90 Days In: What AI Adoption Looks Like

Charlie Cowan: Okie dokie. Share my screen. Desktop 1... And... Let's move this up here... where have we gone? And... present full screen. Okay! Well, good afternoon, good morning, depending on where you're dialling in from, and welcome to this week's edition of the Kowalah Wednesday webinar. If this is your first Kowalah Wednesday webinar, we get together at the same time every week and get to spend an hour to talk through something that is top of mind in the world of AI, and specifically the world of business and AI. That might be something that we are seeing in our customers, it might be some new features or product releases. It might be something that we're using internally. And it's such a fast-moving space that it gives us some time to just pause and sort of catch up and share some of what we're learning with you.

Charlie Cowan: Now, if this is your first time in this Kowalah Wednesday webinar, then we're in a Zoom event. You are going to have the opportunity to ask questions. So, depending on what device you're on, maybe at the top of your screen, the bottom of your screen, you're going to see a chat widget, and I'm going to open that up, so you can chat there, should you wish to. And then, there is also a Q&A widget, and I'll open that up. So if you've got a more sort of formal question, then you're welcome to ask in either of those. I will try and keep my eyes and ears on it, and then my human colleague, Caitlin, is also on the call as well, and will be able to dive in if I miss your questions. We're gonna follow the same format that we normally do.

Charlie Cowan: I'm gonna spend about 35 minutes or so on our main topic of the day, which is around measuring the ROI of your AI program. Then we will spend about 20 minutes on AI in the news. We will look at some of the new releases and updates in the last 7 days. And then a few minutes for Q&A. But as I say, if you've got questions as we go through, then feel free to just ask them as they come to you. And if you're watching this on YouTube, then do ask your questions in the comments, and we'll come back to you there as well. So what is our main topic of the day? I might actually go back one slide, just so that we can, we can see the actual title. On last week's webinar, we talked about planning your first AI project.

Charlie Cowan: As a company, you want to get started, you're finding use cases, and you're getting started. Now, we're forwarding on 3 months, and now the questions are coming about what's the ROI that we're getting from this. Now, this is an absolutely top priority question that is coming in the projects and the clients that we are working with. As we've talked about on previous weeks, the cost or the spend that is going on AI licenses and tokens is going up and up and up. And if you're a CFO watching, then, as we say, 6 months ago, you might have never heard what a token is, and in 6 months' time, it could be the largest line item in your P&L. And so getting a handle on the return on this investment is absolutely critical. It's easy to roll ROI off the tongue. You know, we need to measure the ROI, measure the ROI.

Charlie Cowan: What does ROI stand for? Return on Investment. So in previous webinars, we've talked about the investment, how do you track the cost of these tokens and the licenses, and today we're going to be exploring a little bit about how you measure the return on that investment. This is what we often see in customers when we start working with them. Maybe they've been doing a little bit of AI work, some AI projects, and what we'll typically see is that when you ask them, like, how are you measuring this stuff, you've got anecdotes. So, you know, Lynn in finance, she's been using it, and she's got a great little skill that helps her with the month-end close, you know, and she says it's absolutely great. You know, Henry in marketing, you know, he's been using a prompt to help him research competitors. You know, he absolutely loves it.

Charlie Cowan: You know, this is fantastic stuff, but you can't take that to the board and say that Lynn and Henry are having a great time with it. The next thing that we typically see is just some usage statistics. So, oh, we've got 150 seats of Claude Team. We've got 40 people a month logging in to use their Notion agent. In our ChatGPT analytics, we can see that there are a certain number of GPTs, or projects, or messages, have been sent out. So, again, it's data, it's usage, but again, you can't take that to the board. That's not a return. You can't say, are we selling more? Are we spending less there? And so if those are some of the things that you have, what we recommend that you actually need is to get a hold of some numbers before you even go anywhere near AI.

Charlie Cowan: As you start thinking about your processes, your use cases, you want to start understanding, what's this costing today, so then we can have a number to react against 90 days after we've gone live. And this is the important thing. A typical situation is after the fact, a CFO says, can we prove this? Can we evidence this? Someone in the IT team or the CIO is saying, look, we've got to go and collect some data. But it's too late then. It's too late. You actually want to do this before you start, and so that's what I'm going to talk through a little bit, about how do you go about collecting those numbers, and embedding that into your process so that you've got the data that you need when you need it down the road. So, here are sort of 5 steps I'm going to walk you through, through today's session.

Charlie Cowan: So, firstly, I'm going to talk about the scoring of your ideas before we even start work. We touched on this a little bit last week. I'll introduce you to the ICE framework. Once you've scored your idea, you then want to lock those in before you actually start work on a specific use case or project. Once you've then delivered that use case or that project, which might only take a few weeks or a month to get that thing built up, we want to lock in what was the delivery. Did we meet our expectations on our ICE scoring? Then I'm going to walk you through outcomes. I'll introduce what an outcome is, and when you might record one of those outcomes, which could be 90 days, it could be 6 months after you've delivered your use case.

Charlie Cowan: And then we're going to look at actually measuring this across your overall process, so that you've got something that you can take back to the board, building this up across your entire AI program, and say, look, here is a number that I'm prepared to stand up against. So let's go through each of these in order. So first up is about understanding what your assumption is going into this specific use case. We talked about this last week, and I'll remind you of it now. The ICE framework. ICE is a well-known framework in product management circles. If there's a hundred things that you could possibly build into your product, how do you decide which things to work on? And ICE is a really easy framework to help you with that. And the ICE stands for I is impact. I'm going to go into that in a bit more detail.

Charlie Cowan: How impactful do we think this thing would be if it existed? C is for confidence. How confident are we that we will achieve that impact? And then E is for ease of implementation. How easy do we think it would be to implement this thing? And what you do, you score each of those out of 10, at 1 to 10, and then you multiply them all together. So you get, in theory, a range between 1, 1 times 1 times 1, for something that's low impact, low confidence, and low ease of implementation, up to 1,000, 10 impact, 10 confidence, 10 ease of implementation. And so if you've then got a list of 50 or 100 ideas or use cases, you've now got a really nice ranking. Now, today, what I want to focus you in on is impact.

Charlie Cowan: When I say, let's say, a specific use case is going to be around monthly reconciliation for your finance team. We've got an idea that we could build, I don't know, a skill or an agent that is going to help with end-of-month reconciliation. So I'm going to ask you what's the impact of that? And, you know, if I asked you 1 to 10, you might come up with, oh, I can think this is 3, I can think this is 5. This often depends on who you are and what your role is in the business. If you are Lynn in finance, and you deal with the monthly reconciliation, you might say, well, this is high impact. It takes a long time, it wastes a lot of my time, and if something could do it for me, that would be high impact. But what we definitely challenge our clients on is, what's the business impact of this?

Charlie Cowan: Not just the personal impact. And when I talk about business impact, are we going to sell more? Are we going to spend less? Revenue up, or cash in the bank going up, or costs coming down with spending less, and therefore, there's more cash in the bank. This is something that 6 months down the line, a year down the line, we can be talking to the board, we can be talking to investors, we can be making different investment decisions because we've got more capital on the balance sheet. This is business impact. Second to that, I would say, is productivity, or efficiency. And this is what we hear a lot of.

Charlie Cowan: Well, going back to poor Lynn in finance, if Lynn spends 50 hours across a quarter working on reconciliation, and she's now gonna spend 10 hours on it, well, there are 40 hours of Lynn's time that has been saved, and could be applied to something else. Now, what we definitely see is that, does that 40 hours actually turn into 40 additional productive hours of Lynn? Does that 40 additional productive hours of Lynn turn into a business impact, that we have sold more, or spend less? Typically, no. We don't see that. And so it can be an easy thing to score on the way in, productivity, but that can be a difficult thing on the way out to go, have we sold more? Did we spend less? So, it's important, but it's not as important as finding a business impact. And then the third and final is adoption. You know, are people using our tool?

Charlie Cowan: We built this agent, we built this skill, we rolled out some ChatGPT licenses, we gave everyone access to Claude Tag, people have accessed it, people have logged in. Once again, can you take that to the board in 6 months' time and say, right, this, we've spent X amount on licenses or tokens, and people are logging in. They say they like it. Well, well done. That doesn't actually change the direction of the business. So, when you're thinking about impact, we really want to focus on that business impact. Revenue up, costs down. Now, when should you be asking for these business impacts? And when should this be a real sort of mandatory criteria? Well, our view is, not at the capture point of view, and not when you're asking people for ideas.

Charlie Cowan: We do a thing that we talked about last week, about running a discovery out to a specific team, or to your whole company, and saying, we want to get ideas from you about how AI could impact your work, the team, the company's outcomes. And at this point, you know, we don't want to prevent someone or putting a barrier in place because they haven't got the numbers about what that business impact would be. So at the capture phase, it's kind of, you know, free-for-all. Give us your ideas. We want to get everything. But the next step is what we call promotion, which is, having scored all of these ideas, we now need some data about what we're going to work on. So we might have done the ICE scoring, and someone may have said, well, you know, I think the, you know, the I, the impact of this use case is 8.

Charlie Cowan: Well, it's at this promotion point where we need that impact slot filled. Why do you think it's 8? Why are you coming up to that? What's the business impact of that? What's the number today? What do you think the number's going to be in the future? How's that going to change the money that's coming into this company, or the money that's going out of this company? What you'll often find here is that people either don't have that number, they can't get that number, or they haven't really thought through what that number would be. It's just, mmm, maybe they'll default back to, well, it takes me 4 hours, and it's gonna save me half that time, so it's gonna save 2 hours. Well, and how often is that going to happen? Once a month. And is it just you? It's just you.

Charlie Cowan: Okay, so what you're telling me is your use case is going to save you 2 hours per month. Not very impactful, versus another one, which might drive more sales. So that really is a gate. We need that impact, and their confidence, and their ease of implementation filled out and scored before we go on. And then at the end of the delivery of whatever that is, a skill or an agent, we want to go through that again and validate, is your estimate, your assumption, correct? So what does that look like in the Kowalah platform? And you don't have to use the Kowalah platform, you might be doing this in just a spreadsheet, but this is an example of what we call an opportunity. You could call it an AI use case, an AI idea. And this is closing the books in 5 days, not 20, which is one that we looked at last week.

Charlie Cowan: And so we've scored this, 9 for impact, 8 for confidence that we could achieve that impact, and 7 for ease of implementation, so a score of 504. And this is the kind of detail that we need to get into the impact. Right, it takes 23 hours per close per entity. We've got 12 entities, and we're closing the books every single month. And finance stops working two weekends a quarter. You know, these are real bits of data that can be fed into what comes later on. Now, how are we confident that we would have that impact if we could save that? Well, here is some detail about that, and we've got some data points in our ERP, and how are we confident, when I say confident, how have we scored the ease of implementation? We've got some data points here.

Charlie Cowan: Before you go through the promotion exercise of turning this into a project deliverable, or an expert request, if you're a managed service customer, you're going to be able to have this conversation internally. Do we believe this? Is it evidenced? If we come back to this in 6 months' time, are we going to be able to point to the right kind of data points? So this is a really nice checkpoint before you go anywhere near starting work on a specific use case. So let's say we've decided we're going to do that use case. We've built it, you've built it, your other SI has built it, and now we've just delivered that skill, that agent, and we do a rescore. And why do we do this?

Charlie Cowan: Well, we want to know that when we did that ICE scoring up front of, I can't remember what the numbers were, 9, 8, and 7, you know, how accurate were we? So, having just completed that delivery, we want to know, right, do we still think that the impact is going to be the same? Now, we've not achieved any of this impact yet, because we've only literally just shipped the deployment, the skill, the agent. But, earlier on, we thought it was going to be 9. Do we still think it's going to be 9, or have we learned something through the process that would change that? In terms of our confidence that we can achieve that impact, are we still that confident, or would we have changed that? And the ease of implementation, well, we thought it was going to be easy.

Charlie Cowan: Did we actually uncover that we needed some data that we didn't have access to? We needed to create a new MCP server that we didn't have access to? Would anything have changed that? 30 days after we've done that deployment, then we're locking that in, so that it's not changing afterwards. Now, the reason why this is so important is that what you should be doing is constantly scoring new use cases that are coming into your backlog, from running discoveries into different teams, different business units, and so on. And you want to refine your ability to make sure that when you're doing an ICE scoring at the start, that you're more and more accurate, so that you're picking the right things to work on. So this is a really nice checkpoint to say, right, we scored it at, you know, 9, 8, and 7, 504. In this scenario, actually, it came out at 360.

Charlie Cowan: So we need to go back and think, well, why did we miss that, and would we change anything in our future scoring? Next up, I want to talk about a thing called outcomes. And so, an outcome is really the proof point. It's the evidence. You might call it the receipt that you can think about, expenses that you might be doing internally in the business. You know, you've got a receipt, and you submit that. That is the proof point of what was spent. So this is the opposite of that. That's the, you know, the receipt for the return. Now, an outcome, you can log at any point in the future. You can log multiple outcomes against a specific use case, and they can be of different types. I've listed a few here. We've talked about them on the way in. Have we won specific revenue because of this use case.

Charlie Cowan: Did we close a specific deal? And the reason we closed this specific deal was because of one agent that we built, and that was influenced revenue, and therefore we can tag that. Did we increase the size of a deal, because we were about to sell it, and a skill that one of our salespeople was using, helped them to negotiate a higher value deal? You know, these are tangible examples. Cost reduced. Because we built this agent, we've now migrated off a piece of SaaS software that we were spending $560,000 a year on. And here is the, you know, the evidence of that from our finance app platform. We were going to hire 3 different business development reps, but because we've now got a BDR agent, we are now avoiding those hires, and we've allocated the spend somewhere else. You know, concrete evidence that we've not spent money.

Charlie Cowan: And then, as we talked about, some lower value ones. Well, maybe there's some productivity. We talk about lovely Lynn in finance. Well, maybe Lynn has, you know, confirmed that she is no longer spending 20 hours a month on closing the books, and she's able to take on other projects that she wasn't able to do otherwise. Maybe there's an adoption milestone. We've rolled out Claude Tag, and we know that 85% of our people have spoken to Claude within Slack at some point. But really, it's these top two. Revenue won, cost reduced. And you can keep adding them, keep adding outcomes as you find them, as you're going forwards. And you tag that towards the opportunity, the deliverable that you put in. Outcomes, super, super important. What I would say here, just looking at some of these example outcomes, depending on how you have staffed your AI operations team.

Charlie Cowan: This is a role for someone, or certainly it's part of a role, is to go on the hunt constantly, finding the people that have been rolling out these use cases, or using these use cases, and going and finding real scenarios. If you've built a, you know, a sales agent, and it is helping your sellers to negotiate and close deals, go and speak to your sellers, and find out examples of where that agent has helped, and find deals where they were using the agent, and they would happily link that agent to that specific deal. This is a research exercise. Go out, find these things, and get them tagged as outcomes. The final thing that I would add on, on this sort of ROI topic is, so far, we've talked about individual use cases. We've got an opportunity, we've scored it, we've built it, we've done some tagging of some outcomes on the, yeah, after the fact.

Charlie Cowan: But a use case is just one part of an overall process, and we're big believers that, as you're thinking about your AI program, you should be thinking about your company as a collection of processes, and each of those processes have got process steps. And each of those process steps is an opportunity for improvement and to get some business outcome. So, as well as looking at the individual use cases, take a look at the overall process. We think about an individual process step as being one of these four kinds of dispositions. So it might be human, at the start, very manual. A human does this thing, they, you know, pick up the phone, they send an email, they move information from one system to another. The next step up is AI-assisted.

Charlie Cowan: And so now that human is using a skill, maybe they're using a plugin, they're doing something that is complementing their work, but it's the human that's doing the work, certainly leading it. Then you've got full automation. So this is, I don't know, maybe you're using Zapier, maybe you're using N8N, a deterministic workflow, that is handling this, but there's no AI involved in that. And then the fourth is AI autonomous, so you can think of this as agents. This is where an agent is handling that process, and they're only reaching out to humans for exceptions, maybe some human-in-the-loop approval, but ultimately, the agent is the one running the process. And as we look across your entire company, all of your processes, all of your process steps, then we're looking generally to move things from the left to the right, and these are AI sort of development opportunities to move things from left to the right.

Charlie Cowan: What we tag up in these processes, in these process steps, is how many hours a human spends on that process today, and how many hours a human would spend on that process step in the future, maybe because it's moved to, yeah, AI-assisted or fully autonomous. So using our example of this month-end close process, we can look across the whole process and say, well, at the start, there was 44 hours per month, per entity, of human effort on this overall process, and 21 afterwards. We can then start to add all of that up, and say, right, well, by saving 23 hours per close process, per entity, that's times 12 entities in this mythical organization, times 12 closes, because there's 12 months in a year, at an assumed hourly cost for Lynn, we've got a concrete number here that we can take back to the board, for this one use case.

Charlie Cowan: So this is just kind of a different lens. Earlier on, we were talking about use case by use case, and here we can look process by process across the whole year, for that organization. So, what does it look like to bring all of these five steps together that we talked about at the start? So, number one is scoring your opportunity, your idea, your use case before we start. So, we looked at, what did we say? I can't remember the math, but 9 times 8 times 7, or something like that, gets you to the 504. We think that this is going to have that kind of impact and confidence and ease in the business. We then get all of the detail, the evidence about why we think that.

Charlie Cowan: We're going to find numbers from our ERP, we're going to find the source material that is going to give that evidence, we can then get a sign-off, and then we're going to promote that use case, and we're going to start work on it. Having built the agent or the skill, we're then going to do a rescore. Actually, now, we think that this is slightly lower impact, our confidence is the same, and we think slightly lower on the ease of implementation, because of what we've learned. We can feed that back so that our future estimates get better and better. 90 days later, maybe 6 months later, we're adding in outcomes. We've gone off and we've found out that we closed this deal that we wouldn't have closed. We've got rid of this SaaS platform that we wouldn't have got rid of.

Charlie Cowan: We have avoided 3 hires that we would have hired, and we can add outcome, outcome, outcome, and we can log these against the initial use case and the project. We're then able to roll all of that up, looking across your entire process. In this case, it's going to be your financial, sort of month-end process, and we can start putting actual numbers to that, that can go into your line item, into your P&L. So now, you roll that up across not just finance, but marketing, sales, not just EMEA, but the US and APAC, not just this business unit, but that business unit, and you've got something that to balance out the I on your ROI, of the tokens and the licenses and the spend, you've now got some R return to be able to have that conversation. Now, this is such an important topic to be thinking about.

Charlie Cowan: I was chatting with the CEO of a client the other day, and we were talking about this topic of, oh, you know, we're going to be spending hundreds of thousands towards millions on tokens this year. And I said, yeah, but if I could tell you that for every $100,000 you spend, I can show you 500,000 of return, what would you do? And he said, I'd give you another $100,000, and another, and another. As long as you can tell me that I can turn $1 into $5, then I'm all in. And this is what this is about, and hopefully, as you see through this process, the whole way through, you're gonna have those numbers, as you go through to be able to give either to your senior leadership team, to the exec, and to the board. So what should I recommend that you do between now and next week, if you're inspired to get started?

Charlie Cowan: Well, the first thing, you're probably working on a few use cases right now, so go and think about those use cases, and then go and work out what you think the impact is for that use case. Give it a number. Where's the evidence? Where's the data that you have got that is going to give you that business impact? Don't get distracted too much by productivity, or efficiency, or adoption. Are you going to sell more? Are you going to spend less? And if not, is it the use case you should be working on? With that, let's move on to AI in the news, and quite a few things going on over the past week, so let's dive in. If you were on last week's webinar, you will know there has been some big shakeups in Google. Google's main AI platform is called Gemini.

Charlie Cowan: And they have a division called DeepMind, which they'd acquired many years ago, which is really where a lot of the brains of the AI program at Google had come from. Last week, we talked about how there was being a big sort of leadership shake-up at Google, with the CEO of DeepMind stepping into a chief scientist role, some of the lead scientists there moving outside of the organization. And really, it's a big shake-up at Google to go, look, why aren't we really a frontier challenger that's taken seriously in the AI space? Who do I put in that? Well, I put OpenAI, and I put Anthropic in that space. Certainly in the, yeah, well, yeah, that's who's there. And organizations are not really seeing Gemini as a credible threat to that. So, here come some of the new releases.

Charlie Cowan: Only, I want to say, yeah, 3 weeks ago, 3.6 Flash came out, and then this week, 3.7 Flash came out. Now, Flash is their more lower cost model that is coming out, and so Gemini really pushing on the availability, pushing on the lower cost. And the fact that for Google Workspace customers, well, you've already got Google, you know, we can give you a bit of Gemini for a small additional fee, you can, you know, upgrade to get more access, you know, and, you know, why would you bother with OpenAI or Anthropic? I have to say, in my world, in the Twitter communities that I'm hanging out in, I'm not seeing too many people that are using these models day-to-day, but it's important to know that Google is doing everything they can to try and, yeah, push, push, and push forwards in the space. Next up is OpenAI, and this is about speed.

Charlie Cowan: This is a limited preview of their ultra-fast mode, of their new model, GPT-5.6 Sol. Now, if you've been on our last few weeks, GPT-5.6 Sol is their most powerful model, and is being used for a lot of cybersecurity and more complex use cases. But these more powerful models, whether it's 5.6 Sol from OpenAI, whether it's Fable from Anthropic, are taking time, because they are very sort of intelligent and reasoning models that take time to think through their response. And a new sort of battlefield is how fast these models can run. So, Anthropic have a fast mode, and I think I'm recalling that if you have fast mode on Opus, it might consume two and a half times the cost, something like that. So, it's a significant expense on an already expensive model. And this is the first attempt of OpenAI to go down the same route.

Charlie Cowan: What jumps out to me here, 14 times the speed, so it's not even like, you know, 2 or 3 times. It's super fast, 750 tokens per second. And some of the use cases that OpenAI are talking about here, so immediate incident response, sort of finance research, you can think about some of these sort of share trading platforms, algorithms, where speed is the order of the day. The fact that you can get models to return at this significant increase in speed is going to be a new battlefield. Third up is SpaceX, formerly known as xAI. As you may have seen over the past few weeks, Elon Musk has been rolling up a few companies into one. So, xAI was the name of his AI company, and they ran through a merger or acquisition to fold that in under SpaceX recently.

Charlie Cowan: So now, what was xAI is using SpaceX's data centers, and there's a lot of, you know, benefits for Elon's team over that. They've also purchased a company called Cursor for about $60 billion. We should probably do a bit more detail on that in another webinar, but Cursor, I'm making this up, but maybe they're 3 years old? Maybe it's 2 and a half years old. Not an old company at all. It is a development platform for engineers, and it's been purchased by SpaceX for $60 billion, and that deal closed just this week. What that gives is SpaceX a really strong seat at the table, because they've got the data now, they've got the data centers, they've got the capital, and they've got a development platform that so many engineers are using right now. Last week, we talked about Grokbot being launched, which was a product developed by Cursor, but has been renamed as Grokbot.

Charlie Cowan: And at the same time, they have launched Grok 4.6, which is the new model that is now running from SpaceX. Now, what I think is really interesting here is that over the last week, I've seen so many positive mentions and comments from, I say quasi-independent, commentators, through my X and LinkedIn feeds, about Grok 4.6, and also Grokbot as a, as a harness, as an agentic tool. I think it'd be fair to say that only a few weeks or months ago, people had counted xAI out. A lot of senior people, original founders from there, had left the business. People felt that, you know, Elon was, you know, struggling to have a position in the AI world, and that he was subletting his data centers to Google and to Anthropic, which he is. And now, the vibe that I'm hearing is that, no, we've got a serious contender here.

Charlie Cowan: Now, most of our audience for these webinars is enterprises, and am I confident that an enterprise company is going to connect up their CRM, their ERP, into Grokbot and Grok data centers? You know, I'm not sure. But the general vibe is that this is a very competent model, competent tools. And a vast majority of developers are using Cursor already, so they're in that platform. So definitely one to keep an eye on. Fourth story this week is about an acquisition, which I think has just closed. I don't know if I've seen an official announcement, but as you can see here, it's been reported by Bloomberg. So, Stripe, you may be familiar with, is a financial platform that helps developers, SaaS platforms, product companies to manage their payments flow. So, if you've ever bought something online, you've probably bought it through Stripe.

Charlie Cowan: Basically, it was an API for finance systems when, before Stripe, it was very, very difficult for developers to get payments into their applications. Now, what's interesting here is that Stripe is buying OpenRouter. OpenRouter is a company that makes it easy for developers to connect to multiple models. So, if you imagine you're a developer, maybe you're building a platform, you want it to be an AI-powered platform, now, through OpenRouter, you can have one gateway with one API, and then you can have rules that point to different AI models, depending on what the request is. Now, this is becoming really, really important, because as token costs go up, as we talked about in the ROI section, companies are trying to figure out, well, do I need to route every single request to Fable? Do I need to route every single request to GPT 5.6 Sol? Probably not.

Charlie Cowan: And so OpenRouter can be a way to provide this a gateway that allows you, as the company, to direct a little bit about where that goes. So, really interesting. Stripe, still a private company, would have positioned itself as sort of the finance gateway for developers, is getting into the AI gateway for developers. Could be a really interesting direction for that company to go. And if you are an enterprise, you're very likely to be an OpenRouter customer, or a similar type of gateway, gateway routing platform. And then the final bit of AI in the news this week is an announcement that Anthropic is now going to be watermarking all text that comes out of any Claude model. So this could be in Claude Desktop, it could be in Claude Tag, it could be through the API, it could be something you're generating in Claude Code.

Charlie Cowan: If you're writing text, or using Claude to write or edit text, then it is going to come with a hidden watermark that would allow other people to determine that Claude was involved in the creation of that text. Now, this is off the back of a thing called the EU AI Act, which companies don't have to qualify what I'm saying, so I'm not exactly sure on that. I was going to say they don't have to subscribe to it, but Anthropic has elected to subscribe to it. What they've also elected to do is to make this not just an EU thing, but actually, whether you're a user of Claude in the US, Asia, anywhere else, then the same technology is going to be in place. Now, the key thing here is that it is a signal. It is not evidence. So, two sides to this.

Charlie Cowan: One, if you have a watermark, it does not mean 100% that Claude wrote that content. It means that Claude was involved in that content, so you might have written something, you know, 30 pages of a report, and then you've gone to Claude to get it to help edit and proofread it, and maybe make some tweaks to some of it, and you're going to carry a watermark because of that, even though you've done the core of that work. The opposite is also true. If there is no watermark, it doesn't 100% mean that no Claude was involved in that, because it might have been maybe too short, or maybe it's gone through 3 other different platforms before it has got to you checking for a watermark. So I think this is, you know, early stages for this. I would keep an eye on it. I wouldn't be too, you know, worried.

Charlie Cowan: You know, I've got to keep, you know, checking if watermarks are here, and I can't use Claude for anything that I'm doing on. But just get an idea that this is the general direction of travel, and I'd expect to see other providers that are gonna roll out similar kind of watermarking features. All of my feeds, whether that's X, LinkedIn, are all about AI slop. How do we get rid of AI slop, where people are not caring about their work? This is not about don't use AI at all. It's just don't be lazy. Do your research, write the first draft, work with Claude, do your iterating and your proofreading afterwards. Keep the maintain the quality of your work, rather than just creating slop. So, with that, I've not seen any questions that have come through online. There's no worries about that.

Charlie Cowan: If you're watching this on YouTube, then feel free to add them into the comments section, and we will come straight back to you. Next week, we're gonna get into a product that I've talked about on the last couple of weeks, which is Claude Tag. So Claude Tag is a really interesting, I guess, a transition from giving every single one of your people access to an AI license. Now, that could be Claude, but it could be ChatGPT, it could be Gemini, it could be Copilot. We've often thought about, oh, people need to be given a seat, they need to be given access to it, and how do we decide who has one? How do we monitor it? How do we make sure that they're using the right, you know, the right models? Claude Tag actually gives all of your people access to Claude via Slack.

Charlie Cowan: Now, of course, you've got to be a Slack client, but this will feed in in the future, no doubt, to Teams and Google Chat as well. But the idea here is that instead of giving everyone an individual user, you actually create Claude as an individual license using API, and then everyone's able to chat with Claude from any channel or DM directly, and you pay in a slightly different way. This gives you some great new control that you don't have if you just give everyone access to their own license. So, we look forward to talking to you all the way through that next week, what Claude Tag is, how it works, how to set it up, some of the controls, and some of the pros and cons that you might have over giving everyone an actual license. With that, thank you very much.

Charlie Cowan: I hope you are now, well, sort of set up to go improve the ROI on your AI use cases. Look forward to hearing how you get on. I'll chat to you next week.