Claudeforce Explained: Three Ways to Connect Claude to Salesforce

Charlie Cowan: All right, good afternoon, good morning, depending on where you're dialing in from. Welcome to this week's special edition of Kowalah's Wednesday webinars.

If this is your first Wednesday webinar, we get together at the same time every week for an hour to talk about what's going on in the world of AI — specifically enterprise adoption, organizational change, and transformation relating to AI. It's such a fast-moving space, and this gives us time to take stock and think about what we're doing within Kowalah, what we're seeing with our clients, and what's happening in the broader industry. Today we're going to be talking about a very topical topic: the integration between Claude from Anthropic and Salesforce.

I'm very excited to introduce and welcome Dennis from our team. Dennis heads up delivery. Dennis and I worked together previously at a leading Salesforce boutique, so we've both got an interest — from our Salesforce background and from what we're doing with Anthropic. Dennis, maybe you'd like to say a few words and say hello.

Dennis Groenveld: Yes, hi everyone — Dennis Groenveld, indeed heading up delivery. It feels like coming home again. I started, when was it, 2012, I believe — with Charlie, doing all kinds of exciting stuff with Salesforce and Google, around Europe, around the world even. And now, Claude. So this webinar, for me, is coming full circle.

Charlie Cowan: Yeah, beautiful. Well, before we get into today's topic, here's how the flow of the session will work. We've got an hour — we'll probably spend 30 to 40 minutes on the main topic of the day, Salesforce and Claudeforce. For those in the Salesforce ecosystem, it's Dreamforce this week, their annual festival in San Francisco, so I'm sure we'll have plenty to talk about there. Then we'll move into our recurring segment, AI in the News. My goodness — I say "AI in the news," but AI has literally been the lead story on the news here in the UK, and I'm sure around the world, so we'll talk about why that is and what's useful for you to think about in your own organizations. Then we've got time for Q&A.

If you're watching live, you don't need to wait until the end to ask a question — if you've got a question about Claudeforce or AI in the news, we're in a Zoom meeting, so depending on your device, you'll see a chat widget and a Q&A widget near the top or bottom of your screen. Post in the chat if you want others on the call to see your comment, or use the Q&A widget for a specific question — either works. Dennis and I will keep an eye on both, and Caitlin, who's on the line as well, will keep us on track if we miss anything.

If you're watching this on YouTube — and we know more people watch on YouTube than watch live — feel free to add a comment or question in the comments section of the video. That'll come to us, and we'll come straight back to you.

So with that, let's move into today's topic — and everything Dennis and I are going to chat about comes with a lot of love. We've spent a lot of time in the Salesforce ecosystem, so we're talking from experience.

Let me say, with a smile on my face, that Salesforce is very good at marketing lots and lots of stuff and seeing what lands. A few weeks back there was an announcement of Claudeforce — one of the first times Salesforce has put another company's brand on a product they're selling. My understanding is that Claudeforce is the partnership, and the result of that partnership is a product called Salesforce in Claude — we'll talk about that in a minute.

Then, just last night at Dreamforce in San Francisco, they announced AI Force. The first time I read that, I misread it "Who is Al?!" — I had to look twice before realizing it said "AI Force." AI Force is more about the plumbing and infrastructure behind all of this.

Dennis, you've been digging into this — what are all of these things?

Dennis Groenveld: First of all, I need to say that whatever we say here is going to have a very short shelf life — the naming and the product positioning have turned the world upside down just in the last two weeks.

In that time, Salesforce reversed a renaming exercise from around 2024: Sales Cloud was going to be called Agentforce for Sales, and it's now back to Sales Cloud. I believe about 24 products went back to their old names.

Right now there's a real AI soup of product names. There's AI Force, which is a marketing layer — I'll come to that later. There's Agentforce, which are the actual agents. There's the Agentforce co-worker, which is an in-app assistant. And there's Claudeforce, which — like you said — is the partnership between Anthropic and Salesforce as two companies. That goes much further than people realize, because Salesforce has decided to use Claude Code for their engineering teams, and Anthropic has adopted Salesforce for their own CRM — so there's a lot going back and forth there. The product itself is called Salesforce in Claude.

Then there's Slack's four services: Slack CRM, Slack Code, the Headless Toolkit, Data 360, Customer 360, and a very interesting one — KOA, which is their own model. I can't promise you that a week from now these things will still have the same names.

Charlie Cowan: It's moving so fast. You mentioned Agentforce there — I've been out of the Salesforce ecosystem a little bit. My understanding was that as Salesforce was really getting going on AI, with Benioff all-in on it, Agentforce was a platform inside Salesforce that helps you build agents, but you could plug in other models — it was the application layer that would connect to your OpenAI, or whichever model you chose. What do you understand Agentforce to be?

Dennis Groenveld: Agentforce is the set of agents that live inside Salesforce. Drawing on my recent Salesforce experience, I'd say it's been successful, but not the breakthrough everyone expected. If you look at the partner ecosystem, there are a lot of proof-of-concepts out there, but not a lot of real, big projects — and I think it also took time away from Salesforce's sales teams, who forgot to sell the core product. For two years they've talked about Agentforce, Agentforce, Agentforce, Agentforce, and with the current renaming exercise — back to basics — I feel they've come to their senses again. Agentforce is here to stay, definitely, but right now Salesforce is positioning itself as the platform that safeguards permissions, keeps your data consistent, and safeguards the gateway to the flows — so it's open for everyone.

Also, it didn't make the news yesterday, but there were wider partnerships announced with Google and Gemini, and with OpenAI — so Benioff is in bed with all of them at the moment. But yes, Agentforce lives inside Salesforce.

Charlie Cowan: I think — maybe a year or so ago, I was listening to a podcast with Benioff, and at that point I think he was bullish that Salesforce could be the hub, with other tools as the spokes. I think he's since come to realize that Salesforce itself, as a system of record, is a spoke — and the hub is going to be someone's Claude. As a seller, you're working in Claude; Claude is connected into your calendar, your email, your Granola notes, your Google Drive, and Salesforce is a spoke of that. If you're not partnered up and getting your data into that hub, you're going to see traffic to salesforce.com decline as people choose not to go to one SaaS platform or another.

Dennis Groenveld: Yeah, that's correct.

Charlie Cowan: Let me pull up a couple of additional slides — I think you were talking about these different platforms that have launched. There's an interesting angle here, which I find ironic — or coincidental, at least: what Benioff has actually got is one of the biggest hubs and entry points into AI, and it's not Salesforce, it's Slack. I think it was around six years ago that Salesforce bought Slack for something like $25 billion. At the time, some quarters derided that — why is Benioff overpaying for a comms platform? That's not what people use Salesforce for. But as time has gone on, modern tech companies — certainly ones like Anthropic and OpenAI — live in Slack. It's their communications layer; not Teams, not Google Chat, Slack. And Slack is now becoming an entry point for all sorts of AI agents — you can chat with the Kowalah agent in Slack, the Linear agent in Slack, Claude Tag in Slack. Even though that's not talked about much, I think it's going to work in Benioff's — and Salesforce's — favor. Not sure what your opinion is on that, Dennis.

Dennis Groenveld: Slack getting this new position is great for Salesforce, I think, though I don't believe Benioff knew this when he bought it — we didn't have an AI revolution on our hands then, so I won't grant him that foresight. What I do think is that Salesforce really understands how it needs to change its position as a SaaS platform: it's not the UI that counts anymore, it's everything underneath. People are going to consume Salesforce through their AI of choice, or their UI of choice — and that's the layer underneath AI Force. It's what we thought, the day before yesterday, was called the Headless Toolkit — because Headless 360 has been renamed back to Salesforce Platform, like it used to be. That platform lets you build your own UI, which is great, because Salesforce was never the greatest with UIs.

Charlie Cowan: Exactly — you're going to want to start building this stuff yourself. Let me show a few things here as we build out the conversation.

Salesforce in Claude is the product, built under the Claudeforce partnership. What is it? You go into Claude — ideally as an admin of both Claude and Salesforce — and click on Customize. There you'll see skills, connectors, and plugins. Connectors is what we're talking about here: search for Salesforce, and you'll find this beta version, "Sell, serve, and operate at scale."

This is a connector in the same way you might connect Granola, Google, or Microsoft — specifically, it's a connector into Salesforce's Headless 360, or Headless 360 MCP. I'll show you that page in a minute. It exposes four tools: describing the schema (what Claude can see in your schema), discovering individual records and objects, dispatching (getting work done in Salesforce), and a read-only tool. You can also find the documentation, which I'll show you in a minute.

Depending on your role, you can connect for your team or just for yourself. If you click "connect for your team," it's not quite that simple, because this is beta — you need to request access. When you click on that, you'll end up on the Agent Exchange page, where you enter details about yourself, your Salesforce org, and your Anthropic org, then sign up to a few things and wait for approval. It's an open beta, so you will get approved, but there's some form-filling involved to get set up. Once approved, that enables the connector.

That gives you access to... here we go: "The number one AI meets the number one CRM. Welcome to Claudeforce. Put Claude to work." There are plenty of videos you can watch to see it in action. At launch, it's focused on sales — no Service Cloud, no Marketing Cloud, just Sales Cloud — but it comes with 37 skills. When your sellers are in Claude asking questions, it's the combination of the skill plus the connector that makes it work.

You can scroll through the list of skills — deal signals, pipeline review, generating close plans, handling objections, account planning, customer voice, being the most prepared person for a scheduled meeting, and more. You might have already built quite a few of these skills for your sellers, so you'll want to look through the list and work out which are generic ones from Salesforce that are appropriate to your business, versus where you'd rather enable your own skills. For anyone who hasn't built out a large number of these skills already, this is going to give you a real head start — they're built around general usage of Salesforce, so it's a good place to begin.

The only other thing I wanted to show is the documentation. From the connector area in Claude, clicking "documentation" takes you to salesforce.com's docs for MCP Servers, Headless 360. You can read through exactly how it works, and importantly, see the available tools: Discover, for helping Claude figure out what's in your schema; Describe; Dispatch, for going and doing something in Salesforce; and a read-only Dispatch tool.

This is interesting, because if you're building your own MCP servers, you might assume that wiring one into something like Salesforce would mean thousands and thousands of different API calls. But the best practice from Anthropic is to have fewer tools that do a lot more — so here, all of Salesforce is exposed through just four tools, each with quite a lot of complexity underneath.

Dennis, I covered a lot there — the setup, the connections. Anything you'd want to add?

Dennis Groenveld: I think this is a great start for sales teams and sales leaders who want to consume their Salesforce data through Claude. Working at Kowalah, of course, we do more than Salesforce — we automate whole workflows, and I find the layer underneath this the most exciting part. That's the Headless Toolkit. Claudeforce — the partnership between Claudeforce and Salesforce — is built on top of the Headless Toolkit, so you can do this yourself. I look at Salesforce in Claude a bit like a reference app.

But if you want to build a whole workflow that spans five other systems plus Salesforce — say, self-service actions from your website that update Salesforce — you can build those applications now, on top of Salesforce, using the underlying plumbing that powers what we're calling Claudeforce. That seems to be what the whole world is doing.

And I think that's the future of Salesforce, really — they become the back office, sitting on the permissions. That's another thing worth flagging to everyone: it's not uncommon for sales staff or sales leaders to have accumulated more permissions than they should within Salesforce, without even knowing it — meaning they can do things they're not supposed to. The first thing Claudeforce will do, in this case, is figure out what's possible: it uncovers, on its first run, everything it can reach. So I'd argue that before you turn this on, you should review your permission model. You could burn your fingers if you don't — that's a good exercise to do first, because AI is going to uncover all the nooks and crannies you can get to, ones a human wouldn't even know existed.

Charlie Cowan: There are two things I think of here. One is that when your sellers are using Salesforce for Claude, or any connector they've turned on in their Claude, Claude is interacting as that individual. So, for example, when I'm logging in or asking Claude to search Gmail, it's logging in using OAuth to my Gmail and searching my Gmail. When it searches Google Drive, it's looking at files I own or that have been shared with me. Same with Slack, or HubSpot, which is our CRM. So what you're really doing is taking that individual salesperson's persona and using it to find information.

That's different from what we might call managed agents — where you want an agent that can run a forecast, or help people prepare for meetings, on its own. That's where you separate the identity of an individual running something on their own laptop from an identity an agent uses. Which leads me to my second point.

It's very tempting, as all these connectors come out, to check for security issues and then approve them into all of our systems. Here we've got the key ones we use — let me show all of them. Internally we use Google; we should have Granola in here somewhere, which we absolutely love; and DocuSign for our contracts, among others. So it'd be natural to just connect everything up.

The problem with that — or the myth — is this: if I say something like "help me build my forecast," as a human I'd actually be accessing multiple different systems. I'd be checking my CRM, my call transcripts in Granola, my emails, maybe my calendar, Slack. All of these things feed in as context to build that forecast. But if you only use the standard connectors — the tools coming directly from providers, like the 37 Salesforce provides — that tool only interacts with Salesforce, because that's the only thing it could possibly know.

The way we've actually implemented this within Kowalah is by building our own MCP server, which sits at the top here — this is our Kowalah admin connector, for our own people. I won't go through it in detail, but these are tools that let our internal employees manage our clients and their work. Behind each tool are multiple APIs — it might be looking at Granola, at HubSpot, at Xero to check whether invoices have been paid, at email, and so on. What we're doing is creating tools that are intent-based rather than tightly scoped to one system.

So my recommendation to anyone thinking about these connectors: definitely use them as a starting point — someone's done the work for you already, building the connector and some tools to use it. But as a leader of a team or a process, I'd go up one level. Instead of just sprinkling AI magic over your current process, think about what the new process should look like, and whether you need to build your own connector to support that. Any reflections on that, Dennis, or what are you seeing?

Dennis Groenveld: Two things, and it's all about the money. Every connector has a number of tools attached to it, and those tools carry descriptions — a lot of metadata. Every time a tool is loaded into your conversation, that metadata becomes part of your context window.

So here's a free tip: in a chat, click the plus icon, go to connectors, and right at the bottom there's a "tool access" setting. If it says tools are already loaded, all the tools you've turned on are part of your context window, and you pay real money for that. If you set it to load tools only when needed, you lower your bill. That's today's free tip when you've got a lot of tools turned on — it's really a trade-off between cost and availability.

The last thing on pricing has to do with Salesforce in Claude again: you pay on two ends. Salesforce prices access to the Headless Toolkit based on consumption — how high that is depends on your own Salesforce license, so it's different for everyone. And on the Anthropic side, you pay the normal token costs as well. So, yes, you pay at two ends.

Charlie Cowan: I'll go back to the slides to wrap up this section in a second, but we've talked a couple of times about "Headless 360," and I think you're going to hear more and more about this. What does headless mean? It means there's a system, an application there, but as a human you're not actually going to the UI — it's just the database in the back end. In this scenario, Salesforce is the system of record, tracking your opportunities, your service cases, and so on. The "head" is Claude, or it might be ChatGPT, or Gemini, or Copilot.

This comes up a lot in client conversations: people say, if I'm not going to Salesforce anymore, and it's just a database and a schema, could I code that myself and build it? Why do I need to pay Salesforce, either for user licenses or for traffic and consumption? I'll share my thinking, and then I'd love to hear yours, Dennis.

It's easy to think that SaaS platforms like Salesforce, Workday, Xero, or NetSuite are just a database with a form on the front — it's a meme, a simplification. My experience, having worked within the Salesforce and Workday ecosystems, is that there's a lot of business process, security, and reporting behind them. Just think about sales teams, territories, and quotas: for a system of record like Salesforce or Workday, there's a lot more than a form on the front of a database — there's workflow, security, and governance. Anyone who thinks they'll just build their own Salesforce will quickly run into the reasons why these companies are so big. From a sales perspective alone — territory management, forecasting, quotas — it's way more than a table with contacts and opportunities in it.

So I think you're going to see more and more of these headless platforms, and the charging model is going to shift more and more from users to consumption of access. I don't know exactly what the charging model will be for Salesforce in Claude, but we're seeing this across other platforms: you'll get a certain amount of capacity or consumption for calls into the platform, and that's what you'll pay for, rather than paying per seat.

Dennis Groenveld: Yeah, that's very similar to how cloud platforms — GCP, Amazon, Azure — work: consumption-based metering.

On the build-your-own question: ever since building applications with AI became cheaper than building them the old-fashioned way, the build-versus-buy decision has definitely shifted a bit toward build. I'm with you that you shouldn't try to build an equivalent to Salesforce — it's not just a database. All your business rules and permissions are in there, and if you look at the flows alone, they're huge. So no, I wouldn't say build Salesforce yourself.

That said, for other, smaller tools — a lot of the tools in HR, for instance, like a holiday registration system — I'd be very tempted to build that myself, because the number of business rules and permissions you'd need to code are way less than in Salesforce. I do think building headless systems will happen more and more. But I won't advise anyone to build Salesforce — that's not going to be a successful outcome, in my view.

Charlie Cowan: The phrase I always go back to is a quote from Jeff Bezos, which is to focus on what makes your beer taste great. If you're a brewery, you should focus on brewing your own beer — maybe on the pubs where you sell it — but you don't build your own payroll system, because it doesn't make the beer taste any better. I carry that through in every project we work on: if something meaningfully changes the customer experience, your employees' ability to deliver that experience, or your ability to bring new products and services to market, that's the kind of thing you should build, because it's a differentiator. But the way you pay people is not a differentiator for you — and in many cases, the way you track an opportunity or close a support ticket isn't going to be a differentiator either.

Think about what makes your beer taste great. We've covered a lot of this already, so I won't go through it all again — we'll share it in the show notes so you can access it. We talked about governance and permissions, and showed exactly how to find the connector in Claude, where the documentation is, and the four tools you get access to. If you're a Salesforce customer, you can fill in the open beta — it's not a closed beta — and you'll be given the details. Pricing hasn't been announced yet.

So: prepare for launch. A couple of things you can do between now and having access — as Dennis said, remind yourself of your permissions and who's got access to what; maybe launch it with one team before rolling it out more broadly; and start thinking about how you're going to pay for this, and which budget it comes from.

My final point here: we work with a lot of clients who have their own products that they sell to their own customers, with their own portal or UI. I'd encourage everyone to be a consumer of some of these headless-product connectors and start thinking: if it's good enough for you, it's probably good enough for your customers. How do you turn your own SaaS platform into something headless? More and more, the theme I'm seeing online is: if you won't give me, as a customer, access to my data in my chosen AI, I'll find a product that will — I want to use my agent, not your agent. So flip that around and think about your own customers: they want to be talking about your data in their Claude or their ChatGPT, not coming to your own platform. If you're not already thinking about that, I'd start now.

Let's talk now about AI in the News — Dennis, I'd appreciate your input on some of this too. It's been quite a week.

On September 9th — about a week ago, just after last week's Wednesday webinar — a researcher, a Brit working in San Francisco, resigned from Anthropic. He said he'd spent the last three years doing pre-training research, first at OpenAI and then at Anthropic, and that neither company is acting responsibly — that they're racing ahead without worrying enough about safety. If you go through that thread and the follow-ups, there's talk of a greater-than-10%-chance that we're all dead within ten years. One follow-on quote put it at 50%; other people, from both OpenAI and Google, said much the same thing — that this is real.

That's why I said AI in the News really made it into the news: certainly here in the UK, this was the lead story on BBC News at Ten. Anthropic was talking about it; Donald Trump was in Ireland last week talking about this being fearmongering. The guy in the bottom right is also a Brit — Jack Clark, who's one of the founders of Anthropic — and he was talking about the idea of a kill switch, and how we slow things down. It's really been quite a week.

My view — and I obviously don't work for one of the frontier labs, so I don't know exactly what's going on inside — is that the pace of development keeps accelerating, and people are rightly a little nervous. But there's also a bit of posturing about who's further ahead, who's most nervous — "we're more nervous because we've got the more powerful model." I think there's a bit of marketing that goes into this too.

What does this mean for us as business people — what can we do about it? Both as Kowalah, as a service provider helping people on this journey, and for our own business and team, our job is to implement and use these tools responsibly. We can't control the model, but we can control how we use it and what we use it for — which is why we work, internally and with our clients, to educate people on the risks. Dennis, any thoughts to add? No worries if not.

Dennis Groenveld: It's a bit of the world upside down — depending where you live, I guess. Usually it's companies complaining that governments are trying to over-regulate them. Right now I see all these frontier labs saying they want to be regulated, and then I hear President Trump say he doesn't feel like regulating at all. It's literally the world upside down. Let's just enjoy the ride and watch the show, I'd say.

Charlie Cowan: Exactly — we can only control what we can control. At the meta level — meta as in the word, not the maker of Facebook — you've got the Western labs: OpenAI, Anthropic, Google, and so on. And you've got the Chinese labs, which are more on the open-source, open-weights side. The view is that if we stop in the West and Chinese firms don't, we're heading into a kind of Cold War or space-race situation, where everyone has to stay at the front. Then there's the question of where Europe sits in that, with Mistral or other local models — how do they play?

Definitely one to watch, and one of the reasons I wanted to put this in AI in the News: it was actually in the news. More and more, I'm sitting watching the news or seeing stuff online with my family, and the topic of work is appearing in mainstream media, and we're talking about it. I can't remember a time when CRM or HCM was a top-line story on the news, so this is definitely becoming more prevalent.

On models being released, a couple worth noting. Gemini 3.8 Live came out — a voice model — and, as always with new models, they show benchmarks proving they're better than what came before. What you're seeing here is Gemini 3.8 Live, with extended thinking, sitting above what was previously the top model, GPT Live-1, from OpenAI.

Voice is becoming a frontier. It's great that you can type, but it's even better if you're running customer support desks or sales teams where agents can communicate in real time and actually speak to people. More and more, this is going to become the way you get support. Do I want to wait 45 minutes to book a call with a doctor? Do I want to wait 45 minutes to speak to the government to sort out my tax, or to arrange a new bin for my house? I'd much rather speak to a well-trained agent immediately. So voice, I think, is going to become even more important — and beyond voice, we'll see video avatars alongside those voice models.

I think about people who work remotely — and not to get too dystopian about remote work — but anyone who says "I can do my job 100% remote, but I'm on a call, speaking to people, on video" — unless your job requires you to be physically in contact with someone at some point, you're ripe for becoming an agent via voice or video, because the things you think make you human right now — talking and video, like what we're doing here — are going to be solved fairly soon. So unless it's having lunch with someone, or running a workshop, I think these roles are going to be affected. Any thoughts, Dennis?

Dennis Groenveld: I don't think you could have phrased it any better. Your edge, as a human, over AI, is being human.

Charlie Cowan: Yeah, yeah.

The next two releases were targeted at the same user demographic. Last week OpenAI launched ChatGPT for financial services, focused on banks, investment firms, and private equity — you can see examples in spreadsheets, valuation analysis, buyer screening, and pitch-book preparation: models, skills, tools, and connectors built around this use case.

I really believe the battle isn't about the models anymore. The models are already far beyond what Lynn in HR needs, or what Frank in Finance needs — whether you're using Gemini, GPT 5.6 Sol, Opus, or Fable, they're good enough, and they're going to keep getting better. That's not your limiting factor or your differentiator. The differentiator is the harness — the application, the tools that let the model work and be powerful. That's really the front line now, and I think you'll see more and more of these frontier model vendors targeting users by use case.

So we had OpenAI launch ChatGPT for financial services, and a couple of days later Anthropic launched Claude for financial advisors — a slightly different angle, focused on financial advisors specifically, with new connectors supporting tools like Wealthbox, Vanguard, and Zocks. It comes with skills built around an advisor's day: advisor onboarding, investments, compliance, estate and tax briefs, pre-meeting prep. We've been working with financial services clients building out skills that do exactly this kind of thing, so you can see how vendors are targeting these individual personas and job roles. Today it's Claude for financial advisors; we've already seen Claude for legal, Claude for sales and Salesforce, Claude for HR teams. They're really going to go after these individual roles and industries — in a way that I think makes it very risky for other companies to build a separate platform targeting the same use case, when the model vendors themselves are going to do it.

Dennis Groenveld: If you replace "advisors" with "financial-services" in this URL, you'll see Anthropic already launched something called Claude for financial services — that's on anthropic.com/news, from around July 2025, if I remember rightly.

Charlie Cowan: Oh, okay.

Dennis Groenveld: So the "financial advisors" announcement is the role Anthropic has just announced, while "financial services," as an industry vertical, is something they already had, from about a year ago.

Charlie Cowan: If you're in the financial advisor space, there's a webinar this week — "Inside Claude for Financial Advisors" — and I'd definitely recommend joining. I've been on a few of these role-specific webinars; I was on a legal one the other day with 10,000 lawyers on it. The presenter said he'd been in law his whole career and had never seen 10,000 lawyers join a webinar before. So if you're in this space, I'd recommend joining that one.

Dennis Groenveld: It was probably the most expensive webinar.

Charlie Cowan: Exactly — everyone on the clock, billing by the minute.

Final thing in AI in the News, linking back to where we started around trust: Anthropic published a threat intelligence report — "Detecting and Countering Misuse of AI" — covering the types of risks and cases where people had tried to misuse or hack Anthropic's platforms, and how they've addressed them. You can find the Anthropic threat intelligence report at anthropic.com and download it; it goes into a huge amount of detail on individual attempts to exploit Anthropic. The industry welcomes this kind of transparency, and it shows just how wide-ranging and varied people's attempts to break into these models are. Cyber is definitely a top priority at the moment.

With that, we come towards the end of today's webinar. Dennis, thank you very much for joining — it's been wonderful, and we'll definitely do more of these.

Next week we've got an exciting spotlight webinar, welcoming a couple of the team from a company called Electric Twin. Anyone who's worked in retail or consumer products will know about traditional market research — sending people from a polling company onto the street with a clipboard to stop people who fit a certain demographic: a mother, aged 40, with two young children, walking out of Sainsbury's — "I need to interview that person. Why did you pick up those nappies? How do you think about buying nappies? What would encourage you to switch brands?" That kind of market research is slow, expensive, and prone to error, because it's skewed by the questions someone asks and who they choose to stop.

What Electric Twin does is turn this into an AI-powered process: you build synthetic databases, or synthetic audiences, and then run your questions against them. For example — from their website — "Ask parents, as an audience, which back-to-school brand earned their money this year." It's fantastic for brands wanting to run this kind of research automatically. I know you'll have loads of questions — I did — so Doug and Jack are joining us at this time next week to walk through the platform and talk all about AI-powered market research.

With that, Dennis, thank you so much — fantastic to chat.

Dennis Groenveld: Thanks for having me.

Charlie Cowan: Excellent. Until this time next week.

Dennis Groenveld: We'll see you then.

Charlie Cowan: Bye, everyone.