--- title: "Transcript: Claude Tag: Give Every Slack Channel Its Own AI Teammate" webinar: "Claude Tag: Give Every Slack Channel Its Own AI Teammate" date: 2026-08-26T15:00:00Z duration: "1 hour" speakers: ["Charlie Cowan"] topics: ["Tag @Claude into a live Slack thread and watch it break down a real task", "Turn on ambient mode so Claude follows up on stalled threads unprompted", "Scope tools, data, and memory per channel so teams don't leak each other's context", "Compare Claude Tag to Cowork and Claude Code — which one for which job", "Pick your team's first channel to pilot it in"] recording: https://youtu.be/m9iOnHf-738 canonical: https://kowalah.com/resources/wednesday-webinars/claude-tag-ai-teammate-in-slack/transcript source: Kowalah Wednesday Webinars --- # Transcript: Claude Tag: Give Every Slack Channel Its Own AI Teammate **Recorded** 26 August 2026 **Duration** 1 hour **Speakers** Charlie Cowan ## About this session Anthropic just replaced its Slack bot with Claude Tag — a shared AI teammate every person in a channel can tag, steer, and hand work to. See how it works, what to automate first, and how to scope its access before you roll it out. ## Topics covered - Tag @Claude into a live Slack thread and watch it break down a real task - Turn on ambient mode so Claude follows up on stalled threads unprompted - Scope tools, data, and memory per channel so teams don't leak each other's context - Compare Claude Tag to Cowork and Claude Code — which one for which job - Pick your team's first channel to pilot it in ## Recording https://youtu.be/m9iOnHf-738 ## Transcript **Charlie Cowan:** All right, let me share my screen, which is... I'm gonna share this window... down here... and... alright, let me just open up the chat and the Q&A... There we go. Brilliant! Well, good afternoon, good morning, depending on where you're dialling in from. Welcome to this week's edition of the Kowalah Wednesday webinar. If this is your first Wednesday webinar, we get together at the same time every Wednesday and spend an hour talking about things that are top of mind in the world of AI, and specifically business and AI. So these are often things either that we are using or discovering internally. They might be things or ways of working that we're seeing in our client projects. New releases, new models, and so on. **Charlie Cowan:** It's such a fast-moving space, and it gives us a time to reflect and then take what we're learning and share it with you, so hopefully it becomes something useful that you can take into your own organization. Now, today, we're going to be talking about a really cool topic called Claude Tag, which I will come to in a moment. But let me walk you through how we'll handle today's session. We'll spend probably about 30 minutes or so talking about Claude Tag, the main topic of the week. Then, we will go into our usual, regular AI in the News segment, where we look at the last 7 days of releases, updates, and talk through what we think is most important and impactful for you in your business. And then we've got some time for some Q&A at the end. **Charlie Cowan:** What I would say, though, is that don't hold any questions until the end if they relate to what we're going to talk about, about Claude Tag or the AI in the news. We're in a Zoom event, so depending on what device you're on, maybe at the bottom of the screen, the top of the screen, if you hover, you'll see there's both a chat window where you can chat, and ask a question there. There's also a Q&A button as well, and that's for, you know, a more formal question. It'll come up on my screen, and I'll hopefully do a good job of noticing it, and I'll come back to you. As I say, you know, don't hold on to any questions for the end. Feel free to ask them as you go through. And so, let's dive into our discussion about Claude Tag. **Charlie Cowan:** And let's start by framing one of the challenges that we're seeing in companies, and especially in large enterprises that are moving on to the enterprise plans of AI models, including Claude. The default is, right, we're gonna give every single person access to a Claude account. So that could be Claude Desktop, they might be accessing Claude on the web, it might be Claude on their mobile, but ultimately, the individual has got access to a Claude account. And they have the power, the right, to pick and choose any model. Now, if you're a typical human, you're gonna say, I want the best, I want the best. And for anyone that's not close to this world of AI and tech, how are you supposed to know the difference between Haiku, Sonnet, Opus, and Fable? **Charlie Cowan:** All you know is that one is bigger and better than the other, and therefore, I'll probably use the bigger and better one, even if I'm just summarizing my one-to-one with my manager. Now, if you're on team plan as a company, then everyone is limited by the number of tokens that they can consume, either on a 5-hour limit, and then a weekly limit. So, ultimately, it doesn't matter what model someone chooses, they run out, they run out, until they get reloaded again. But on Enterprise, that's not the case. At Enterprise, it's pay-as-you-go on all of the token usage. And so, whilst you can put limits on individual users and workspaces and groups and so on, the reality is that we're seeing clients that are experiencing bills that they hadn't planned for. And part of the reason for this is not just, oh, can't you just limit, limit, limit? **Charlie Cowan:** Well, you're still giving many people the choice over what models they can use in which circumstances. And often, the human at the other end isn't the right person to be deciding and making that decision. They don't know. So that's one of the problems. The second problem is, where is work happening in your organization? Now, if you're an organization that uses Slack, then Slack is very much where that work is happening. If you're in Teams, it's likely in Teams. If you are one of the companies that uses Google Chat, then maybe that's where the work is happening. But that is where the flow of work is happening. It's where groups are coming together to discuss stuff. And if you're then pulling people out of that, say, oh, now I'm going to ask Claude about this thing, and then report back, you're taking the person out of the flow of work. **Charlie Cowan:** So that's another one of these problems, and so what we're going to talk about today is about how do we limit that exposure around budget, governance, and so on. And also, how do we make sure that Claude is where the work is actually happening? And this is what Claude Tag is. Claude Tag was launched, I want to say, maybe, the time moves so fast, but maybe it was 8 weeks or so ago, fairly recently. It is only in Slack at the moment. So what I'm going to show you and walk you through at the moment, you need to be a Slack user as a company. I would imagine, from the way that they phrased it, you know, at this time, it's just in Slack, that we would expect to see something similar arrive in Teams, Microsoft customers, and maybe in Google Chat for those that are using Google Chat as well. **Charlie Cowan:** But I think you'll get the idea, and then when this is rolled out to other platforms, the same thing will exist. So a few things to keep in mind before we get into this. I talked about this already, meeting people where the work happens. So, across clients, across ourselves here at Kowalah, work happens in spaces, channels, where people are collaborating. So, you might have marketing channels, you might have sales channels. Your salespeople and deal teams very likely have deal channels for individual deals that you're working on. You're gonna have product channels, you're gonna have developer channels. This is where people come together and they get work done, and the idea here is about bringing Claude into those channels, rather than having those people having to jump out and go and have a conversation elsewhere. **Charlie Cowan:** The second concept to think about here is that when you're having a chat with Claude in Claude Desktop, in Claude Web, it's really your identity. So when you say, go and check Google Drive, when you say, go and check Notion, when you say, go and check Linear, Claude's running off to go and do those things, but it's using your identity, so it's seeing what you see. Now there's some pros and cons to that. So the pro is, well, everything's, like, tightly scoped to what you can see, and so that's good from a governance perspective. But it also means that unless you're the one that is, you know, asking, unless your laptop is open, then Claude can't see these things, so you're very restricted to the human doing the work. **Charlie Cowan:** With Claude Tag, Claude actually gets its own identity, and I'll talk to you about this in a moment, but you'll likely give Claude its own Gmail account or Microsoft email account. You'll give Claude its own access to HubSpot, its own access to Salesforce, its own account in Linear or GitHub, and that's where it will be able to access those things itself, as itself. The third thing that we shall have a look at is that you are able, as the administrator of Claude Tag, to be able to choose the model, not just as the overall, but on a channel-by-channel basis. You know, if you've got salespeople that are working on individual deals, do they need Fable? Do they need Opus to be able to do this? Or actually, is Sonnet a very capable model, and you can manage that directly? And that feeds into that budget, and the budget per channel. **Charlie Cowan:** I want to say that in our marketing channel, for example, you know, there is a budget of $200, $1,000 per month, whatever that might be. And it means that you can really limit your exposure on a channel-by-channel basis, and that gives you some control that you wouldn't have otherwise. So, let's have a look at what this actually looks like. Now, if you're familiar in Slack, then you'll know your way around Slack. There's two ways that you can interact with Claude Tag once you've set it up. It's gonna, just to be clear, it's gonna show up as Claude. There was a previous Claude integration, and when you turn on and enable Claude Tag, it'll walk you through switching from the legacy one into Claude Tag, but it's called Claude. **Charlie Cowan:** And, as a Slack user, you can just directly chat to Claude in a direct message with Claude, and you can set up channels, where you can at-mention Claude. So, in this example, and what I've got here, let me just see, actually, if I need to switch what I am sharing, which I think I am, let me change what I'm sharing, and see if I can go to Slack. Where is my Slack? Too many windows... too many windows... What am I trying to share? Let me go here, screens... Can I find Slack on my... Where to share... Hold on one second... Let me open Slack. I'll give it one more look. Otherwise, I will... no idea, no idea why I can't share Slack in, let me stop my share. And then I'll just try it once more. Slack, here we go. Okay, so here is my Slack. **Charlie Cowan:** And, now just for our demo purposes, we've got a channel here called Meet Claude Tag. And Claude is in this channel, and just beforehand, I said, hey Claude, I'm doing a little demo on Claude. Can you let me know what systems you've got access to, and how I can best use you? Happy to be the demo. What I've got here, Slack, yeah, you watch this channel. GitHub, all 18 of our Kowalah dev repos, got access to all of those. I can attach the relevant one, read code, open PRs. A sandbox, artifacts, and memory and schedules. Okay, so I might just say, okay, great, can you take a look at the Kowalah Agent Hub repo? I'd love to know a little bit about the different agents that we've got live in the Kowalah Agent Hub, and maybe we can think through creating a new agent there. **Charlie Cowan:** So just to mention, I didn't tag Claude specifically in that. I didn't do at-Claude, like I've got, if I did down here, I could add in Claude, because this is what's called an ambient channel. When you add Claude into a specific channel, it's just listening all the time, and it will make a decision about whether it should respond or not. So, very much like how you would be chatting with humans, yes, you can at-mention it, but you don't need to. It can just pick that up automatically. Couple of other things while we're here. You can, if you're the admin of a channel, you can then configure that channel itself. **Charlie Cowan:** I'm going to click on this button, and I'll show you what this opens up in my browser when I switch my sharing, but it's going to give us some abilities to configure what happens specifically in this channel, where we can give it a bit of specific training. So I'm going to click that, and then I'll just come back to my Slack. I can see that Claude is taking a look at this. Let's see how quickly Claude comes back to that. I'll pop back over to this a little bit later when Claude has finished doing its work. It's busy looking at all of our repositories, which is lovely. So let me stop sharing. And, oh, Claude's back, what were we saying? Oh, it's built out a task list. **Charlie Cowan:** So I'm finding and attaching the Agent Hub repo, I'll read how the agents are defined and registered, I'll then summarize the live agents and what adding one involves, and so it's built out a task list, so it's getting to work. I'll pop back to that a little bit later. Let me go back to our screens. Let me stop sharing. And then go back to my screen here. And let's share again. Google Chrome. Alright, so I mentioned a minute ago that when I was in that channel, I would click this button, and it would tell us how we could configure that channel. And so because I'm the owner of that channel, I'm able to configure Meet Claude Tag at that channel. And so what can I do? So, the responding automatically is what I talked about, it being an ambient channel. So when off, Claude replies here only when at-mentioned. **Charlie Cowan:** Anyone in the channel can also change this by asking Claude in Slack. So we have this on, and so Claude just gets involved in the chat. Now, you can imagine if you've got a deal channel, where you've got a salesperson, a sales manager, and a solution architect chatting about a specific deal, Claude's there listening, and when Claude thinks it can be, you know, helpful, and has got something to add, Claude will chime in without you having to at-mention Claude the whole time. You can then start to configure that a little bit more by providing some channel-specific instructions to guide Claude about when to chime in, what types of things she should help with, you know, help us with product roadmap, help us with negotiating, help us with, you know, planning emails out to clients, whatever that might be. So this is just general, channel-specific instructions. **Charlie Cowan:** In addition, you can start to customize what connectors an individual channel has got. Now, two things here. One is connections, so what are things that can be, you know, are configured for this channel. So if it was, I'm going to stay with my sales scenario, you might have specific connectors into HubSpot, for example. You might have specific connectors into your pricing tool, for example, in here that Claude has got access to. And what you can also provide is access to specific plugins, which is a holder for skills. So, here we could give this Claude access to our negotiating skill, our prospecting skill, our proposal builder skill, and that can be enabled to enhance this Claude in this channel in a way differently to the other Claude in other channels. And the third thing we can customize is what's called a routine, and here you can think about a scheduled task. **Charlie Cowan:** So, imagine that we've got a company-wide sales channel, and Claude is in that company-wide sales channel, we might have a routine that says on Monday morning, we want Claude to provide the latest forecast update in the channel. And that routine is gonna run in the background, it's gonna look at HubSpot, it's going to look at Salesforce, it's gonna run its skill about forecasting, and then it's gonna pull that into Slack. You know, morning everyone, this is the latest, and here's my, you know, assessment of who needs to do certain work. So these are the things that, as a channel owner, you can configure on a channel-by-channel basis. This is really important, because this is how each of the different Claudes and different channels can act in a very different way, compared to each other. **Charlie Cowan:** So that was a bit about you as a consumer, or a channel owner, or a member of a channel. Now let's have a look at some of the things that you can do as an admin to set up and to configure your Claude Tag. So, I'm in my Claude settings as an administrator. I'm in my organization settings here, not personal. So this is where you're adding in members, capabilities and so on, and you're going to see Claude Tag on the left-hand side. Now, if I scroll up to the top, you'll see a few things that may be on, may be off, so enable it if it's not enabled, allow DMs, can all of your individual people message Claude directly, or just in channels? Channel spending limits. So let's go to manage here. Now, we've only got one channel that we're using for testing at the moment. **Charlie Cowan:** I'll explain a little bit why in a moment, but this is where we could say, for this limit, I want to use a custom spend limit in this channel. And the limit is going to be $200 in that, and that is monthly. That then means that anyone that's using that channel, we can limit it to that. It also means that you can restrict which channels have got access to Claude Tag as well, so you've got a lot of control over what is being spent here. Let me go back to Claude Tag here. So that was your channel spending limits. Now, where is Claude Tag working? As I mentioned, at the moment, only in Slack, but you could imagine a world where this is gonna come in and say Teams later on, or Google Chat, and a few other things. So that is set up there. Can we manage that? **Charlie Cowan:** Yes, yes, yes, restrict in your organization. Okay, all good. So that's setting up Slack. Then we've connected it to our GitHub. We are obviously a development software company. We've got access to GitHub. This allows Claude to interact and take actions in your GitHub repository, so get that set up as well. Now, what can Claude do? By default, you're going to set up your Slack connections, and later on I'm going to show you how you configure that by a channel. But as a default, you're going to set up your default Slack access. If you do not do anything else in a channel, then this is what the experience will be for each of your users. So, Claude Tag version, we want the new version, not anything legacy, that old version that I said. We want Claude Tag being used. Now I'm going to introduce you to a thing called an access bundle. **Charlie Cowan:** So I said at the start that when you use Claude on your desktop or in the browser, what Claude sees is what you see, so it's just an extension of you. But now, Claude can have its own persona, its own access, and this is great, because you're able to give different Claude different access to different things. And they're called access bundles, and so I'm going to just jump over to the left here, where we've got access bundles, and I'll show you a little bit about what is going on here. So we've just set up one for these demo purposes, which is called Sales Claude. And if I click on that, you'll see what does Sales Claude have access to. So, firstly, credentials. We can give Sales Claude access to different systems in its own right. So, we might have, let's think, I mean, there's loads here, but we might have HubSpot as a sales platform. **Charlie Cowan:** We might have, well, we've got Apollo here for getting contacts. We might have Notion, where we're doing certain work. We might have Google, Gmail, Calendar, and Drive. Now, the key thing here is, you're not connecting it as me, you're connecting it as Claude. So, working with your IT team, what I would recommend, well, so what I recommend, so Anthropic recommends, is that you set up claude at kowalah.com as a Google account. Now, if you're going to create multiple access bundles, you might create sales-claude at kowalah.com, marketing-claude at kowalah.com. And this is creating these different identities and entities that you can give different permission to different tools. Now, why this is important is because, depending on your different use cases that you're going to start to think through, different channels may require different access and different permissions. So, in your Finance Claude channel, for example, you may want to give access to Xero. **Charlie Cowan:** Think that's going to be here? No, it's not going to be a custom tool. But there may be, let's think, maybe there's QuickBooks. No, that'd be a custom tool as well. I don't know if there's any finance ones on here. You'll figure it out. But depending on what systems and access you want to give channels to, you're going to create different access. Snowflake might be a good example, then. You're going to create one access to Snowflake for your finance team, which has got real detailed stuff about pricing. I don't know, maybe payroll and things like that. There may be another Snowflake access, which is for salespeople to get access to information on deals, for example. That would be a good scenario. So, that's a reason why you'd have multiple access bundles, rather than just having all in one. Secondly, access to repositories. **Charlie Cowan:** So, you may not want all of your channels to have access to your code repositories, that might not be relevant. Domains. When Claude is working in this channel, what should it be able to do? Can it access specific external domains, or is it restricted into just a certain subdirectory? Plugins. I mentioned a few of these. So, some of these will be provided by Anthropic. So, you know, there's an Apollo plugin, there is a Box plugin, basically teaches your Claude how to use that integration. And then you may have some that are your own. I wonder if I've got some Kowalah ones down here. No. Kowalah, oh yeah, we've got the writing toolkit plugin. Okay, so this is how you can create these plugins centrally, and then deploy them. This would mean our writing plugin, which I absolutely love, would be useful in the sales channel. **Charlie Cowan:** Hey Claude, can you help me to write an email out to this customer? Claude's got access to our writing toolkit for how we like to write and our tone of voice, for example. And then thirdly, what did I say? Thirdly? 1, 2, 3, 4, 5. Fifthly, we've got instructions for how to use this access bundle. You can provide a bit more instruction about how Claude should be using this access bundle, what it's got access to, you know, who to ask for help if it gets stuck, things like that. So that's your access bundle created. You can create one, you can create multiple of them. Let me go back to our Claude Tag, here, where we're setting up our default Slack. So, access bundles, well, we want to use the Sales Claude access bundle here. **Charlie Cowan:** This is where you may actually have a default one, and that's got quite restricted in terms of the access that it provides. And then you might have individual ones, for different teams, where you open up that access, or you provide it a bit more access. Repositories, we talked about that. Connectors, we talked about that, so you can provide these over and above the access bundle. The plugins we talked about. Okay, and then more custom instructions. Now, you're getting a feel here for these custom instructions. They are multi-layered. We've seen custom instructions at the channel level that the manager of the channel might put in Slack. We've seen custom instructions at the access bundle level, how to use that access bundle, and we've seen custom instructions at the Slack overall level, which can be managed by the administrator. **Charlie Cowan:** So, beware of these different levels, and making sure that they're not conflicting with each other. You don't want one saying, do this, and the other one saying, don't do this. And then next up, we've got some advanced configuration. Now, just as a reminder, what we're configuring at the moment is overall Slack. We're not doing it at different individual channels. I'll show you that in a minute. So, the default model for all of Claude Tag across all of your organization, well, we can set that to Sonnet. Sonnet 5 is a great model, and for most knowledge work is absolutely sufficient. So, we can set that at the default, and that immediately restricts all of this, you know, high expenditure from individuals going, oh, I want to use Fable 5 to plan my one-to-one with my manager. So, that is a, you know, immediate default to go for. **Charlie Cowan:** Setting the environment, we can set that up. Allowing Claude to work with channels with guests. So, do you want Claude to be able to interact when you've got customers in a channel? Now, that might be good, and you might set up your Claude to be very effective at helping customers by giving them plugins and skills that are useful. You may well have, in your customer success team, shared channels with key customers. Well, imagine bringing Claude into that conversation, and having Claude support those channels. Can the channel members change the channel instructions? We just talked about that. And a few more things about permissions, channel, okay, happy days. So, those are a few more of the settings that you can have. Now, that is your overall Claude Tag, and when you're first rolling this out, you could just leave that as it is. **Charlie Cowan:** What you then might want to do is then say, right, well, now I want to create individual channels with different settings. So, this is our Meet Claude Tag demo one. So, what Claude Tag version do we want? Inherit from the top version. Access bundles, inherit from above. Everything else, yeah, inherit, inherit, inherit. Advanced, inherit, inherit. But if I want to, I could change it. So, where we've got Sonnet 5 as the default for all of Slack, if this is our, I don't know, our executive, strategic planning channel, and only our senior leadership team are in here, well, maybe we're saying, well, actually, we do want Opus 5 for this, because we're working on more complex analysis, and so on. **Charlie Cowan:** So this gives you this ability to define what's happening at an individual Slack channel level, and it's super easy just to add in more channels as you go, and then to define them slightly differently. So we covered quite a bit there, but that's basically how you're setting it up. Let me go back over here. We've covered this, so governing centrally, scoping the access bundle by bundle with those access bundles. We looked at deciding which models you can use, and we looked at individual channels being able to have their own budget as well, that you can set, so you've got some control over that. So what I would suggest is, if you're, well, number one, are you Slack users as an organization? **Charlie Cowan:** If you're not, then maybe you stopped watching a while ago, but if you're Slack users, I would be turning on Claude Tag, and I would be creating at least a default access bundle. I would be creating at least a claude at yourcompanyname.com account in the key systems that you think Claude would need access to. And then I would share that out and give that access in a couple of tightly scoped channels. You could turn off the DM setting, so that you're not, you know, opening this up to thousands or tens of thousands of people DMing Claude all day, just keep it in a few channels, and just get learning, familiar with how the access bundles work, how you can fine-tune the instructions for each of those at Claude's. Alright, hopefully that was helpful. **Charlie Cowan:** If you've got any questions on that, then feel free to either let me know directly, you can email us here at Kowalah. If you're watching this on YouTube, then you can add a comment, and we'll come straight back to you as well there. Oh, before we do go over that, I wanted to show you a couple of things that, I just, let me just stop sharing, and then share Slack again. Where am I going? Share, Slack, Slack, here we go. Claude was busy off working there, if you remember, so I asked, can you take a look at the Agent Hub repo? I'd like to know a little bit about the different agents. And so, Claude built a task list. It's cloned the Agent Hub repository, I've read the registry, I've confirmed there are 17 agents registered, and I've posted the summary. And here we go, 17 agents are registered in the Agent Hub. **Charlie Cowan:** Da-da-da-da-da, we've got some revenue agents, Brian, McMahon, and Coco. Delivery agents, ops and exec agents, and we've got some behind-the-scenes agents. And so this is Claude interacting. You can see how I would very easily be able to, let's just carry on chatting. Great, can you help me to scope out a new agent that I want to build? Can you start to put that in the GitHub repository as a pull request that I can go and review? You just start getting work done in the place where your team are collaborating around this work already. And then the final thing that I wanted to mention to you, I touched on this earlier, well, so why aren't we using Claude Tag fully here at Kowalah? And that's because we're probably one step further along than where Claude Tag is at the moment. We already use Claude managed agents, and we use them extensively. **Charlie Cowan:** This is our agent hub, and here we've got all of these agents that are deployed, as Claude was mentioning. These are our revenue agents, our delivery agents, our ops agents, our product and finance agents. And if I go to one of these agents, let's say, where's Carolina gone? Oh, she's here, there we go. Carolina's our project manager agent, she runs all of our projects. These agents are running independently. They're set up in our Google Workspace, carolina at kowalah.com, and we're able to chat with them in Google Chat, because we use Google. We don't use Slack as our primary chat. And we get everything that we want from chatting with these agents. We give them individual names, and we treat them really like members of the team. We love chatting with them. We can't use Claude Tag because we're in Google Chat rather than Slack as our main collaboration tool. **Charlie Cowan:** But secondly, these would all be called Claude. You know, Claude in a project manager channel, Claude in a finance channel, Claude in a product management channel. And we love the fact that they've got names and personas, and they're, you know, people that we've worked with in the past, and we feel this really brings them to life. I've got no doubt that Anthropic may go down this route, because it really helps you to know that one Claude is different to another Claude, is different to another Claude. You know, these are all running at Sonnet 5 in the background. It's exactly the same architecture in the background, but we've got this personality over the front of them, which, you know, our team really, really enjoy. Okay, let's dive back into the main deck for today, and switch over to AI in the news. I'd say a little bit of a quieter week this week. **Charlie Cowan:** No sort of big releases like we've had in some previous weeks, but a few things that is worth talking to you about. This is about to come out, because it's saying rolling out later this week, but one of the challenges that Anthropic have had, and Claude users have had over the previous weeks and months is you've really got sort of 3 different products in one surface. You've got chat, you've got Cowork, and you've got Code. And, for the majority of knowledge workers, you can take Code out, they're not coding, so then you're stuck with chat and Cowork. If you're not familiar, chat is really, you know, Q&A. I've got a question, you come back to me with something. Cowork is about building a task list, and taking action on something, and maybe running over a long period of time. **Charlie Cowan:** Often, Cowork runs on your desktop, your laptop, and so it's got access to files and systems, and it can do stuff. What Cowork has been doing in the past is managing all of that memory locally on your laptop, and what they're now doing is merging this so that for your remote Cowork sessions, it's going to use the same memory that your chat has. It's all about bringing this thing much closer together, and taking the decisions off individuals who shouldn't need to have to worry about what memory is where. So, this is going to be turned on as a default at the organization level, if you're on team or enterprise, and then users will have, as you can see down the bottom here, users can turn this on individually, and they'll be able to edit and delete everything that Claude has saved themselves. So, yeah, look out for that over the next week or so. **Charlie Cowan:** Next up, also on Anthropic, is announcing that Bain has joined the Claude Partner Network. Now, for those of you not familiar, Bain is one of these large, management consulting businesses, up there with, I guess, the McKinseys, the BCGs, these types of organizations that are helping the world's very largest organizations with their strategy and, you know, technical thinking. They are rolling out Claude themselves to 20,000-odd employees, and then they're embedding Claude in the work that they're doing with their clients. I think it's just another validation of Anthropic really being right at the heart of how enterprises are thinking, and this is good validation of Claude's fairly young partner network, but is gaining some great partners at the top level. **Charlie Cowan:** Next up, and I will do my best to translate some quite technical stuff here, OpenAI announced, a few, we'll say months ago, that they were starting to build out their first custom inference chip. So the chips that are working in their data centers to run inference, which is when you are having a chat with ChatGPT, and ChatGPT is working a way to give you an answer, inference is what is happening, and these chips run that inference. And their strategy is that if they can design their own chips that really understand their models, they'll be able to surface answers more quickly for you as a user, but by doing that, that will reduce their cost and their compute required in order to service the increasing demand that they are seeing. **Charlie Cowan:** So, I won't even attempt to digest what this chart is showing, other than to say that on the y-axis, you've got mixed tokens per second per kilowatt. So, this is the number of tokens that are being used in order to serve something up. And, obviously, the green bar is better than the blue bar that it was. There are plenty of other charts that you can see. We'll put this in the show notes, but they've got plenty more different dimensions about how they're measuring this chip's models, peak per-user decoding throughput, more throughput per kilowatt. This is all about the environment, it's about cost, it's about giving you quicker answers. So that was recently announced just yesterday, 25th of August. If you are Google users, then a couple of new releases that you might start to see in your Gmail and your Google accounts. So the first is Google Chat. **Charlie Cowan:** So, Gemini was previously a separate tab that you would click on to go and chat to Gemini somewhere differently, where, I mean, this really builds on what we talked about, AI in the flow of your work with Claude Tag, is that now you're going to see Ask Gemini in your Google Chat tab. It's where you're chatting already with all of your other colleagues. Well, now you can chat with Gemini just here. So, take a look at that. I would imagine more and more of this is embedding Gemini and being able to tag Gemini in the chats you're already having, following Claude Tag's lead. The next one is a similar kind of thing, but in Chrome. So, I don't know what I am sharing, how much you can see, whether you can, oh, you should be able to see my entire browser here. **Charlie Cowan:** So, you'll see this Ask Gemini tab up at the top, and if I click that, I'm going to get Gemini here. And I'm sharing the weekly webinar deck. Hey, Gemini, can you tell me what we are looking at? And so, now we can have Gemini working in and around the Chrome pages that we're looking at. If you've been using Claude in Chrome, you will get a good feel for what this looks like. You are looking at slide 16 from a presentation deck titled Weekly Webinar Deck, which demonstrates the Gemini integration. And I haven't tested this out yet, but no doubt we can add in skills and a few other things, a bit like we can in Claude in Chrome. And one of my favorite use cases for Claude in Chrome is not asking what I can see here, it's fill in some things. **Charlie Cowan:** So, if you've got a form, maybe it's, you know, I don't know, an RFP response that you've got to fill in, maybe it's a Google Doc, maybe it's a spreadsheet, you can start to have Claude filling that stuff in by looking at what is on the screen. So, have a play around with that if you are Google customers. And that brings us to the end of today's Wednesday webinar. I haven't seen any other questions in the chat, so thank you very much for joining live. If you do have any questions and you're watching this on YouTube, then feel free to add your question in the comment, and we'll put that there. Next week, we're gonna go back to a topic that we're talking a lot about with our enterprise clients, which is to focus on the workflow, not on an individual task or use case. **Charlie Cowan:** Where we are seeing so much impact is where a company looks at a top-level domain, a top-level process, such as your sales process, such as your product design process, such as your supply chain, and looks at that process end-to-end, and then goes, right, how do we rebuild this for the AI world, rather than looking at individual items in it as an individual use case. And we'll use an example, a great post that was posted on X a few weeks ago from the CTO of Uber, who ran two-week sprints across all these different workflows, and just rebuilt those workflows, and got a much greater impact than just picking off the individual use cases. So, I'll see you same time next week, Wednesday, the, I think that's gonna be the 2nd of September. Look forward to seeing you then. Thank you. --- This transcript is published by Kowalah, a UK Anthropic Implementation Specialist. Webinar page: https://kowalah.com/resources/wednesday-webinars/claude-tag-ai-teammate-in-slack All transcripts: https://kowalah.com/resources/wednesday-webinars/transcripts