Redesign the Workflow, Not the Task

Charlie Cowan: Good morning, good afternoon, depending on where you are dialing in from, and welcome to this week's edition of the Kowalah Wednesday webinar on the 2nd of September. And, certainly here in the UK, the weather has taken a turn, it's getting a little bit cold, the kids are going back to school. And it feels like everyone is getting back to work and figuring out what they're gonna do in the world of AI. And that is exactly where our Wednesday webinars come in, to give you a time each week to take a breather, to focus on what's going on, get an update, and think about how all of these advancements in the world of AI can be applied back into your business. And we'll get on to talking a bit more about our topic of the day in a moment.

Charlie Cowan: But if this is your first Wednesday webinar with Kowalah, we get together every week, and we follow a similar flow. So, I'll spend about 35 or 40 minutes or so talking about our main topic of the day that I'll introduce in a moment, around redesigning your workflows, rather than picking individual use cases. We'll then have our recurring segment of AI in the News, and we'll take a look at some of the top stories from the last 7 days. And I have to say, this week, there have been quite a few, some in the last 24 hours that we're going to talk through. And then we've got an opportunity for you to ask any questions. Now, if you're watching live, then feel free to ask your questions as we go through. You don't need to wait until the end.

Charlie Cowan: If you've got a question about what we're talking about in the main topic, or AI in the news, then feel free to chime in. You're in a Google, sorry, a Zoom event. And you'll see, depending on your device, you've got, at the top or bottom of your screen, both a chat widget, where you can chat, either to me or to others that are on the webinar, or a Q&A widget, where you can ask a question, and that will pop up. I will keep an eye on it. I've got both screens up, so I'll keep an eye there. And my colleague, Caitlin, is on the call as well, can assist if I miss anything. If you're watching this on YouTube, then feel free to ask any questions in the comments, and we monitor that, and we'll come straight back to you as well.

Charlie Cowan: So with that, let's get into today's main topic, and this was inspired by a post that I saw on X a few weeks back, and let me actually go to the actual post, which was, which is here. And it was, let me see, when was the date? July the 7th, so it was at the start of the summer. And it was from, I believe, the head of product at, so, chief technology officer at Uber. And he's talking about how they've been driving AI adoption, and specifically agentic AI adoption across the organization. He was saying that Uber, as you would expect, is heavily using AI across their development teams, their engineering teams, but what this post was about is about driving AI adoption beyond engineering into finance, legal, operations, and beyond. And so, in his post, he talked a little bit about the process that they've gone through.

Charlie Cowan: And this really struck me, because it's something that we see, and something that we believe, and that we recommend to our clients. So, he talks about pairing an engineer who's completely AI-pilled, completely gets their way around Codex or Claude Code, and knows how to actually do something with an idea, and to go and pair them with a domain expert that is in the business unit. So that's someone in finance that knows about the finance process. That's someone in sales that knows about the sales process. And they run this two-week sprint. And so what happens in that two-week sprint? Well, the first couple of days is just shadowing that domain expert. We sometimes call this ride-alongs, you know, if you are going to spend time with a salesperson, you might spend time joining their sales calls, or going out on their sales meetings.

Charlie Cowan: If you're spending time with legal, you might watch them as they negotiate a contract with a client. So ride-alongs, a day in the life, but something where this engineer gets embedded in what that functional expert is doing. Having mapped out what the overall process is that that function or individual is working on, then it's about prioritizing individual opportunities within that process that are going to have the most business impact. And we're going to talk about this a little bit later. Having prioritized the use cases in that overall process, then it's about building. You know, is this a hackathon? Is this a mini sprint? I've heard this called a bolt, you know, two days of building stuff, to focus on those initial use cases. Validating what has been built. So, yes, with that individual domain expert, but with other people in the team. And then, shipping. Shipping something on day 10.

Charlie Cowan: And then what iterates into the second week is a little bit of monitoring that, and seeing what needs to be improved in days 11, 12, 13, and 14. So, 2 weeks, get this sprint done, and repeat that, repeat that, repeat that. So, I thought this was pretty interesting to see. He also shared in his post, and we'll give you a link to that post so you can get this graphic directly, he then called that an agentic pod. How have we driven AI beyond engineering? And so, it's that process with that timeline I shared you. Pairing, understanding, identifying, building, validating, and shipping, in that two weeks. So you might be watching this and going, right, Charlie, great, this, I need to go and do this. We're not Uber, we're not, you know, heavily backed and with a team of thousands, so how are we going to approach this ourselves? Which workflows are we gonna pick?

Charlie Cowan: If we're going to pick one or two, how do we know where to focus our efforts, if we've got these constraints? So I'm going to show you two different lenses, that both we're using, and that I'm going to give you the source material that you can then, you know, follow that yourself. So the first is a fantastic book called Rewired, and I have Rewired on my desk as I speak. It is my number one recommendation to CEOs that we partner with at the moment. Read this book, because if you read that book, you'll have some great best practices, you'll also know how we're thinking. And it'll allow us to have better conversations. Now, this book was written by the team at McKinsey. McKinsey, as you may well know, large management strategy consultants working for some of the top companies in the world. And this book, second edition, came out in April of this year.

Charlie Cowan: And it is really written for a CEO of an organization. It's not written for the CIO, for the CTO, although they're welcome to read it. It's written for the CEO, the person that is tasked with driving the success of the business, and needs to lead the company into this new world of AI. Now, I'm gonna give you a couple of images and graphics from that book. But the real premise is that, where a lot of companies have worked with AI over the past year or so has been to let chaos reign by just experimenting, giving people licenses, coming up with some use cases, see what bubbles up, and we get some value. And what Rewired really lays out is that from the very top of the organization, we need to have a business strategy about this, and pick the two or three big fires that are going to drive real sustainable value across the organization.

Charlie Cowan: So how do we do that in the organization? Well, in any industry, whether you're retail, CPG, hospitality, manufacturing, there are going to be a couple of levers, or leverage points, that define the way that your industry makes money. That may be the way that you convert leads into sales. It may be the way that you deliver your service to your customers. It may be the way that you're able to price your products. There are things that are right at the heart of how a company in your industry derives value. And you can't change those. You assume them as a company. Let me go back here. So having done those, you're then defining your business strategy. Now, this is not an AI strategy. AI might be part of this, but you've got a business strategy, and that might be about opening up in new markets.

Charlie Cowan: It might be around opening up in new regions, launching new products, changing your pricing, hiring great people, but you'll have a business strategy. So, having done your business strategy, you're then gonna set your goals, and you may follow OKRs, objectives and key results. You may follow V2MOM, vision values, and all of the rest of that. But, whatever that is, you're going to have some goals, probably on a quarterly or an annual basis. What Rewired then drops you down to is thinking about your business as a collection of domains. What is a domain? Well, a domain is typically an end-to-end process, so it could be your lead-to-cash process, it could be your finance process, it could be your hiring process.

Charlie Cowan: It could be, you might call it a customer journey, you might call it a value chain, but it's this area of your business that, when you describe it, everyone knows exactly what that thing is. So having described your domains, the different areas of your business, you're then going to pick 2 or 3 of those that are the most impactful in your organization. Within each of those processes, there's then going to be a couple of levers that really drive the value in that individual domain. And so you could think of these as the KPIs. You might have your OKRs up here, your goals, your objectives, and key results. Down here are the KPIs, the numbers that that individual domain tracks.

Charlie Cowan: So, in sales processes, that might be your conversion rate, it might be the size of your deal, it might be the time to close. These are the levers that, if you move them in the right direction, are really going to change the success of the output of that domain. Then you go down one level further, and you start thinking about, right, what are the solutions, the things that we could build that would sit within that domain that could drive those different levers? So what is the thing that we're going to design and to build? And within that, there may be 4 or 5 different use cases, and by use case, I might mean, you know, a prompt, a skill, an agent, something that sits within that. And so Rewired lays out this whole sort of cascade that you can fall down as you decide what to build.

Charlie Cowan: And so, let's get to this slide that was flashing up earlier. So let's give an example of that. So, as a domain, an end-to-end process, you could think of as lead to cash. So, as a B2B organization, you may get leads in from your marketing team, and you then take them all the way through your sales process before you sign a contract with that customer, and you pass it on to another domain, which would be your customer onboarding and service delivery domain. So, in lead to cash, well, what might be some of the value levers that sit in that domain? Well, here's a couple. Well, if we win more qualified deals, that would be a lever. You know, win more deals, we're gonna make more revenue, that is an important lever. And what about closing deals faster?

Charlie Cowan: If instead of it taking us 12 months to close a deal, we close it in 5 months, well, we're likely to lose less, because time kills deals, and we're likely to be able to give a better customer experience because we're able to keep their momentum up. So those could be two value levers. Dropping down one level, well, what might be a solution that supports that value lever? Well, on winning more deals, maybe we could come up with a lead scoring engine that's gonna help us to focus on qualified deals, and we spend less time running after things that we're never gonna win. And maybe we can win more, because if we were able to accelerate and automate our proposals so they're a higher quality, they're more consistent, well, that would be a really good solution.

Charlie Cowan: And then within each of those, there may be 2, 3, 4, 5 different use cases that we're going to build. And here, I've described these all as agents, and they may well be agents. We might come up with a lead triage agent, a lead scoring agent, a lead contact agent. These could all be discrete use cases, using different technology, different data sources, different outcomes, but they're all connected to this one solution. Now, there are a few reasons why in the book Rewired, they say this is such a great approach compared to lighting a thousand little fires. If we look at this lead scoring engine and these three use cases, number one, we're solving a process end-to-end.

Charlie Cowan: Yes, lead to cash, but this winning more qualified deals, we're going to be able to solve that end-to-end process, because we're providing use cases that go along the whole of that process, instead of doing a little bit in finance, a little bit in legal, a little bit in HR. So, we solved the process end-to-end. Secondly, it's very likely that these use cases are all going to rely on the same data sources. Well, we're talking about leads, we're talking about our sellers, we're talking about the content that goes into our proposals and our pricing. If we build the data foundation once, all of these agents are likely to be able to use the same thing. They're likely to be able to pass the output of one agent straight into the next one. So we get a benefit from that. And the third benefit is change enablement.

Charlie Cowan: It is one team that are focusing on this process, one human team, and so having built these three or four or five use cases around a lead scoring engine, we can teach one group of people how to interact and change the way they work in this use case. If we build something for finance, something for legal, something for HR, we've got to run around and teach a small group of people how to use one process that might not have the right data, and we just lose that economy of scale from focusing all of our efforts in one area. So that's Rewired, and I absolutely recommend that you read that book. It is, like I say, my number one recommendation for a CEO that is driving AI transformation across their organization. Now, in that, we talked about domains. A domain, an end-to-end process, a value chain, a customer journey.

Charlie Cowan: How do you decide which of those domains you should focus on in your organization? And I want to introduce you to another book that I highly recommend if you are a CEO or any senior leader that is responsible for leading a team. And this is called Playing to Win, and I think I want to say this came out in about 2014. It might even be older than that. It is a relatively old book, definitely pre-AI. But it was written by A.G. Lafley, who was, for many years, the CEO and chairman of Procter & Gamble. And then his consulting sidekick, a guy called Roger Martin. And in it, they codified strategy. They codified the strategy that they use to run Procter & Gamble. And it is just a fantastic book, and it is applicable into any organization. So what is this codifying of strategy?

Charlie Cowan: Well, they describe it as five cascading questions, and I'll walk you through each of those now. So the first is, what is our winning aspiration? If we're going to build a strategy, we need to know what the goal is. How do we come up with a strategy if we don't know where we're aiming to get to? And so, with our AI lens in it, our question might be, what does this company look like if AI genuinely worked? And the key thing here is, this is not what is our AI strategy. It's not that we've got a strategy that everyone adopts AI, everyone knows how to log into Claude. This is, what is our company strategy? What does our company look like if AI genuinely worked, in terms of the experiences we provide to our customers, our suppliers, what markets are we in? So, what is this going to enable, unleash for us.

Charlie Cowan: Second up is, the second question is where to play. And so when we talk about where to play, we're thinking about what markets do we want to play in. We're thinking about what geographies do we want to play in. We're thinking about what parts of the value chain do we want to play in. And so, when we think about our AI decisions and picking these domains, we're thinking about where can we have the most amount of impact. Let's not go and focus on deploying AI into every single one of our teams. Let's pick which ones we want to focus on. And this is one of the reasons why I love this book, because the authors talk about strategy being about making painful choices.

Charlie Cowan: So, if your choice is every domain, and I'm going to show you which domains are available in a minute, but if you say, well, we're just going to do everything, we're not going to make that choice, you're not being a leader, you're not defining a strategy, you need to pick, and you need to say no in order to have made that choice. The next question is how to win. So, having chosen our where to play, we now need to say, in those where to plays, how are we going to win? So what is it that we have got that no one else can get that gives us this unique right to win in that where to play decision? So for us, that is going to be thinking about, what data are we sat on that gives us access to something that our competitors don't?

Charlie Cowan: Is it the approach that we're going to take to embedding AI into our workflows? Is it how we're going to reconfigure our teams to be able to go faster? What is our how-to-win choice? And I'll come on to that in a second. Next choice is our must-have capabilities. What are the things that have to be true for us to enable to win in our chosen markets? So, do we need to have certain platforms? Do we need to have certain people? Do we need to have certain data? What is it that we need to get to the start line to be able to have that? And then finally, how are we going to track our success? And they call these management systems. So, do we need a cadence of reporting? Do we need certain KPIs?

Charlie Cowan: Do we need certain OKRs that we are going to set, so that, 6 months down the line, we can look back at what we've done, and we can know that we have been successful? So these five questions are at the core of this book, Playing to Win. What is our winning aspiration? What are our where-to-play choices? What are our how-to-win choices? What are our must-have capabilities? And what management systems must we need? And a great way to think about this is that your where-to-play and how-to-win choices are actually intertwined. When you choose one market where to play, that is tightly aligned with its how to win. And so I've given you some sort of examples of the questions that you might ask when you place the AI lens over these choices.

Charlie Cowan: So in terms of where to play, we're thinking about which functional domain, which processes are you going to be focusing on? What are the different steps in that process that make up that process, where you can focus? And what kind of data have you already got that would be useful to making those decisions? And then within that, how are we going to win? Well, by applying AI, by applying agents, how are we going to do something that our competitors absolutely can't do? So I definitely recommend that book. Now, I've talked a few times about domains and, you know, how to pick which one. You may have come across this before, maybe not. This is from an organization called APQC. And if you Google APQC, you will find their process categorization framework.

Charlie Cowan: Now, it's open source, you can chuck in your email, and you can get the PDF article, which describes all of this in great detail. They'll give you Excel spreadsheets, which describe each of the different sub-processes in great detail, and they've even got industry-specific ones, so I definitely recommend that you take a look at that. But what they've done is break down, and this is their cross-industry taxonomy, every company is a collection of processes, and at the very top level of these sort of domains or process groups, you could call them, are these operating processes of 1 to 6. So, as a company, we have to develop our vision and strategy. We've just talked a little bit about that. Having developed our vision and strategy, we need to develop and manage some products and services. R&D, we've got to build something to sell.

Charlie Cowan: Having done that, we need to go and market and sell those products and services to our customers. Having sold a customer, well, we need to make sure that we've got physical products to be able to deliver to them, if we're in a CPG or retail, so we need to manage our supply chain. If we're a professional services organization, we need to deliver those services. And we need to manage our customers, and make sure they have a great customer experience and they buy more from us. Those are the end-to-end, top-level processes. And supporting that, we've got a number of management and support services. So, I won't go through each of these individually, but you've got things like managing our finance team, we've got things about managing our people and acquiring great talent, managing our IT, managing our partners, making sure we've got the right risk and IT frameworks in place.

Charlie Cowan: And any company is going to have each of these running. Now, individually, between each of these, you've got all of the sub-processes, and it goes down 3 or 4 levels. It's a great way of taking a lens on your business. So you could think about these as domains. Now, my point here is that, when you start to look at your company like this, you start to think, well, where is the value lever, in our industry? Is it in managing enterprise risk, compliance, remediation, and resiliency? If we get that really right, is that going to change the direction of our company? Hmm, maybe not. If we manage our information technology really, really, really well, is that going to double our revenues? Hmm, maybe not.

Charlie Cowan: I would tend to recommend that it is these operating processes up at the top that are the things that are really going to change the direction of your business, and are the things that you might want to focus on as your key domains. This is a graphic from Rewired, which helps you to think through this, and they talk about balancing potential versus feasibility. So, on the y-axis, going up, you've got from low value to high value. You know, if we get this thing right, this domain, this process, will it be really valuable to our organization and the goals that we have got as a company? And then on the x-axis, left to right, low feasibility to high feasibility.

Charlie Cowan: There's absolutely no point coming up with something that is really, really highly valuable, but is very, very difficult, because we don't have the data, it requires third parties to get involved, customers wouldn't accept what we're going to create, employees wouldn't accept what we're going to create. So this is a nice sort of lens to, once you've figured out what your top-level domains are, to then lay them out and go, right, high value, high feasibility, let's focus on those. A couple of lenses that you can place on figuring out the value. Number one is the scale of that domain, or of that process. Is this something that affects a lot of people across the organization, whether that's internal, whether that's customers, how many people spend time on that?

Charlie Cowan: So think about your sales organization, where you may have hundreds or thousands of sellers interacting with tens of thousands or hundreds of thousands of customers, big scale. Think about your finance team that are working on closing the books at the end of the month. You know, it's going to affect a small number of people that are working on that process, so scale is one lens to think about. Next up is repetition. How often does that thing happen? If this is once a year, you define your annual strategy and OKRs, yes, it might be important, and it affects a lot of people, but it's not repeating on a daily or weekly basis, and so is that, you know, a really valuable use case?

Charlie Cowan: Whereas, maybe, you know, how you support your customers or service your customers is happening on a daily basis and repeating, you know, very regularly, and would be, we have high value from that perspective. Business impact. Now, this is why I talk about the leverage points for your business, I talk about OKRs for your business, I talk about the KPIs in your business. When your leadership team get together, when you have the board meeting, what is the dashboard that they put up on the screen? What are the numbers they're looking at? You know, is it pipeline? Is it sales? Is it average order size? Is it support tickets? Is it time to first resolution? These are the things that you want to tie your AI processes and domains towards. You know, is the leadership team looking at, you know, I don't know, number of open job descriptions? Maybe, maybe not.

Charlie Cowan: Focus on those KPIs at the top, because that is what is going to make your AI program be impactful. And then the fourth one is a bit about feasibility. Are the agents going to be able to see what they need to see to be able to do the job? If not, if this data doesn't exist, if it's impossible to get hold of, if it's very fragmented, it's not to say that that is a no-go, but this also comes into filtering. If there's another use case that's got the same scale, repetition, or business impact, you're gonna make a decision over who's got the best access to data. This is a quote also on one of the pages in the book, Rewired, which I love. Pick battles that are big enough to matter, and small enough to win. And that is that phrase, or that chart, about value versus feasibility in a nice caption.

Charlie Cowan: Pick battles big enough to matter, and small enough to win. So, having picked your domain, and in this example, I'm choosing a lead-to-cash domain as being the one that would be important, because it's about driving pipeline, it's about driving sales. And then within that, you've got the multiple sub-processes, and here, I've just mapped out a process to go from an inbound lead to a qualified deal. Oh, sorry, to a qualified lead that is going to be handed off to a salesperson. So, a very small sub-process in a sales team. And then what you're gonna do is you're gonna map that out. You may have this already mapped out in Swimlane somewhere, or you can do this in the Kowalah platform, we can help you to do this, or you can just use a whiteboard and some sticky notes is the best way to start.

Charlie Cowan: But you're gonna map out the steps of those process, of that sub-process, and you're gonna say, right, well, an inbound lead lands from marketing, someone came to an event. We're then going to enrich it in some way, with who the client is, who the contact is, what do we know about them, what industry they're in, all this kind of stuff. We're then going to draft some outreach, we're going to maybe call them, we're going to send them some messages, we're going to connect with them on LinkedIn. We're then going to call them, a human is going to call them, we're then going to book a meeting, and we're going to hand it off to a salesperson who can take it forward. For example, that would be a very sort of standard initial process.

Charlie Cowan: For each of those sub-steps, you're then going to give it a disposition as to what happens today, and this is what is along the line here. Automated, human, human, human, automated, and human might be your current state. But what you're then looking for are what are those levers that, if we can improve that, it would be really, really valuable. So getting through this process really quickly by AI handling this, an agent to enrich and score, an agent to draft the outreach, an agent even to do the outreach, could well be how we can drive this and get a meeting booked with that customer in minutes, rather than waiting for a human to do that work, and we lose that prospect, because we know that time kills deals. So that's the key thing.

Charlie Cowan: Domain, map out the process, map out the steps in the process, and then start thinking of the individual use cases that can support along that overall process. What you'll then be able to do across your organization, having picked those domains, is to go, right, sales and marketing is our number one domain, and what are the AI opportunities and the process gaps within that? And so, definitely think about looking at your company through that lens, rather than just individual use cases. What I would recommend, as we come to the end of sort of the main section on that we've talked about today, is pick one function, pick one domain, have a think through some of those lenses that we've talked about, whether that's sales and marketing, whether it's customer service. Pick someone that understands AI in your organization.

Charlie Cowan: Maybe they've come from your engineering team, or maybe you get external support for that, and go and do that day in the life, just like we saw from Uber. Go and spend a couple of days doing a ride along, go and see how they work, understand it, map out those processes. Either on a flipboard, or in the Kowalah platform, map it end-to-end, figure out where AI could be really, really meaningful, and then start working, building those use cases, and building it end-to-end. We'll put the links to these books in the notes to this webinar, but the Rewired one is my number one recommendation, and then Playing to Win for just a higher-level view on strategy, absolutely. Okay, so that brings us on to AI in the News, which is our recurring segment to look at what has been launched or updated in the last 7 days.

Charlie Cowan: And there is so much going on at the moment. I've just picked out 3 or 4 things that have caught my eye over the last week. So the first, and this happened just last night, Anthropic launched two new models, upgrades to their real top-level, largest models, Fable and Mythos. So just to bring you up to speed, if you're not too close to this, Mythos is their top-level model, which is actually restricted only to people that are working in cybersecurity and the sciences, working on sort of clinical research. I mean, it's so powerful that they've wanted to restrict it, because if it gets into the hands of bad actors, then there are severe cybersecurity risks. What they then did was to release a model called Fable, which you could call a sort of Mythos Lite.

Charlie Cowan: It is the Mythos model, but with some certain constraints in it to prevent it from taking certain actions if you were to try and focus it on cybersecurity, it would say, you know, I can't work on that, you need to pick a different model. Now, since Fable and Mythos came out, there was some vibrant feedback, I would say, from the community about how many times Fable fell back to Opus as being a different model, about how many times it said that it can't work on certain tasks, and just in terms of the speed and also the cost as well. And so, last night, Fable 5.1 and Mythos 5.1 were announced. You can go and take a look on Anthropic's blog for more of a description of what it does. But this image that I've grabbed from the blog post summarizes it fairly well enough.

Charlie Cowan: All you need to know is that the y-axis going up on the left is the score on agentic coding, so on writing software. And so, the higher up it is, the better it is. And then the logarithmic scale on the x-axis, so logarithmic, meaning that it just gets bigger and bigger and bigger as it goes further to the right, is the effective cost per task in US dollars. And so here, further to the left is better, bearing in mind that it's logarithmic. And so what you can see is that the sort of greenish dots are the Fable 5.1, and so if you just take, at max, the dots, I should say, are various thinking levels, so if you just look at the max thinking level on 5, and go, right, well, what is it now, max, on 5.1?

Charlie Cowan: Well, it's gone from 70% up to, whatever that is, 73% on the coding, and it has gone left from just shy of $20 to just shy of $10. So nearly half the cost at an improved score. And so we're generally going to see more and more of this. In the blog post, you can see I put a bullet point there. Typical workloads are costing about 25% less, or 45% if it's agentic work, and this is due to the way that it's handling a lot of the caching. You and your teams should not be using Fable left, right, and center. This is for your most challenging work, this is for planning work, this is for strategy work, it's for your most complex coding work. Day-to-day, you should be using Sonnet or Opus, but this is a really good advancement for those more challenging problems.

Charlie Cowan: Next up, and in the same vein, is a new model which has yet to be released, called Astra from OpenAI. And so, if Mythos is the frontier sort of cyber model for Anthropic, then Astra is the frontier cyber model for OpenAI. And, after the hoo-ha, I should say, of the rollout of Mythos and it being sort of taken off the market and pulled back again, OpenAI want to take a more considered approach, for their own branding and marketing, and so they're doing a lot of announcements and sort of gradually trailing the release of Astra. And so, yesterday, I think this was, they put out a blog post about some of that sort of ongoing work. And this, I thought, was a really interesting chart.

Charlie Cowan: As all of the model providers do, they try and test their upcoming models to break things in terms of the cybersecurity sort of framework and threat that they're working in. And what you can see here is the percentage success rate that the Astra model has against this test framework called Exploitbench versus, versus GPT, what am I looking at, 5.6 Sol, which is the current sort of frontier model from OpenAI. And the point here is that it's just going exponential. The scoring rates are going up and up and up, and then there are different charts presented in that blog post that show it getting up towards 100% on various measures. The point here is that these models, these frontier ones, are very, very dangerous if they get into the wrong hands. And these model providers are balancing defenders versus, well, offenders, if these models get into the wrong hands.

Charlie Cowan: And so they want to make sure that defenders, being you, us, anyone that is running an organization, has got the right tools and protections in place to defend against a bad actor who might have an open source, open weights model that is of a similar standard moving forwards. We're just trying to stay ahead. Next up, back to Anthropic for this one, and this was the announcement from Anthropic of a research preview of a thing called the Model Hardware Standard, which defines how agents, AI models, agents, can operate physical hardware and devices, primarily in clinical and manufacturing scenarios. So you could imagine, in a biochemical lab, where you've got the sort of robots moving around, doing testing, running experiments, and so on. And what this defines is that standard.

Charlie Cowan: So, you can see the video that they presented when you go to Anthropic's website, but here you've got an agent over here, here you've got the model hardware standard, and then here are all the different tools and hardware items that the model might be controlling. Now, this is really interesting, because it opens up, you know, what's beyond that. We're thinking about more robotics, thinking about more about cars and physical items that might be in the home or in business. So definitely one to think about. I would imagine that as the pace of these models accelerates so fast, that hardware and robotics is going to come into normal, you know, daily work across so many different businesses very, very quickly. And the final AI in the news for this week is about data residency.

Charlie Cowan: Now, depending on what industry you are in, you may have requirements in the contracts that you sign with clients to only host or handle their data within certain geographic restrictions. I'm based here in the UK. We are outside of the European Union, so companies may have restrictions to only handle data in the UK. In the EU, it may be in the EU, and various other countries around the world. Now, up until now, OpenAI has definitely been leading on this, compared to OpenAI versus Anthropic. OpenAI on Enterprise has had the ability to set data residency, not just on where the data is stored, but also on inference, so where your AI prompts are handled. And you've been able to pick from, I don't know the number, I'd say, let's say, 12 or 20. It was a large number of different residences that you could choose. But that was only for enterprise clients.

Charlie Cowan: And today, they've just announced, or yesterday, they've just announced that this is for GPT Business workspaces as well. This used to be called ChatGPT Team, now ChatGPT Business. And you can now choose the EU and US, and I'm sure many more will come in the near future. This is definitely a decision point. It has come up in client conversations that we have had, where the client's not been able to choose Anthropic as of yet, because inference and data residency is not offered directly, natively, through the Anthropic platform, outside of the US. So, if you're in a regulated industry, that might be something to keep an eye on. Now with that, I've not seen any questions that have come through, so I am able to take questions. If not, if you're watching on YouTube, then feel free to ask any questions through the comments. What that brings me towards is next week.

Charlie Cowan: And on next week's webinar, I am absolutely thrilled. We're going to be inviting Jamie Chung from 9fin to the webinar. Jamie is the Chief of Staff to the COO at 9fin, a global provider of financial data into the private and public credit markets. And Jamie's gonna talk us through how they ran their AI week, which happened the end of June. Now this is gonna happen one hour earlier than our normal webinars, which are 3pm UK time. This is gonna be at 2pm. But you can register for this using the QR code, or go to kowalah.com, go to our resources, and you'll find webinars. And we'll spend an hour with Jamie walking through how Jamie planned the week. What happened during that week, some lessons learned, and some advice to share to you if you're planning your own AI enablement week. And so with that, thank you very much.

Charlie Cowan: Thanks for joining, whether you were live or whether you're on YouTube, and look forward to seeing you next week.