Scoping Your First AI Implementation: From Brief to Production

Charlie Cowan: Okay... Alrighty... That's... Put that on. Do not disturb. And move a few things up here. Okay, I'm sharing my screen. Good to go. Well, good afternoon, good morning, depending on where you're dialing in from, for this week's Kowalah Wednesday webinar. We are in the middle of our, I think, our fifth heat wave here in the UK. A lot of people are away on holiday, so thank you to those of you that are tuning in, either live or on YouTube. We get together every single Wednesday at 3 o'clock UK time, and it gives us an hour to just talk about things that are top of mind in the world of AI, and especially in the lens of business use of AI. And today we're going to be talking about a topic that's coming up so much in, certainly, our new clients. Which is, you know, how do we plan the first project, the first deployment?

Charlie Cowan: Where do we apply our efforts to get the most impact? So, we'll spend a little bit of time on that. If this is your first Wednesday webinar, then we'll follow a fairly similar process to how we do normally. We'll spend probably 35 minutes or so on the main topic of the day, which is, as I talked about, planning your first project, your first deployment. We'll then have about 15 or 20 minutes to look at AI in the news. This is our recurring segment when we look at what's happened in the last 7 days and help bring you up to speed. As I say, there is so much going on that it can be hard, even for us, who are living this day in, day out, to keep on top of everything. And this is your opportunity to get that fast track to know what's going on.

Charlie Cowan: And a little insight, it's been quite a crazy week. There is a lot. I think I've got 5 or 6 different things I'm going to talk about in AI in the news. And then, we've got some time for any questions, but if you've got questions as we go through today's session, and you're watching live, then feel free to ask as we go. So, we're in a Zoom event. You can either hit the Q&A button, and then you can ask a question there. There's also a chat button there that you can ask in. I'm obviously presenting, I've got both the Q&A and the chat window open. Feel free to put your question in there. And then Caitlin, my colleague, is also on the line and is able to point me in the right direction if I miss anything.

Charlie Cowan: With that, let's get into the core of today's topic, which is scoping out and getting started on your first AI program. Now, we're working with a lot of organizations who, we hear a lot of the same thing. You know, I'm sure we're behind. I'm fearful that we're missing out, that everyone else is further ahead. And having made that commitment and that decision to invest in an AI program, it's what's the first thing that we should be doing? Often, that is, we need to do some kind of training or change enablement, but increasingly, you know, we need to build something that people can use. And what I'm going to walk you through today is some of the lessons learned and best practices that we're seeing across our organizations, across our clients. So, firstly, what are the missteps that you can make when it comes to building something?

Charlie Cowan: And these are things that we see quite a lot, either before we've got to a client, or often when clients are asking us what they want to work on. The first is the pet project. You might have heard of this in all kind of decision-making in a company, which is HiPPO. The HiPPO decision, which is the highest paid person's opinion. And this is not a great way to make decisions. Well, the person that is top of the org chart, or highest up in the individual meeting room that you're in, says, well, this is what I want, and I'm in charge, and therefore, that's what we're gonna do. What you often see here is that you're building something that solves one person's problem, and not the team's problem, not the process's problem, not the company's problem, and so this is definitely a gotcha to look out for. The second is the thing that demos really well.

Charlie Cowan: Now, a lot of people are building some crazy things in Lovable, in Replit, in even using Claude Code, and being able to bring something together that looks really, really impressive, especially if it's, you know, combining, I don't know, maps or nice sort of interfaces. This thing looks really good. But once again, does this actually impact the team? Does it impact the process? Does it impact the output of the company? Or is it just a snazzy dashboard that one or two people are going to have a look at now and again? The third gotcha is tackling the most difficult process, the most challenging thing. I'm gonna talk a bit about month-end close processes later, but you can think about that as something that's quite challenging. It involves a lot of people, it involves a lot of systems, it involves a lot of data.

Charlie Cowan: And if you gravitate towards the most difficult thing, then this can often be a challenge, because a few months in, you've still got nothing to show for it, because you're still building it, or it doesn't quite work. And that leads into the fourth one, which is certainly what we see in large enterprises, the endless pilot. We're not quite sure what we're aiming for. We've got a vague idea. We're not sure the data is correct, we're not sure we've got access to the right integrations, and so this thing just keeps going, keeps nudging forward, but never makes it to production. And these are all alerts that you want to miss when you're building out your first use cases. So if that's not what you shouldn't do, then what should you do?

Charlie Cowan: Well, if I was going to phrase it in one sentence, I'd be saying, you need to be looking for a use case that's going to give the biggest impact, that you're confident about delivering that impact, and that it's easy enough to ship. And so there's 3 words in here that are really important. Impact, confidence, and ease. Now, you might have come across this before. If you're in a product team, you might have heard of the ICE score, and scoring things against impact, confidence, and ease. We're going to look at this a little bit further as we go through, but you need to be driving business impact from your use case. If it doesn't drive business impact for the team, for the company, then what's the point? You need to be confident that you're going to achieve that impact, because if you're not, then you're just, you know, hallucinating.

Charlie Cowan: And it needs to be relatively easy to deploy, because if you can't do it, because you haven't got access to the data, you haven't got the integrations, you haven't got the systems to be able to do it, then again, you're going to be stuck in that never-ending pilot, the thing that never gets shipped. So the biggest impact that you're confident about, and that's easy enough to ship. So where do we start looking for these types of use cases? Well, I'm going to go through 3 graphics now, 3 or 4 graphics, that I'm walking every leadership team through, when we start one of our engagements. So the first, I talk about this a lot, so if you're on last week's webinar, you'll have seen this slide as well. There is a fantastic book, I'm going to show it on my screen here, you might be able to see, called High Output Management.

Charlie Cowan: It was written by a guy called Andy Grove, and for many years, he was the CEO and then chairman of Intel, so the memory, and then the chip manufacturer. And this book was written not last year, not two years ago, but in 1983. So, before the dawn of email. In fact, in the book, he talks about the incoming electronic mail as a new way of communicating. So it's an old book, so why am I mentioning it here? Well, he looks at driving a company, at leading a company, as a series of production lines, a series of processes. And, whether that's finance, whether that's legal, whether it's people, sales, every team is a production line. You have an input on one side, so if you're a sales team, your input might be leads that are coming from marketing.

Charlie Cowan: You then apply labor to it, people, and then you have an output, which in the sales team example would be signed contracts, customers that are ready to be invoiced. And this production line is what you, if you're the VP of Sales or the Chief Revenue Officer, you own that production line, and your job as a leader, this box on the right, your job as the Chief Revenue Officer is to drive the highest output, so the most amount of sales, at the required quality, these sales don't churn and leave after one year, for the lowest effective cost. So, we need to hire people, but we want to hire the, you know, the fewest and at the most effective cost to drive that required quality and the highest output. The same is for legal. The same is for marketing. You have an input, you apply labor, your people, and output.

Charlie Cowan: And this is the book that Andy Grove is writing, 40 or so years ago. Now, re-reading this in the world of AI is really, really insightful, because you start thinking about that job of the leader is exactly the same as it was 40 years ago, driving the highest output at the required quality for the lowest effective cost. And when you think about AI, one of the first things that you might fall into is thinking, right, how do I give AI to my people, and they will start optimizing their bits of the process. So here we think about giving people ChatGPT, giving them Copilot, giving them Claude, maybe giving them some training, and saying, look, here's a tool, can you use it to do your bit of the process a little bit faster? What you find here is that, A, you have a bell curve of people.

Charlie Cowan: Some people will lean into AI like nothing else, this is right up my street, you know, I'm self-learning at the weekends, I'm building skills, I'm building little plug-ins, and say, absolutely, this is helping me. But you'll go through to the middle of the bell curve, where people are like, oh, I'm not too sure, I haven't got time, I'm too busy for this. And then you'll go to the people that are actually sort of against AI. You know, I'm not going to use this. I like typing into my laptop. I like working in my Excel spreadsheet. I've only got 5 years to go till I finish work. I don't want to change the way that I work. And so even though you do apply this training and change enablement to the team, you're going to find this uneven distribution of whether that changes.

Charlie Cowan: And even if everyone took on what you trained them on, what you're going to find is that that doesn't really change the output. It makes each individual in the process more efficient, but it does not necessarily drive the output. So you'll often hear that companies will say individuals are experiencing 20-30% productivity gains. But are we then selling 20, 30, 40% more as a company? No, we're not. Well, where's the difference? Well, that efficiency in the box is not turning into the output of the box itself. So instead, what I'd encourage you as a leader of an organization to think about is the right-hand side of this image here, which is, instead of just applying AI to the existing cogs that are in the existing box, is to go, right, how would I rebuild this process for an AI world? And that might be redeploying the people on different parts of the process.

Charlie Cowan: And it might be having AI or agents within that AI sort of category picking up entire parts of that process. That's going to allow you to drive a much higher quality of output and higher volume of output at a lower effective cost. So that's Andy Grove, and Andy Grove's black box. We might come back to that a little bit later. Next up is this image that I also come back to a lot. We draw it a lot on whiteboards in our workshops, and this is the Gartner AI Opportunity Radar that I first came across a couple of years ago in, yeah, 2023. And it's just a great way of thinking about different use cases across your organization. So you've got two axes here. You've got the horizontal axis that goes from everyday AI, you know, we're summarizing emails, we're sending meeting updates, we're planning for a one-to-one with our manager.

Charlie Cowan: Over to game-changing AI, which is the AI that's gonna change the direction of your company, maybe even change the direction of your entire industry. And then you've got from internal operations, down at the bottom, so the things that we do in our organization with our own people and processes, up to external facing. So, yes, this could be customers, as I've got on the slide here, but it could be suppliers, it could be candidates that are thinking of joining your company, it could be investors. And as you go through this, the four quadrants, it's really easy to start thinking about the different types of use cases there. Now, I won't go through all of them individually, but what I will say is that down in the bottom left, everyday AI and internal operations is where companies gravitate towards at the start. It's what they think of as the use cases.

Charlie Cowan: We've got a finance process, we've got an HR process, we've got a legal process. I know what this costs us today, because I can look at payroll, I can see we've got 10 lawyers, and I think if we could do it with 5 lawyers, then I know exactly what number would drop to the bottom line. If I think about the right-hand side, this can be a bit more difficult to think about and to estimate. So the bottom right is core capabilities. You can think of this as the guts of how your company runs. Our supply chain, our R&D, our internal operations.

Charlie Cowan: I always think about Amazon here, the logistics of getting an order shipped out, arrives the next day, it's not the right thing, you go online, you say, I want to do a refund, you drive down to the garage, you've put your thing in the cabinet, and you get your money back in your account a few hours later. You know, that is core capabilities of the guts of the business. And then top right is new products and services, new value propositions, new things that you can take to market that you couldn't do without AI. So you might think of markets that you can sell in now, different languages, maybe. Maybe you can sell something for more, because it's got an AI-enabled component in it. Now, if you were to draw a diagonal line from the top left to the bottom right, generally, everything to the bottom left is quite defensive.

Charlie Cowan: So it's about saving money, it's about stripping cost out of the business. And there will be ROI in that, but there's a natural flaw to that. You know, if you've got 10 people in legal, you know, the minimum you could have is zero. And so there's a natural point that you can get to, and you don't want to have zero, because you need people to negotiate contracts. So, you know, there's something that you can do there, but also everyone else is doing that, so it's not a differentiator. If you take that diagonal line and go to the top right, and think a bit more about core capabilities, a bit more about products and services, a bit more about anything front office that is facing your prospects, this is where you're coming up with new stuff.

Charlie Cowan: And this is more challenging to come up with, because it's an unknown unknown, and so often this requires, you know, a few workshops and discussions to think about the potential there. But definitely keep this in your mind as you're thinking about different use cases. And then the third dimension I want to talk to you about, we talked about Andy Grove's black box and processes, we've talked about the AI Opportunity Radar from Gartner, and then the third, this comes from us, from Kowalah thinking, is splitting out AI for the people and AI for the company. So I talked about this at the start. You know, a lot of organizations have started by thinking about, I'm gonna go and train our people on how to use AI, and if they are trained, then they will create magic for us, and something will happen. But this is very much inside Andy Grove's black box.

Charlie Cowan: Whereas, if you start thinking about AI for the company, and actually redesigning the box, you've got something that persists as individuals move on from the business, as people change roles, they change teams. You, the company, own the way of working, and you've injected into that. And so you'll hear a lot in our conversations about AI for the company and AI for the people. And even where the people are coming up with ideas, how do we promote that and take ownership of it at the company level? So how do we go and find these use cases, now we've got these ideas? Well, I'm going to introduce you to a concept that we call, talk about, called a discovery. We launch a discovery out to your people. That can be either to the whole company, or it can be to individual teams, or groups, or regions, or business units.

Charlie Cowan: And we're really going out there to go and uncover their ideas of where AI could impact, both themselves and also their team and the company. So there's two ways of running this. So one is, you run this as a workshop. So, let's take the example of your finance team, because we're gonna follow this thread as we go through. So let's say we've got, you know, 15 people in our finance team, and, or finance leadership team, maybe, and we're gonna get them into a workshop in a room, and we're gonna run them through this exercise, where we're gonna gather use cases from them in person. The second approach on the right is, well, we've not got everyone in a workshop. Actually, there's too many of them, and they're distributed across different teams, different units, different offices.

Charlie Cowan: And this is where we send them the question, and we get them to provide their responses, and we consolidate them. And sometimes, you'll find people are doing both of these, so we run the workshop, we gather a core set of use cases, and then we go out to the wider team. When we're going out to the wider team, there's a range of different channels that people can come back in, so it's not just your traditional fill in a form. That's very 2000s. No, this is more interactive, chatting in Slack, chatting in Claude. Wherever it is that the work happens today is where we want people to have this conversation and think about the work that they're doing, and where these use cases could be most impactful. What is a discovery, in Kowalah terminology?

Charlie Cowan: Well, a discovery is this process of going out to your people and saying, look, I want to gather use cases around this specific topic. So, it could be specifically to do with finance, which is the example I'm going to give you here, but it could be anything across an individual team, business unit, or the overall company. When you're setting up a discovery, the first thing, oh, we'll go back one, is we're going to provide them with a brief of what is it that we're looking for here. So you can really guide your people when they're getting involved in this discovery. So you may have some strategic objectives. You may call them OKRs, Objectives and Key Results. It may just be your strategic goals.

Charlie Cowan: So, we're going to plug that into the discovery, so that when your people are responding to this, we're really guiding them to come up with ideas, with opportunities, with use cases that support the company's objectives. AI for the company, not just AI for the individuals. We're then going to decide who this is going to go to. And so, like I said, this could be the whole company, but it could be a specific group of people. Who are the people that are closest to this process, closest to what is happening, are going to have the best, you know, ideas of what's broken, or where the opportunities are? And then, as an admin, as the discovery continues, you're going to see these come in, and you're gonna be able to track them in the Kowalah platform, or via the Claude connector that we'll give you. So that's the wrapper, the discovery.

Charlie Cowan: What does that look like when you're an individual? Well, if you're just filling this in in the Kowalah platform, then it's going to look something like this, where maybe I'm in finance, and we're doing a discovery around finance use cases, so I might say that this is going to help me with the month-end close process. And, here I'm going to type in a little bit about what that might help with. I'm just going to talk, because it's a lot easier. We have our month-end close process, we use NetSuite as our finance system, but the data needs to come from a range of different systems and accounts, and there's a lot of different people that are involved in this, because we run multi-region and multi-entity setup. And therefore, our typical month-end close process is taking sometimes up to 20 days to get completely closed.

Charlie Cowan: We think that there are a lot of repeatable exercises in here where AI could be helpful. So, we're gathering that information there, happy days, the team, I'm gonna say finance, ops, and then they're gonna log that, and then that goes into your pipeline. Now, I was just putting this into a slide here as an example, but imagine having that chat in Slack, in Teams. It's very just relaxed about talking about where you think the problem is, where you think the challenge is. If you were doing that in Slack or Teams, you might add in a couple of attachments. It's a lot easier than just filling in a very long form that is logged somewhere. So that's capturing all of the use cases.

Charlie Cowan: What you're then going to do, once you've got these 30 or 40 use cases that have come in from your team, is that as a core AI team, as a leadership team, you're then going to score each of these use cases using the ICE framework that I talked about. So, impact, confidence, and ease of implementation. So, our opportunity here is close the books in 5 days, not 20 days. This is the high-level idea, you know, now we're gonna score that. This is where you want to do this in a workshop environment, and you want the right people that are involved. Not just the person that does this today, but maybe some people that are around at that process and that team. So, question number one is, what would be the impact of this?

Charlie Cowan: So, if we could close the books in 5 days, not 20 days, what would the impact be to the business? You know, are we going to sell more? Are we going to be able to make better decisions? Is it going to save us money? And here's where we want to think, not just about the individuals that are doing this, I'm sure this would be impactful to them, but is it impactful to the finance team? Is it impactful to the company? And so you might say, right, well, this is 8. At the moment, it takes 3 people 20 days to do this. I know what that costs. If it saved us 15 days on that, well, yes, maybe we don't need that many people to do it. Maybe they could be focused on something more strategic.

Charlie Cowan: I think more of the value would be that if we could close our books in 5 days, not 20, we can make better decisions that are going to give us 25 days before the start of the next month, rather than just 10 days. So, I think that's the strategic impact. The next question is about confidence. How confident are we that if this thing existed, it would have that impact. So the question here is not, how confident are we that we can close the books in 5 days, not 20? That's going to come later in ease. This is more, how confident are we that if we did close the books in 5 days, it would have an impact of 8? And so here, we might be saying, well, you know, I think it's quite strategic, but I don't know that it would have that impact, that if we could close the books.

Charlie Cowan: So maybe I'm feeling this is a 5, rather than a 10 or a 1. And then the final one is the ease of implementation. So this is more about, how do we feel about whether this is possible or not? Well, it requires a lot of data, there's a lot of manual work that happens at the moment. It requires humans to be speaking to each other on a phone, so we might be saying, okay, well, actually, ease of implementation, ease to get this, this might be a 3 or a 4, potentially. What this does is then gives you your score, 160 out of 1,000. Because you times each one by each. So 10 times 10 times 10 gets you 1,000, 1 times 1 times 1 gets 1. And so this gives you this scale of being able to score these things.

Charlie Cowan: If I just go back here to this example, I didn't, this one isn't the exact score, but this gives you this example here of where we've got 5 opportunities, they've all been scored, and now we can really see, right, well, this one's at the top. It might not be the most impactful, but because we're confident in it, and it's got a high ease, actually, that's why we're going to do it. Whereas there might be something that is, you know, very impactful, but it's very complicated, and so that might be, you know, down the bottom here. So this gives you your ability to fan out, go and get lots and lots of use cases, and then come back and then work out what are the one or two, three things that you should be working on. And our guidance to clients is, you're not running one discovery.

Charlie Cowan: Each leader of each team should be running a discovery into their team. Head of legal, head of finance, head of marketing, head of Germany, head of this business unit, depending on how it is. Get to the people that are closest to the work, and closest to the process, and start uncovering these high-impact use cases. So now you've scored it, and you've picked, you've scored all of your different use cases. What you're gonna do then, and we do this in the Kowalah platform, so you don't need to worry about it, is score it against two different, these dimensions. Or, let's say score it, it's gonna visualize it for you in two different dimensions. So this is your ICE scoring. So, you've got impact across the bottom here, you've got ease on the left here, and then the size of the dot is confidence.

Charlie Cowan: And so here, we've got this close the books in 5 days, not 20, and we can see this is, you know, this would be 1, this is the top-scoring one compared to all of the others that I'm clicking over. And then the other lens is the AI Opportunity Radar, from Gartner. Same thing, we've got the everyday AI over to game-changing AI, and the internal to customer-facing. And up here, we've got some nice use cases, but down here is our, where are we? Close the books in 5 days, not 20. Now, this is quite a useful lens for you to have a look at, because as you capture lots of use cases, just remember what I said. Everyone is focusing on the bottom left-hand quadrant. Existing processes, existing ways of working, making them more efficient. And this may well be the ROI that funds your AI program.

Charlie Cowan: But keep a close eye on some of these things that may not score highly on ICE, but they may be very impactful long-term. And these may be a few that you want to keep in your back pocket as you're building these out, because they're going to have the better impact on your organization long-term. I just want to pause just for one second before we go on. As you'll notice, I'm presenting in Claude Design, and we've created these slides. The way that the slides have been created is it's plugged directly into the codebase for Kowalah. And it allows me to have very interactive experiences as I'm presenting. And it is really pushing forward my, I guess, understanding and belief in what's possible with how you run a slide presentation. Before, what would this have been?

Charlie Cowan: Screenshots, not very dynamic at all, but I definitely encourage you to have a play around with how you can bring a story and a talk track to life, in Claude Design over PowerPoint or Google Slides. So we've now picked our idea, we've picked our idea, and we're gonna do the close the books in 5 days, not 20, because it's gonna be, it's top scoring, and we think there's a major ROI on it, so that, that's fantastic. So then it's about, like, how do we get that thing built? And there's a couple of lenses that I want to share with you here to think about. So one is the layers that you're going to go through, and we're going to start from the bottom on this one. So, keeping in mind, closing the books, as a finance process. So, number one is, what data does that process need to access?

Charlie Cowan: So I talked about NetSuite as an example for an ERP. You might use Xero, you might be using Workday Financials. You may need to pull in information from your CRM about deals that have closed. You may need information from your supply chain, there may be other documents, spreadsheets, and things like that. So really thinking about where do your humans go to get this data today. And can we give eyes, ears, and maybe hands, to your AI to be able to pull in and retrieve that information, make the required updates, and then push stuff back in to those systems. Typically, what this means is that you want to create an internal MCP server. It stands for Model Context Protocol. It's how AI models speak to other systems.

Charlie Cowan: And without even knowing what use cases you're going to come up with through your discovery, I'm pretty confident that those use cases will require AI to have some safe and governed access to your existing systems. And what you're going to want to do as an AI team is to have control over that. This isn't just giving everyone access to a connector in their own client. You want to provide your own connector that you give to your people or to your agents. So, knowing that you're probably going to have, after this first use case, you're gonna come up with 5, 6, or 7 different use cases and agents, I'd be making sure there is a group of people in your AI ops that know what an MCP server is. They have a plan to build one, and they're either building it internally or working with a team like Kowalah to build and maintain it for you.

Charlie Cowan: That's going to give whatever you come up with next the safe, governed access to read and to write into your other systems. Layer number 2 is the window in to this data. Now, when you're, we talked a little bit earlier about AI for the people and AI for the company, and so the natural first thinking is, right, we're going to give everyone access to Claude, we're gonna give them access to ChatGPT, we're going to give them access to Copilot. And the window that the human is gonna have is by sitting in front of their keyboard, typing, help me close the books, help me, you know, write a proposal, help me prepare for my one-to-one. And so it's the human doing the work, they've just been given some AI to assist them.

Charlie Cowan: What I want to challenge you with is thinking about, actually instead of giving them a window through Claude, or Cowork, or Codex, or Copilot, is where's the work getting done today? Well, you've probably got this team working in Teams, you've probably got them in a channel in Slack, and actually, that's where the work is happening today. And actually, to give them access to help them with this month-end process in Teams or in Slack is the right place. Not asking them to move out of that, to go as an individual, to chat to their Claude, or to their ChatGPT, and then move it back in. And in some cases, there isn't an interface at all, because the agent is just doing the work in the background, and it is just, you know, posting to Slack or to Teams. So just have a think about the window in.

Charlie Cowan: Not everything is about giving your people access to another UI to start doing their work, but doing it in a more efficient way. And then the third layer is, how's this thing going to get done? You'll often hear people talking about skills. This is how we can give the AI the ways of working, this is how we do this. This is the house style, this is our playbook. You can think about an SOP, a standard operating procedure. These are the things that we need to document how this new process is going to work, and we're giving that to the AI. So these are three layers. How do we get to the data? What's the window into that data? Is it a human accessing it, or maybe it's an agent?

Charlie Cowan: And then what are the best practices and the ways of working that are going to allow the AI to do this job, closing the books in 5 days? So I touched on there about, is it the human doing the work with AI, or are we going to give an agent this new task instead of a human doing it? And this is going to be a big discussion point for you as you think about your first deployments. On the left, I'm going to say Harry from Finance. Well, let's use Mark, because I'm going to talk about Mark the agent in a minute. So, Mark was a CFO at a company that I worked at previously, and so I know exactly, you know, who Mark was and what Mark did, and so it's easy for me to think about that. So, on the left, you said, right, well, we're gonna give Mark some skills.

Charlie Cowan: We're gonna give Mark a closing the books skill. We're gonna give Mark a tax skill. We're gonna give Mark an M&A skill. All of the things that Mark does today, we're gonna give him some skills, and we're gonna tell him, hey, Mark, you know, next time you're closing the books, remember to use the closing the books skill. This is in the box. We're helping Mark to do the job that he's already doing a little bit more efficiently. But it relies on Mark changing the way that he works. Mark needs to be AI positive. Mark needs to know that the skill is there. Mark needs to know that he's gotta go to Claude, and he's got to say, I need to close the books. If Mark chooses to carry on working the way that he has always worked, then it doesn't matter how good the skill is, no impact.

Charlie Cowan: Over on the right, though, we got, right, instead of Mark doing the work, actually, we've built a digital agent called Mark, and we delegate the work to him. So, when we get to the first of the month, Mark triggers, oh, I need to run my closing the books skill, and off he goes and does that. I'll just show you a little bit of Mark. I'm not going to give you a demo of Mark, but just so that you can understand a bit about what I'm talking about. Here at Kowalah, we have an agent hub. These are all of our in-production agents. We've got them split down through go-to-market, we've got our delivery team, our operations team, these are all agents, and then we've got finance here, Mark, our digital CFO. Mark is live on Google Workspace, any of our team can Gchat him, and Mark has got access to a number of skills.

Charlie Cowan: And so here's how he manages cash flow. He provides us with financial snapshots, he assesses the profitability of deals that we're doing, he provides financial coaching to any of our leaders that want to understand how to manage a P&L, a balance sheet, and cash flow. And Mark's also able to interact with our accounting system and track all of our payments. So instead of us giving a human CFO skills that they use, we built a digital CFO instead, and that delays us needing to hire a human one. So just think of that lens a little bit. AI for the people, AI for the company. AI for the company is you just build agents that do these things, rather than building skills to give to the people that are doing it today.

Charlie Cowan: And then final bit on this whole process of how do you bring your first AI use case to life, is a little bit about the process. So, first off is defining what you want this thing to do. What does it mean to close the books in 5 days instead of 20 days? What does good look like? Spending time with your existing team, mapping that out. Step two is designing it. How's it going to work? Is it a skill for the humans to use, or is it an agent that the humans are gonna delegate to? And then launching. And my key point here is that this is not like a typical software project, where you get to, you know, final sign-off, testing, go live, and hypercare, and then, you know, you're done. After 3 weeks, you leave it.

Charlie Cowan: Actually, the real work starts once you've launched your new agent or skill, because now, just like a new starter joining the company is when you want to train it, coach it, ask it why it's doing certain things. And I really think that if you think about how you'd onboard a new employee, this is the right way of thinking about bringing on an agent. You're not expecting much on day one. You're in fact expecting near zero. Maybe there's some domain knowledge because of what you've trained in the design, but really, this is the starting point. So, what can you do between now and, you know, next week to get started on this? Firstly, I would think about launching a discovery. We can obviously do that through the Kowalah platform, or you can run that just yourself.

Charlie Cowan: Go and find a business unit, go and ask them, where could AI help with your work, where could it help the team, where could it help the process, where could it help the company? Then go through the scoring process of impact, confidence, and ease to pull out the most priority ones. And then, pick one to get started on. Not necessarily the one with the single highest score, it could be the one that's going to be, you know, strategic to where you're heading in the organization. But, yeah, that's the process. Good luck with it, and of course, we're here to help you. Now, we've got 15 minutes for AI in the news, so I will rattle through the 6 stories. We'll share the deck with you and the recording on YouTube as well. But there has been a lot that has happened in the last seven days.

Charlie Cowan: So, firstly, and I think this is one of the biggest shakeups in the world of AI so far, well, maybe since Sam Altman was fired and then returned to OpenAI about 6 days later. Big shake-up at Google, with Demis Hassabis, who is the founder of DeepMind, which is the main, you may say, the brains of AI at Google, has stepped out of the CEO role to become chairman and, I think, chief scientist. And then, Jeff Dean, who is one of the original brains behind, I think, the TPUs and certainly the early AI models at Google, has left to go and set up a new AI company with a number of his ex-colleagues. And so, you know, I'm not close enough to what's going on within the Gemini team there. But I would say a lot of the view from outside is that this is a bit of a self-implosion going on.

Charlie Cowan: The Gemini models, whilst very competent, have not kept pace with what's going on at Anthropic and OpenAI. And I can imagine Sundar is rattling the cage and saying, look, we need to be more commercial, we need to be up at the front end here. And this is resulting in people changing, you know, their roles and their direction. You know, my perspective is, you know, I'm not seeing customers that are even on Google Workspace. I'm not seeing customers that are coming to the table saying, here's something that we built on the Gemini platform, or that we've trained our people with Gems, and it is changing the way, changing the direction of our business. It really is Anthropic and OpenAI that are leading that. So, we'll have to see what this means to the direction of Gemini and the Google AI program, but definitely one to watch. Next up is around cyber.

Charlie Cowan: So there's been a lot of discussion around cyber in the last few months. If you're on the Anthropic platform, you'll have heard of Mythos, which is their kind of frontier cyber-grade platform. And, on the OpenAI side, they just launched GPT-5.6 Cyber, and they've got some new models coming out, codenamed Astra, I think it is, which are really at this frontier for cyber defense. Now, if you have a spare 35 minutes, I would absolutely encourage watching the YouTube video that I have put a copy of on the right here. YouTube this title, the OpenAI Hugging Face Incident. It is two researchers from OpenAI, and they talk about an incident that happened about 3-4 weeks ago, where a frontier model, which I think is codenamed Astra, was given a cyber task.

Charlie Cowan: Just to summarize it, they give these new models a task to try and find a vulnerability, and they do this in a sandboxed environment with no access to the internet. And this is obviously for safekeeping. We want to test the model, but we don't want it actually going out across the internet. In this video, the researchers describe how there was a package that was linked to this sandbox, and what the models were able to do was to figure out vulnerabilities in this package, and reprogram it, reconfigure it, basically to get access to the admin keys, and then give the models permission to go outside of the sandbox and get internet access. And once they had that internet access, they were then able to go to Hugging Face and try to solve this theoretical problem that they were trying to solve. And, anyway, it was uncovered by Hugging Face.

Charlie Cowan: OpenAI then realized the situation. For a very technical topic, these two researchers do a very good job of explaining what happened and how, and it will give you some real insight into how powerful these models are. What's the lesson here for us as, you know, business users? The models are very, very effective. Where they wrap up the video, they talk about how there are two sides of cyber, offense and defense. And at the moment, the models are in the favor of offense. If you're a bad actor, and you've got access to some of these frontier models, then you can use them in, well, I want to say you can use them for nefarious purposes, you can, and the model providers are trying to limit that. The point of the recorded video is they need, they're encouraging people to think about the defensive side of it.

Charlie Cowan: How do we make sure that our organizations are ready for these models when they get into the hands of bad actors, and how can we prepare ourselves to, you know, push back on these potential risks? So, cyber, definitely a good topic, definitely go and watch that video, even if you're not technical and not in cyber. Third update, this was, I want to say yesterday this was, let me look at the date, yeah, 11th of August, this came out. As you may or may not know, Grok is the AI model that comes from what's now SpaceX, Elon Musk's AI outfit. It was called xAI, and then it was merged into SpaceX. In addition to that, SpaceX have acquired the platform Cursor, which was a coding platform, I think that was for $60 billion, but one of the most widely adopted AI development platforms by coders, and is now owned by SpaceX.

Charlie Cowan: What was announced yesterday is Grokbot, and you can think of this like an agent that runs up in the cloud, and you give it access to your systems, like Salesforce, like your HR platform, like your HubSpot, whatever that might be, and then you delegate tasks to it. As we were talking about agents earlier, this would be something that you delegate to. If you're in the Anthropic space, you might think about Cowork, where you've been delegating things. If you've turned it on, you might be thinking about Claude Tag, where you delegate things. That would be an agent out there. In the OpenAI space, you may be using ChatGPT Work or ChatGPT Codex, depending on the name of the day, but this is all about delegating work, and so Grok is sitting in that space.

Charlie Cowan: I've not used it, but I see that reading the tea leaves on X, the general feedback seems to be, you know, quite positive, that this is an interesting new dimension. So, one to take a look at. Am I confident that enterprises are going to hand over the keys to their sales force or their ERP system, to SpaceX? I'm not sure. We will have to see. But definitely one to keep on your radar. Next up is Anthropic confirming that they are building, in-house, chips. And this is linked also to the next story that came out, I think last night I got the email, that the preview pricing of Sonnet 5 has been locked in at a lower cost. So typically Sonnet was at $3 input, $15 output, and they've locked in $2 input, $10 output. Why is this linked to the in-house chip team? Well, it's all about reducing the costs of running AI.

Charlie Cowan: This is, you know, the frontier at the moment, is how do we give people this top-level intelligence, but at the lowest cost possible? And Anthropic is definitely not in the lowest cost area right now. So, they'll be thinking about ways that they can cut this cost per query so that they can drive down and continue to be cost-competitive with OpenAI. And of course, you know, Copilot or Gemini, if you're using those models as well. Linked to that, and I think this is my final one for AI in the news today, a couple of days ago, Meta, formerly known as Facebook, has open-sourced its most powerful model. Now, this is called Muse Spark 1.2, and introduced Muse Glimmer, which is another model that you're gonna see a little bit about.

Charlie Cowan: And now at the frontier of open source, you have got Meta, you've got Kimi, we talked about a couple of weeks ago, you've got DeepSeek. You've got a few of these providers that are providing these open source, or open weights models, I should say, that you can start to run on your own devices. Now, who's to know whether this is going to become competitive with Anthropic and ChatGPT from OpenAI. But, as companies start to think about their token budgets and how they're managing those costs, increasingly, I think you're going to see some of these open source, open weight models, get fed into the mix that companies are deploying to their people. One more, and this was just today. I saw this about half an hour before we started the webinar. And that is that Lovable has just raised $400 million, at, as you can see there, a $13.3 billion valuation.

Charlie Cowan: Now, if you've not used Lovable, it is in the vibe coding category, I would say, and I don't mean that in a negative way, but it allows non-developers to bring an idea to life. A bit of history, I would say, Lovable has maybe 18 months of product-market fit. They were around beforehand, but they were, you know, striving hard, they were iterating on the product, and then they renamed the product to Lovable, I want to say in, like, November time of 2024. So, yeah, a year and eight months ago, and suddenly it went crazy through last year. So, going from really a standing start to a $13.5 billion valuation in just 18 months or so, I think is absolutely crazy. So definitely one to take a look at.

Charlie Cowan: If your teams are using Lovable or creating things yourself, just creating things themselves, just be aware of that, and, you know, they're really becoming an enterprise-grade platform. With that, I haven't seen any questions come through, so I'm just going to wrap up by introducing what we're going to be talking about next week. If this week was all about planning your first AI project, then next week is all about 90 days in. How do we measure what people are doing with that thing? How do we track? How do we report? How do we come up with an ROI so that we can evidence this back to the organization's leadership? So, same time next week, you can sign up for this just at kowalah.com slash resources slash webinars, and we will see you this time next week. Thanks very much.