even animations
- Claude Code for non-developers — Step up from the web interface and CoWork and start using the CLI
Sign up for all upcoming sessions in one place:
Join our free AI webinars
Practical AI topics, no jargon, free to attend. New topic every couple of weeks – always something you can actually use.
Building a Claude assistant that knows your business
18 Jun 2026
---
[Transcript to be added][00:00:00]
Rob Brighton: Good afternoon and welcome to today's live stream. In this session we're talking about building a cloud assistant that actually knows your business, not a generic chat bot, but something trained on your context, your processes and your way of working. Crispin Read is going to walk you through the fundamentals of building a skill effectively so you can come away with
[00:00:31]
something that you can start building and using straight away. We'll have some time for Q&A at the end hopefully, so drop your questions in the comments where that's on LinkedIn or YouTube today and we'll try to pick them up for the Q&A at the end. Time for me to disappear and Crispin Read over to you. Thanks a lot Rob. Yeah Rob's been cautious about the Q&A because I just realized that I haven't timed this but I'm definitely hoping to get done in 20 minutes so we've got some good time for the questions. The question is always kind of the best bit in this webinar series.
[00:01:06]
This is what we're looking at today, is building a cloud assistant that actually knows you and your business. By the end I'll show you the skill that I've built up that I can just voice know and it can be a drafted email, can add task boards, add notes, adjust notes, find space in people's canons, can do loads of different stuff. That's kind of the end point and I want to stress that this is not like a setup for you to copy. I'm going to show you how I built my VA assistant
[00:01:43]
so that you can build yours. So we need to see a lot of people on Instagram and TikTok and what have you and they'll say, comment this and I'll send you the skill. Those things are again, they're generic and they are and the purpose is to get you to comment and to boost the numbers. So this is a bit more involved because it's about how you do the actual work to get your one to work.
[00:02:13]
And as ever my mantra of people process tools in that order. In this case people is you so we're going to kind of skip straight away to process and documentation and that's where this all comes to par. So I'm not a techie person. This is something as a techie person myself and this is something I've heard a lot my career, especially early on. The leaders above me would always be distancing themselves from the work that the technical people did. That's the
[00:02:47]
techie people, that's their job. I don't have to do that stuff because the techie people do it. And that's a weird thing. I always thought that was weird, of course, because I was like elbows deep in all this stuff and I didn't think it was so difficult. But over the last 20, 30 years, the tech has become ever more and more easy and accessible to learn. But still in some cases, people just haven't bothered. PowerPoint is what 30 years old or whatever and Google slides
[00:03:18]
and people still and word, people still create word documents without semantic headings and without proper word processing, which is crazy. They didn't learn how to do that stuff. Anyway, this thing of like I'm not a techie person, I think it is these days, it's no longer valid. Everybody uses tech every day. And the things that you were doing, the things that you need to do to get
[00:03:48]
loads more out of large language models are a little bit more complicated than learning Instagram. Little bit more complicated. We're not asking people to refract a JavaScript functions or bash scripts. Your interaction with technology is levels above that now. So the good news is that it is pretty easy to learn, pretty easy to progress. And also, as a leader, you don't need
[00:04:23]
to know everything. You don't need to know all of the stuff about how it works. You just need to know enough to lead it and to make it work for you. And those are the foundations that I'll give you today. So who am I to be talking about this? So if you don't know me, hello, I'm Crispin Read. I started working tech in the olden days, like before Google existed, back when people used to say www. as a prefix for a web address. And I had a career in software development and digital
[00:05:01]
strategy and UX. And the code is actually my hobby project that got out of control a while ago, when I became really interested in digital skills training and in particular how we learn complicated digital skills in a work environment that's practical. Last year, I spent a lot of time working with Skills England and the Alenture Institute to lead the research and development for the new level four AI and automation apprenticeship. And during that process, I talked to dozens of
[00:05:39]
high growth SMEs about their projects, about their problems, about their barriers. We also had some research collated by Innovate for hundreds more businesses about their issues and about their wants and needs. And then that led to the not just the creation of the apprenticeship, but also like our understanding of AI and automation as digital adoption, as digital transformation inside teams and small businesses in particular. And also I live this, right? I run
[00:06:17]
a business at Code as Guild. We've had crazy successes with improving efficiency and productivity with AI and automations. And that's what I want to share with everybody. So what's our problem? If you've been to some of our other webinars, you would have seen this slide before. It's a thing anthropic, a couple of months ago, of the stretch of where AI is theoretically implementable across
[00:06:50]
different industries and functions. And then the red, that's the blue bit. And then the red bit is where people are actually using it. So there is an enormous amount of potential here that is not being used. Even though we're doing bits with AI, we're not getting the efficiency gains that we should be. So, you know, we've got, there's research is consistent across everywhere, you know, between 10 and 55% is available for efficiency gains by using these tools. And what that means,
[00:07:26]
you know, what does that mean for a business, even if you look at the floor of like 10%, if you're a business that's turning over 1.2 million, it's 120K a year of available efficiency. So, in the next 20 minutes or so, I'm going to talk a bit about the main concepts that you need to understand to build your own stuff. And the example demo that I'll be giving you is my
[00:08:01]
VA skill that is like my routing for most of my tasks. And in that, I will talk to you about the kind of the words that we hear, like context and tokens and things like that and explain them, because it's important that you understand those in order to understand how the system works. And then once you've got that bit, basically, you're just talking to the thing to create stuff.
[00:08:31]
So, I'm very proud of this slide, because it took ages. And you would have heard things about the context window, context, tokens. So, let me explain that. So, at the start of your chat, which is over on the left side here, you set the context, you've written your main prompt, and then you get a response, and then there's another prompt and response prompt. Now, all the little dots, they represent tokens. So, you can think of those as a word, but really,
[00:09:02]
they're a bit smaller than a word, but you're burning through these tokens to make this happen. Now, your LLM can only hold so much in the context. So, once you start getting, the chats start getting long, your LLM starts forgetting the stuff that was earlier on. So, you have to keep bringing it back in. So, this is when you talk about things like a system prompt. This is a thing that we're loading in at the beginning of the conversation. Prompt engineering is writing
[00:09:41]
a really good prompt at the beginning, or being very specific about what you're asking so that you are giving context in a very efficient way. And context rot is another thing. So, that is when you've burned out of the context window. So, you're starting to get worse results because your chat is forgetting things. It's an easy way to think about it. So, talking about the history
[00:10:13]
of that, 12, 18 months ago, we were writing huge prompts, sides and sides of hundreds and hundreds of words to set the context for a chat so that we could get those results quickly and efficiently. So, to think about it, context is what it already knows or what it's holding in the current context. And then the prompting is what you are typing. So, how do we nowadays, we don't
[00:10:47]
tend to write these huge prompts. We don't tend to go backwards and forwards restating things again and again, or we don't want to because that burns up tokens and it's inefficient and it doesn't give you great results. So, context engineering. This is setting up the context so that your chat is efficient. That's the easiest way to think about it. So, the extra things that have landed inside
[00:11:18]
LLMs. I'm using Claude as an example in this demo, but very similar things exist with other LLMs. They might be called different things and set up might be slightly different. But essentially, they all have memory now. So, memory is the things that are remembering based on the things you keep repeating or the things that you say are important. So, that loads automatically. That's always getting loaded into the chats that you're having. So, it's setting that context with all the stuff in the memory and all the stuff in the custom instructions and the things that you set up as like, this is
[00:11:52]
who I am. This is my job. This is how I want you to talk to me. This is how I want you to respond. I want you to be super formal or I want you to be really informed. Those sorts of things. So, that stuff is going into your chat every time. One of the things I should say is you can control that memory because if you're getting repeated mistakes, you can say, you keep saying this, but actually, the response I want is this or we should think about this in this way or this is how my business operates or this is what I want to happen at this age. Put that in your memory
[00:12:27]
and then the chat will go and update the memory. So, that's a really quick loop where you can improve your experience just by using the memory. Now, I've put a thing here of project. Now, we did a session last month about setting Claude up for your team. Again, Claude, you could substitute chat, whatever. It's more about the organization as people. How do we organize ourselves in a shared digital environment? A lot of that we were talking about projects as in, so projects can group
[00:13:00]
certain things, skills and knowledge basically. In this case, you can think about a project as you are the project. What is the knowledge that's specific to you? What are the skills that are specific to you or your business? So, skills are repeatable tasks. So, you think about what's the examples I've got here? To draft a proposal. What are the steps you take to draft a proposal? What does it look like? What's the headings? How do we follow up after a sales call?
[00:13:32]
How do we turn these meeting notes into the actions that we want? What's our format for this sort of stuff? How do we respond to an inquiry? So, those are repeatable things that happen. And then here on our knowledge side, that's more, maybe it's more like the nouns. Knowledge is the nouns, skills are the verbs. So, this is static documents or bits of knowledge. How do we make money as a business? Who's our ideal customer? What's our customer value proposition?
[00:14:05]
What's our B2B elements of value? What's our SWOT or PESL or all of those things? So, all of these things that you could add as different bits of knowledge, that you could add them at different levels. So, this is the basics of setting up how your environment looks. But now I want to talk about where does all that stuff go? Because you don't want to load everything in at once.
[00:14:39]
This is that we don't want to give every single bit of company information that we've got to every single chat. We don't need to load all of the skills for everything for every single chat. You can see the arrow here of when relevant. If you're having a chat and you start talking about a proposal, your chat is going to think, oh, I'll use the proposal skill. Or if it gives you a response, it's like, oh, you didn't use the proposal skill. You could say, you didn't use the proposal
[00:15:10]
skill. And it'll be like, oh, sorry, let me just adjust that for you. Now I'm reading the skill and get that better result. So, it's setting these things up that give you these better results. So, what does Claude need to know about me to be useful? He needs to know things about your business. He needs to know things maybe about your team or maybe things that are specific to a project. So, these skills and knowledge could be specific to you as a person or in your specific role. They
[00:15:43]
could be for a project, it could be for a team, it could be for a business. And if you organise those things correctly, and you could have things that are a cross project or in a team and a project and the business, it doesn't matter. It's just like the organisation of where that knowledge and those knowledge and skills reside in the large language models, that's how can you get really efficient, prompting efficient responses. So, you can bring those things in at the relevant time.
[00:16:18]
So, now, as you're not supposed to do in webinars, I will switch the presentation and we'll look at a demo skill that I've built. Here we are, Claude. So, this is what a skill looks like if I show you in code form, it's written in markdown, which is, I feel like it's the new QR code. The things that
[00:16:55]
came out 20, 25 years ago, everyone hated at the time, but now everyone loves. So, markdown is just a really quick way of organising text. So, when we look at the visual side of it, we can put things in tables, we can have quotes, we can have headings, but it's very, very, very light on file size. So, not only does that mean the storage is cheaper, it's better, but it also means it's
[00:17:30]
much quicker for a large language model to read. It's not like you give it a PDF, we're burning tokens there because of PDF, it's going to take a picture of it, and then it's going to OCR, extract all of the text data and then read the text. Whereas this, it's asky, it's just reading every single little word here or bit of a word that's your token, so it's much, much more efficient. So, I'm just going to whiz through, talking through what this skill does. I guess the important
[00:18:05]
thing is because I have set up my projects, my skills, my knowledge, that makes this sort of thing possible. But I'll come back to how we get here in a moment. What I've got in this VA, I have, this is a demo version, I have anonymized all the stuff, so I'm free to share it. So, lots of the stuff is all made up. There's no emails or stuff in here or board identity. So, this is the main thing about routing logic. Where, listen for these sorts
[00:18:41]
of keywords. If I say drafts and reply, I'm talking about email, it's task creation. There are bits where I keep notes, and that's generally in Notion. There are things where I want to do with Canada, and that's Google Calendar. There are, where if I'm talking about document or writing a document, then that's going to be something into Google Drive. It's a bit more than notes. So, this is my first pass at what I want my VA to do. I'm telling it all about myself, my regular
[00:19:14]
stuff. Now, this is where I'm at with this project right now. It's an ongoing project. It's going to go on for ages. But past iterations of this have been a lot more complicated. I can remember it started up, but previously this has been like a thing on Telegram, a voice note on Telegram that went to NAN and got routed and used OpenAI Whisper to do transcripts and then sent things back to different places. It's been a rag system of like memories and tasks or routing that was held in
[00:19:48]
Notion. It's had loads of different iterations, but this newest one, which is only like a week or so old, is now the skills are so powerful in Claude. I was able to put all of that tech into just, you can just do that and a skill. So, I'm just telling the language model, when I say this, I mean do this. There's got rules about what I want to do. So, don't ever send an email. You're only doing things in draft. You've got your tone of voice from the voice skill. So, I've got a voice skill that says, am I just replying to someone in the team? You can draft something that's more
[00:20:22]
informal. Am I writing to the DWP about something? And then it's maybe a bit more formal. So, this voice skill makes decisions about how to draft stuff for me. And again, you know, this is human in the lead. I'm not automating away my decisions. I'm getting things drafted for me so that I can go in and edit them and change them to what they are supposed to be. We've got this task creation. So, Monday is our like joint thing, but I've got like a personal task list in there, but I can send things to all different tasks and all different boards. And these little drinks
[00:20:56]
out tasks, stuff is in Notion, not in Monday. So, there's some stuff there, stuff about calendar scheduling, recurring meetings. Again, there's more information in other skills. So, what I'm trying to do is be efficient with what's the top level that my VA needs to know. And if it needs more clarification, it can go off to those other skills. It's got things here about these are fake, of course, but different Notion pages I use regularly, different Monday boards I use regularly,
[00:21:28]
different Google Drive folders, and the glossary of like shorthand and abbreviations for what these things mean. So, I can just like babble away and jargon and have that. There's some stuff here about the people in context as well. And scheduling rules. This is 10 to 12 is good. No back to back meetings. Put a 15-minute buffer when I'm saying, so if I give a task of like, find me a time with Francesca where on Monday, it's not going to back to back or go in between one of her meetings. It's
[00:21:58]
going to look for an appropriate time. So, I'll just come back now to my final bit of slides. Stop that one. Back to my slides. Here we go. And so, one of the things that I wanted to make clear is like, obviously, you know, I've kind of
[00:22:31]
rushed through a lot of things, talking about skills and talking about knowledge files and how they're all great. I've got another, the webinar that we did about setting Clawed Up for Teams is on the website. If you would like a link to that, just let us know in the comments and we will pop your link to go and see that one. That explains how projects, knowledge, skills a bit further. Also, connectors, you know, I talked there about like email and notion and calendar.
[00:23:03]
People are often a bit worried about setting these connectors up and letting them run, but we're doing a session, our next session, it's the beginning of next month, it's about setting those up and like a few sort of bits of best practice around that. Also, remiss of me not to mention that we do, of course, run courses around this stuff. So, if you would like to learn about this stuff with a group of people over a short course or a longer apprenticeship, then let us know in the
[00:23:34]
comments and we'll get in touch. And so, my thing to leave on, I have got a few minutes left for questions, so I haven't gone too far over. But I think my point here is your first step is to start the work. So, you start a conversation with Clawed or ChatGBT or Perplexity or Co-Pilot to help you build this stuff up. I'm looking to set up a VA. I'm looking to give you more
[00:24:04]
information about our business and how that works. Help me. What documents do I need? What documents could you help me create? Starting that conversation will get you to the next steps. Also, in the blog post that we will put out tomorrow, probably Monday, I'll add some starter prompts for you for starting those conversations so you can use them. So, yeah, just please do get
[00:24:34]
stuck in. Block out an afternoon next week. You'll feel the difference immediately and incremental from there. All of the efficiencies compound. Thank you. Thank you very much. That was really good. And before you mentioned that you'd anonymized it, I've started to panic thinking that I've been avoiding some of the weekly meetings. I just felt good. Kind of coffee might be a skill, but anonymize all the stuff, put some random things
[00:25:10]
in. And yeah, I think some people have got some crazy job titles too. Okay. So, we've had a DM message in that Francesca shared. How can you share these skills company-wide? That's a great question. So, the skills can be personal. In the customized section where I was doing the demo, the skills are actually split there between your personal ones and then there's shared ones and then there's organizational ones. So, you can have ones that are shared in a project,
[00:25:46]
you can have ones that are shared between specific people, and you can have ones that are organization-wide. That is such a powerful thing when you've got a shared way of doing things and we use that a lot in where we have skills that we are constantly updating as new things happen. And it's a really such a good way of working that we're constantly in this retrospective mindset of reflecting on how that's working, how that's getting better, and then updating the skill
[00:26:17]
through generally, through having meetings about it and then taking the transcript and getting an update to the skill just from that. I think there's a kind of a relevant kind of follow-up question. It's quite long. Do you share any of the skills with the rest of your team? That might be a slight delay on this. Information about the business seems like something that would be relevant to be shared with your whole team rather than managed individually. Yeah, absolutely. I mean, again, I'm kind of plugging the webinar we did a few weeks back of that team setup. So, when we were
[00:26:51]
going through that, we have projects that give us that hierarchy. So, these skills are only available to these people. This knowledge is only available to these people. So, you can be really granular with all of the things. Like when we talk about the connectors next month, we can show how granularly you can be with those permissions too. Question from Tom. Do you externally host
[00:27:25]
always on agents or do you run it locally on a server? Oh, well, different. It depends. Our general Claude team's stuff is all with Claude. So, all of those things are hosted there. We have, this is something for later on as in maybe with the Claude code when we're going to run later in the summer. But we're in the process of getting these markdown files actually out of
[00:28:01]
Claude and then having them in GitHub. So, we're retaining our own information so all our skills and knowledge would be held externally and called in. And that, I mean, that's like a future thing then getting everyone to a localised thing. But the mechanism is still, it is available, basically. So, we have other, I have other projects where we are sharing skills that are specifically private and they are only, they are on a private repository on GitHub and you have to
[00:28:35]
have access to that to download it or to clone it to your, and then run things locally. It's a use case by use case. Yeah, we have something still on Na10 with agents running on them 24-7. It depends on exactly the use case. So, there's a question on YouTube. I don't know how to put it on the stage. But how long does it take to get a VA set up? Well, I guess this has been like an
[00:29:08]
ongoing project of mine for literally years. But, I mean, it's kind of, it's a marked thing that say that's like a probably a three-year project of things that I've been trying to do to the effect of that. But that version of it took, it probably took me at an hour of backwardsing and forwardsing and getting things from old bits of way things are working. I think if I was starting from scratch
[00:29:42]
to build something like that now, not long at all. And I think like, if I carved out an afternoon, like I was saying just now, if you put out an afternoon and then started the conversation, you'll be collecting documents and putting them in knowledge. You'll start to be, you'll start to create a few skills. In an afternoon, you could definitely do like a small company set up, I think, one or two people. And it's not something that you should ever consider
[00:30:17]
done. It's something that evolves over time. Like you said, you have iterated over and over again, but the power of the current Claude models and set up means that it is just becoming even better each and every time you go through the process of improving it. But it is a discipline to make sure that at the end of an output or at the end of a week that you review where it's gone wrong and
[00:30:47]
what you can do better, which is quite often the phrase that I drop into Claude. And so, yeah, they're blocking out some time to do the majority of the prep work, but then making sure that you're constantly updating it just out of habit. Yeah, I think the thing is, I think like, things like this, go back and think back to the first slide of like, oh, I'm not a techie person. These sorts of things used to be so far out of reach of everyone. And now you can talk it into
[00:31:20]
existence. The first version of that that I built took me ages and was reliant on loads of different systems and didn't work very well. And I didn't have the time to put into it, effectively just trying to build an app by clunking things together. Whereas now the technology exists that we can virtually talk these things into existence really quickly. And they can be really personal and valuable. But you've got to do the work, right? You know, it's the whole thing is iterative.
[00:31:54]
It's going to get better. And the best thing you can do is start. I think we'll just have one last one from Abigail. Are there any good reads videos to explain what's possible with this kind of model for beginners? Yeah, so, well, they'll go back to like, touting our own webinars. I think that's got a kind of a flow to it of getting into it. Anthropic have got a great resource section of things to learn. We'll put that into the in the comments as well. And I'll add it to the blog post. And but like, I mean, I keep going
[00:32:29]
to Claude because it's that like our system, but all of the knowledge language models have have training around how to use their products. And there are some really good creators on on on TikTok and YouTube on Instagram, who are sharing really worthwhile stuff. In the blog post as well, I'll put some I'll put some links up to there for people to for people to follow.
[00:32:59]
So, Brill, right, we've gone a little bit over. So just enough time to post the the upcoming sessions that Crispin Read has already mentioned. But thank you so much for your time today. Please keep the comments going in LinkedIn and YouTube. And we will be following up with all of you. Please keep your eyes on the website for the blog post update, which will, which we'll be sharing on LinkedIn as well. And it was great to have your time today, Crispin Read and see you all again soon. Thanks very much. Bye.
.
