---
title: "Transcript: Show Back or Charge Back? Deciding Where Your Next AI Pound Goes"
webinar: "Show Back or Charge Back? Deciding Where Your Next AI Pound Goes"
date: 2026-10-14T14:00:00.000Z
duration: "1 hour"
speakers: ["Charlie Cowan"]
topics: ["What Smart Reports actually shows you: usage, cost, friction, and skill patterns worth packaging", "Showback vs chargeback: two ways to make AI spend visible on a P&L", "Finding the middle 60%: the users who aren't power users and aren't idle either", "Turning a friction pattern into a shared skill instead of a support ticket", "Deciding what gets budget next quarter, not just measuring what you already spent"]
recording: https://youtu.be/Rpzypx6ISDs
canonical: https://kowalah.com/resources/wednesday-webinars/show-back-or-charge-back-ai-spend/transcript
source: Kowalah Wednesday Webinars
---

# Transcript: Show Back or Charge Back? Deciding Where Your Next AI Pound Goes

**Recorded** 14 October 2026  
**Duration** 1 hour  
**Speakers** Charlie Cowan

## About this session

Anthropic just shipped a live dashboard on team AI usage, cost, and friction. See it in action, and learn how to decide whether to showback, chargeback, or upgrade your middle 60% of users before you buy another seat.

## Topics covered

- What Smart Reports actually shows you: usage, cost, friction, and skill patterns worth packaging
- Showback vs chargeback: two ways to make AI spend visible on a P&L
- Finding the middle 60%: the users who aren't power users and aren't idle either
- Turning a friction pattern into a shared skill instead of a support ticket
- Deciding what gets budget next quarter, not just measuring what you already spent

## Recording

https://youtu.be/Rpzypx6ISDs

## Transcript

**Charlie Cowan:** …and talk about what is going on in the world of AI. There's so much going on at the moment that it can be really difficult to keep on top of what is happening. This gives you a bit of time to keep on top, to digest, and to think not just about all the new releases that are coming through, but about how we can implement this in our teams and our processes, and make good use of it.

We're going to follow our normal process, which I'll introduce in a moment. If you're watching online, you've got a chat window. You're in a Zoom meeting, so if you look at the top or bottom of your device, depending on what you're on, there's a chat window where you can ask questions. I'm also opening up a Q&A window. You can use either of them, and I'll keep an eye on them on a different screen. If you're watching on YouTube, as I know many of you do, and you have questions as we go through, ask them in the YouTube comments and we'll come straight back to you.

Today we're going through an important topic, especially for our enterprise clients, who are trying to figure out how to budget for, account for and pay for these AI expenses that are coming through. We're going to talk through showback or chargeback, which, if you're not familiar with those terms, I'll introduce shortly. We'll spend about 40 minutes on this main topic, and introduce some ways in the Anthropic platform that you can use to manage this yourselves.

Then we'll talk about AI in the News. This is our recurring segment, where we look back at everything that has been launched in the last seven days, and this week has been a big one. There's some really cool new stuff from Meta and some really cool new stuff from Anthropic, so there's plenty for me to show you. Then, as I mentioned, we've got time for Q&A. It's great that we can share our experiences and what we're seeing in our clients, but as we progress week by week, we also want this to be an opportunity for you to drop in and ask about what you're trying to build and your own experiences. Feel free to ask your questions there, or at any point as we go through. Don't feel you need to wait until the end.

So, showback and chargeback. How do you account for and cross-charge your AI expenses? Where is your AI expenditure going? For those of you in the UK, we're thinking about our AI pounds, in Europe your AI euros, in the US your AI dollars. Where is your money going?

A little research that came out fairly recently, the State of FinOps. If you haven't got a FinOps team, FinOps stands for financial operations. It's a team focused on managing the finances relating, initially, to cloud expenses. FinOps was built off the back of the rise of things like AWS, Azure and Google Cloud Platform, GCP, and FinOps teams are now being pointed at the rising expenses coming from AI expenditure and tokens. 98% of FinOps teams now manage that AI spend. The bill comes in from Anthropic or OpenAI, and those FinOps teams are being aligned to it. But interestingly, only 8% of those FinOps teams report to the CFO, and it's the CFO who is signing off this massive new line item in the P&L. So how do we square those two things, if the team looking at managing this isn't linked to how those expenses are being signed off? We're going to look at some of the data that can come into these FinOps teams, so they can start to assign where that expenditure is going.

About a month ago, Anthropic launched a new feature in their enterprise platform. If you're working within a company, you may be aware that you've got Claude Team, which is for teams of up to 150 seats, and beyond that you go to Enterprise. One of the main differences people always think about between Team and Enterprise is this. On a Team plan, you pay for the number of users, and each of those users has a certain amount of tokens they can use before they hit their limit. Once you move on to Enterprise, you have a reduced per-seat fee, but the bulk of what you're spending goes on the per-token cost, and you're paying API pricing. This is what concerns companies as they go through that 150-user limit: they start seeing costs potentially increase, depending on how they've set things up.

You get a lot of extra benefits when you go up to Enterprise. It's not all about paying more. You get much greater governance, compliance and reporting functionality, and one of the features just launched as a beta on Enterprise is Smart Reports. So what is a Smart Report? It looks at a subset of your team's chats, conversations and sessions over the preceding period, and provides some analysis of what people were talking about. On the screen on the right, we can see what Claude was used for: different types of work stream, based on the number of sessions. Migrating the shipment tracking service off the legacy scheduler. Automating weekly carrier performance scorecards. Triaging and fixing warehouse scanner sync incidents. This is an example report, so your company and your industry will show different things. But you can imagine, across your 5,000 or 10,000 people, getting insight into what people are talking about and where the cost is going. This shows the spend and the number of sessions: sessions is the percentage in blue, and spend is in red. It's a great little insight for starting to look at what is going on.

On Enterprise, you can also split up your teams by what we call workspaces or groups, so you can look at what marketing is doing, or what a specific region is doing, and look at those chats in that context as well.

You can then look at the different things being created by those people in their chats. Here, while a very small percentage, 9% of the sessions, relate to code, the cost relating to code in this example report is 40% of the expenditure. So very quickly, as a FinOps team or a CFO, you can say, let me focus my attention here. We might have 10,000 people in our organization, but it's our engineers using Claude Code who are driving 40% of our expense. For written content, people writing blog posts or emails or reports, there's very little. For research, deep research, it's maybe a quarter of the sessions and 20% of the expenditure. So you can start to focus your effort on where to go if you want to minimize this and, as we'll talk about later, on cross-charging.

This feeds off the previous slide: a large amount of the expense is coming from very few users, very few teams, maybe even one team, and they're racking up half or so of the bill. So you can dive in not just on the teams but on the individual users running these long-running sessions. And I would say, not just because it's in this example report but because I'm seeing it so much across my X feed, who is running up a massive bill? It's engineers, it's developers, and it's for a few reasons. One, they're typically working within Claude Code. That's their weapon of choice, and they're in Claude the whole time. Two, they're some of the most pro users. They've learned how to build loops, how to build goals, how to build long-running sessions and how to use sub-agents. As they head off to achieve a specific goal or task, they're using all the power that Anthropic and Claude provide, so it's no surprise those sessions are expensive. Compare that with someone in finance or legal who might be doing a bit of QA. Even if they're building some scheduled tasks, they tend to be quite limited. They're certainly not defining sub-agents and long-running goals that run overnight.

You can then drill into the individual sessions and see what was being used. Well, Claude Code, Claude Code, Claude Code. What model is being used? Opus 5, and we've got some Fable here. You can drill in to exactly what was happening in those sessions, and then maybe it's a training issue, maybe it's a skill issue, maybe it's a permissions issue, and we want to limit what people can do within a single session or which models they have access to. Does everyone need access to Fable? If we're honest, no. So you've got a lot of control here to start to manage and learn what's going on.

The topic of this conversation is showback and chargeback, so we're going to move on to that now. With Smart Reports, is that enough to do your billing and your cross-charging? Not really, for a few reasons. It's only reading a sample of your sessions, then pulling back so that you can use a bit of analysis. It's only looking at a subsection of your data, so if you want to look back over the last year, it won't give you that level of detail. And in various scenarios, depending on your industry and the model you're using, if there's zero data retention, you won't be able to access all of that data. But what it does give you is a directional view. Looking across our 10,000 people, I can see that the bulk of our expenditure is coming from this team, from this region. I can see that the bulk of the long-running sessions are using this model, or this part of the platform, which is likely to be Claude Code. And it lets you think about what you want to do next.

You do have access to the data you need to cross-charge, if you choose to do that. Through the analytics APIs, you can pull out all of this detailed information. Because you're on the Enterprise plan, you can pull it back by the different workspaces, and if you've set up groups or custom roles, you can analyze it on that basis. Depending on how you've set up your workspace, you can then link that into the cost centers you've got in NetSuite, Workday, or whatever you use to manage your internal cross-charging. So you can weigh up the two: Smart Reports for getting a sense of direction, and the detail for actually billing and cross-charging.

Before I go into showback and chargeback, I want to point you to some great documentation. We'll put these in the show notes. For getting started with Smart Reports, we've got this lovely article, with all of the screenshots I just showed you. You can go through how to set it up, what you need and who has access to it. That's handy. Secondly, we haven't talked about this, but I wanted to show you how to use analytics chat to ask Claude about your usage. As long as you've got the right permissions and you're on the Enterprise plan, you can go into your Claude account, select Analytics and start chatting. Here are some example questions: show me what's happening in my account, show me the daily spend, show me this versus various teams. Imagine that as a way of building up your conversational knowledge about what is going on. And, a short advert for what I'm going to show you in a moment, you can now build super cool dashboards directly within Claude. If you're an admin or an owner and you've got access to this kind of analytics, depending on your permissions, you'll be able to see this right in your Claude platform.

That's Smart Reports and analytics chat. Let's get back into the deck.

I've used the phrases showback and chargeback. Showback means we give P&L owners visibility of what their teams are using, but we don't charge them. We let the chief marketing officer, the chief revenue officer, the head of service, the head of Germany and the head of Asia Pac know what their teams are spending on AI. It's not to penalize them, it's not to charge them, and it's not to take it out of their P&L. It's there to help them understand what's happening. Maybe there are some… I hesitate to say leaderboards, because we'll come on to this in a minute and we don't necessarily want to penalize people. But we do want to encourage people who aren't using it, and alert people right at the top that there may be different ways to approach it.

This is a really good first step. Let's get some trust in the numbers, build the right dashboards, and make sure the right people are seeing them. Maybe we talk about it in our leadership meetings or our board meeting, and maybe there are some incentives for teams to do certain things to encourage usage. But it's not hitting the P&L.

The next step is to move on to chargeback, where we say, hang on, why should IT, because this is where it normally starts, be sitting on this ever-increasing expense coming through from all of the AI usage and tokens? Why should that sit centrally, at company level? Maybe it should start to be apportioned out to the different business units.

You want to make sure you've got accurate data, which is why showback builds people's confidence and trust in it. Now you're using the analytics API to pull the actual per-user data that is linked to your cost centers. But be careful. Say you've got a P&L owner, let's say the VP of services for Europe, and they now own the token cost in their division. If you haven't planned this correctly, they might immediately start thinking about how to reduce it. They might start taking licenses away from people, limiting usage, or hiring someone to do the work instead of getting the AI to do it. That's a blunt instrument for what should be happening, which is making sure we're working on high-value use cases. How do we track the ROI of the use cases we're using? How do we train people to use AI in the right way? How do we put the right limits and controls in place so people can only use the right models for the right jobs? How do we transfer workloads from humans using AI to agents using AI? These are all great things, and when we move from showback to chargeback we need to have them in place, so that we don't see knee-jerk reactions from the people who own the P&L.

So our general advice is always show back. As soon as you're on Enterprise, you've got access to this data, so show it to people, create dashboards, encourage people, encourage experimentation, and move people along this journey. Then start to look at chargeback once you've got a measure on consumption and some trust in it, and you can give people levers that improve their usage and the ROI on it. If someone is spending more, we can give them an award and a pat on the back. It's not a negative. By using AI, you're driving our company forwards, and we can prove that.

Depending on how your organization is structured and how you do cross-charging today, you may look at this differently, but the way we think about it is threefold. Firstly, the seats. Depending on your deal with your provider, whether that's OpenAI, Gemini from Google, Copilot, Anthropic, whatever, your license may be slightly different. But typically you'll have some element of a fixed per-seat fee per user, which may be $20, $30, something like that. This is not linked to usage, so if you cross-charge it you're really just moving money around. You may choose to do that, or you may choose to keep the seats handled centrally.

You may then also choose to say that for each person we'll give them, say, $50 of usage per month, and that is also held centrally. Everyone gets the consumption and the seat, because as a company we want to encourage everyone to have access, so we cover that cost centrally. And then you might say that everything above that, we now cross-charge back to the business unit in question, because we want the business unit to take over some management, some control and some levers to encourage people to get the best ROI on that expense. As an example, for finance you might have 120 people, and centrally we hold a £20-a-month allowance for each of them, and the rest we push out to the business unit to bill to their own P&L.

I think we have covered most of that. I talked a little about not seeing this as a leaderboard. What are we encouraging, more usage or less usage? So it's not really a leaderboard, it's more of a bell curve, and this is the typical usage that we see in any company.

Down at the bottom, or on the left, you've got non-users, or ambivalent users. I've got an account, maybe I'll go there to ask a question now and again, but I'm not that into AI. I like working in my spreadsheets or my email myself. I'm ready for my pension, or I don't trust AI because of the environment, whatever the reasons are. They're just not engaging.

Then up at the top end, you've got the power users. These are going to be your engineers. It might be someone in your analytics team who has become a power user of Claude. Maybe it's someone in marketing. The team isn't the key thing. It's that these people are all in, and they're setting up scheduled tasks and running them. These are going to be your high-expense users. Where you're looking in your Smart Report at long-running, expensive sessions, they're going to be up there.

And then you've got the bulk in the middle, the 60% or so who are into AI and using it, but who aren't embedding it into repeatable skills. They're not building scheduled tasks, and they're not using agents at this point. This is where we see so much of the potential uplift.

Our general advice is to focus on the people in the middle. How can we get them to have more usage, and more outcome and activity through using Claude, before you start buying more people's seats or doing anything at the fringes? You're going to get more ROI by focusing on these people and getting them to use the tools in a better way.

One note on the people on the right-hand side. Someone spending the equivalent of their own salary or more on tokens isn't doing anything wrong. We do not want to penalize these people or necessarily put limits on them. But we do want to go and understand what they're doing, how, and what problems they're trying to solve, because it may well be that there are basic operational or implementation skills we can give them: using the right model for the right task, defining a loop or a goal correctly, setting up sub-agents. There are many things you can do to dramatically bring those costs down while still having them as your power users and very, very effective.

Here's another nice thing that comes out of Smart Reports from Anthropic. It looks at some of the sample sessions, finds things that are quite repetitive and could be useful across multiple users, and suggests shared skills. For those of you not familiar with a skill, a skill is a way of defining how something gets done in your Claude. The next time you say, create me a sales proposal, or create me a customer agenda, whatever it might be, you don't have to tell Claude how to do it because it's already baked into a skill.

Why is this important when we think about showback and chargeback? Because a lot of these long-running sessions are people figuring out how to do something and iterating on it. Oh no, that's not what I wanted. Oh no, can you use our deck template? Oh no, I've forgotten we should do this. If you can spot these things and turn them into an organizational skill, and give it to everyone in that team or company, you head off all of that iteration and refinement before it has even started. It's a really easy way of improving the experience for that middle 60%, and of reducing the session costs of the work they're doing at the same time.

So we covered quite a lot there. I want to give you some things that can help you bring this to life for yourself. I've shown you the Anthropic articles, and we've also built a template you can use. It's a Google spreadsheet, and it gives a lot of the instructions and guides we've gone through here. It lets you populate some settings about what you're paying for your seats, and some of your inputs: the different cost centers you've got, headcount, who's in each team, what you're paying, and so on. Then you can go through both the Showback tab and a Chargeback tab and start to populate those yourself.

Now, of course, you'll say, Charlie, are you expecting me to build out this spreadsheet myself? Absolutely not. We'll touch on this a little in AI in the News, but if I go to my extensions here, and oh, I've added it in here, Claude. You can now add Claude into Google Sheets and Google Docs and, I should say, Google Slides, so you can have a conversation with your Claude there. It has access to your main Claude, so it has access to your connectors and your other systems, and you could populate this just by having a chat in the Claude sidebar. You might even build a skill, I'm going to show you that in a minute, that helps you with this.

So how do you get that template? On kowalah.com/resources/templates, we have an AI Spend Showback Chargeback Model. Click that, copy the Google Sheet, and it gives you your own copy of that Google Sheet. It explains how the model works. Just give the template to your Claude and away you go. You can literally open this in Claude, or in ChatGPT. It's very cool. There we go: can you read the template, and then help me to do that?

So that is showback and chargeback. If I now go back to our slide deck, where are we? Slide deck, for what we're going to carry on talking about as we go into AI in the News.

A lot has been going on this week. I've got a few slides I'm going to walk you through, but I'm going to start with one that is right at the end, because it has been one of the most impactful things for me this week. As you may see, I'm presenting this slide deck in Claude. I'm doing all of our presenting in Claude right now. Every workshop the team runs, every internal team meeting, and the Wednesday webinar is all being run directly from Claude. I have not created a Google Slides deck since Claude Slides moved into the main chat.

So let me go to a new chat. Here we go. I'm in my Claude here, and if I go to my artifacts here… Artifacts. We already had Docs, Slides and Design, and we've now got two new ones, Dashboard and Motion, and I want to show you both of those.

Dashboard is, as it sounds, an interactive dashboard of the kind you might previously have built in Tableau or Power BI. You would have had to go to your dashboard administrator and ask, can you build me this dashboard? And they'd say, oh, we've already got one that looks a little bit like that, can't you use that? It's very complicated, and a large number of dashboards get built and only ever used once. Claude Dashboards aims to resolve that by making it super easy to build dashboards around anything you already have a connector for.

When I think about Kowalah world, we've got access to the Kowalah platform, so if you're a Kowalah client, you might have access to the Kowalah platform and can build dashboards about your AI roadmap, your opportunities, your projects and so on. You might have access to your email and your calendar, so you can build dashboards about what's going on this week. You might have access to your CRM system. We use HubSpot, so you can pull that data in. You might have access to your call transcription tool. We use Granola. Where these dashboards are really great is in bringing all of those together into one, so you could have a client dashboard where you flick through different clients or deals and pull in information from each of these.

So what does a dashboard look like? I've created one, though I can't show you one based on all of our real data because it would show client information. What I did instead was upload a spreadsheet of all of the aircraft traffic movements in the UK during the pandemic and ask it to build me a dashboard. This was one hit. I haven't done anything other than build the dashboard like this. I gave it the spreadsheet, and it created this dashboard. UK aircraft traffic movements, okay, here are the different movements, lovely. It's created some nice bar charts here, split the data out by airport group, and made some nice interactive elements. I can even search, so if I search for Gatwick, I can filter down. Imagine you have a CRM system. You can get interactive, both to read the data and, depending on the permissions in your connector, to write as well, so you can update things in your CRM system directly in here.

Let me go to the chat for this. This was one hit. I said, hey, can you see this spreadsheet? I gave it access to a spreadsheet, and if so, could you create a dashboard artifact off the data? And there we go. It said, yes, I could read it, here's the dashboard. One hit. So now I'm going to say, okay, that's brilliant. Can you use the Kowalah design system to make this look on-brand?

This will go and rebuild it and make it look really nice. Having built one of these dashboards, plugged into your CRM, your email, or whatever it might be, you can then share your artifact, your dashboard, with others in the team. You can share it with an individual. If you're on Enterprise, you can probably share it with a group, or with everyone in the company, or potentially, if you've got the permission, with someone outside your organization as well. These are going to become really important and powerful for sharing information internally.

I'll let that carry on building while it makes those changes. The second artifact type, launched last week, is Motion. You can think of it as a video, like a stop-motion video, a launch video or an explainer video that shows people how things work. I'm going to give you an example of something we all know about: the story of Goldilocks and the Three Bears. All I asked Claude to do was create a one-minute video of Goldilocks and the Three Bears, and this is exactly what it created. It asked me a few questions. It asked whether I wanted a felt style, which I selected. It gave me three different examples that I could choose from, and I chose this one, and it created this one-minute video. If I were sharing my sound, you'd hear some lovely music that goes along with it. Then I can tweak it. Oh no, I actually want this in Kowalah colors, in these colors. You can really play around with it: different tweaks, the number of beds, how long I want the video to go, how fast, and everything. Anything where you're thinking of explaining something or changing something for customers, you can start creating these.

Okay, Charlie, we're not doing the three bears, so what's a real-life example? Well, we are just launching a series of live training sessions in London. Individuals can come to these sessions, learn all about Claude, get hands-on, and collaborate with people from other organizations. I wanted a little launch video to put on LinkedIn, so I created this earlier today, and this is our video. I fed it all of the instructions about how these sessions are going to run, what we want people to come away with, how they're going to collaborate on a table of groups of four or five on a live situation, and that they're going to make connections with people who are in the same boat. And this is where they're going to happen, when, and how much.

That was a five-minute conversation I had with Claude, and I'm now able to post that video online, and it's all on brand with our marketing design system, so I'm very happy with that. If you think about how long it would have taken to create this type of content before, it would have taken a long time. It would have gone into a backlog for someone in marketing or a content team. Maybe it would have gone to an external agency, and maybe it would have cost thousands of pounds to get someone to design it. And you're only going to get one version of it. Whereas here, I can create 25 different versions of this video that we can use and test out.

So definitely play around with that. All you need to do in your Claude is go to Artifacts and click Dashboard or Motion, and it will open that up. If you're in a new chat, you can choose Dashboard, or type /motion, and it will bring that in. Have fun with that, and I look forward to seeing what you come up with. I love this so much. This was our dashboard, if you remember, for air transport movements. I said I wanted it Kowalah-branded, and so here we go. Here is our Kowalah-branded dashboard for aircraft traffic movements in the UK during the pandemic. Really, really nice, and exactly on brand. You can imagine the dashboards you could create for your team, for senior leadership, for partners, however you want, depending on the data access, and make them really, really interactive and visual. No Power BI skills required.

So what else have we got in AI in the News? Last week, Haiku 5.5 was launched by Anthropic. If you're in Anthropic land, you'll know they have four model families. Haiku is the smallest, then Sonnet, then Opus, and then Fable. Over the last few weeks they've been upgrading the whole family, so some are on 6, but many of them are on this 5.5.

Benchmarks are benchmarks, and charts are charts. All you need to know is that this chart shows how expensive a specific task is, on a logarithmic scale, so it steps up from cents to 50 cents, $1 and so on [exact scale values unclear in the recording]. Up the left is how good the model is at a particular task, so the higher, the better. With all of these charts, you generally want to go up and to the left.

What you've got in blue here is Sonnet, and many of you may have been using Sonnet 5.5 as your daily driver for lower-level tasks. You can see that you can get pretty much Sonnet-level performance on many of the thinking levels of Haiku. But the real comparison is Haiku 5.5 against Haiku 4.5. Haiku 4.5 was way more expensive, at pretty much $1.50 per task on average and a 15% success rate, whereas now, at high, it's 18 cents compared to $1.45, and 61% success. You don't need to be a data scientist to know this is a massive increase. Where you're thinking about low-level tasks, maybe low-level tasks that your agents are doing, Haiku is a great model. We're using it in a load of our agents that go off and do research and pull in information from other systems. We don't need PhD-level thinking for that. Haiku is absolutely perfect.

Next up, and I want to say this came out maybe two nights ago: Meta. Many of you know them as Facebook, but Meta, the parent group, has been really doubling down on AI in the last few weeks. They've had some great success launching Muse as their personal agent, and now a series of new models has come out, primarily Wonder, so Muse Wonder, and Muse Wonderful [the model is called Muse Wonder Full], and I think we've got to give credit where credit's due: Wonderful as a name for a model is fantastic. Benchmarks are benchmarks, but these are all moving up to frontier-level capabilities across the board. Many might have counted Meta out as a frontier-level contender even six months ago, but with these models, and Muse as this personal agent, they are right up there at the frontier, so definitely one to think about when you're considering your model mix.

I think that is most of what I wanted to cover today, without going into too much detail. Let me go back to our deck here. We talked about Claude Haiku. Happy days. We talked about Muse Wonder being launched. And we talked about Claude Docs, Slides and Design coming out of beta, and Dashboards and Motion now being in beta themselves. I haven't seen any other questions come through on the chat. If you're watching on YouTube, feel free to ask any of your questions about anything we've covered, and we'll come back to you very quickly.

Next week, we're going to be talking about your agent org chart. We talk about our agents a lot. We use our agents in every bit of the work we're doing, and we give them all names: Brian, Karolina, John, Krista. The question comes, when do you choose one agent versus splitting a role into multiple agents? We'll talk about some of the decisions we make, and what makes giving a role to two agents, rather than one, a good experience. Same time next week, 3pm if you're in the UK, 10am Eastern, and you can register just by going to kowalah.com/resources. I will see you then. Thank you very much.

---

This transcript is published by Kowalah, a UK Anthropic Implementation Specialist.
Webinar page: https://kowalah.com/resources/wednesday-webinars/show-back-or-charge-back-ai-spend
All transcripts: https://kowalah.com/resources/wednesday-webinars/transcripts