Copilot Cowork vs. Scout with Microsoft's Alev Tamer and Chris Wheeler

Alev Tamer, Sr. Partner Solution Architect at Microsoft, and Chris Wheeler, Senior Copilot Solution Engineer at Microsoft, discuss the latest Microsoft Copilot announcements fresh from Microsoft Build.

• Multi-model strategy: how Microsoft brings together models from OpenAI, Anthropic and Microsoft AI to offer genuine choice

• Critique and Council: using multiple models to validate and challenge each other's outputs for improved accuracy

• The DRACO (Deep Research Accuracy, Completeness, and Objectivity) benchmark and how Research with Critic achieves the highest score for deep research completeness and accuracy

• Copilot Cowork vs Scout: cloud-based fire-and-forget task execution versus local desktop agent capabilities

• Work IQ as the personalisation layer that makes Copilot genuinely relevant to your role and daily work

• Real-world use cases including automated demo pack generation, fitness apps built on Azure, NHS website creation and AI-assisted performance reviews

• Skills and plugins: creating reusable instruction sets for repetitive tasks and new integrations including Dynamics 365 and Fabric IQ

• Data residency and security: navigating Anthropic data processing, EU data boundaries and flex routing for European organisations

Thanks to Shure, this episode's sponsor, for their continued support of Empowering.Cloud

Transcript

Copy

Tom Arbuthnot: Hey, everybody. Welcome back. As we record, it is the day after Microsoft Build. We've just been changing all our prep because now we know what we can talk about, and some big announcements around Autopilots and Claws and all sorts of things. Obviously the two people that I always go to to decode this are Alev and Chris, and we'll see if they agree and get some of their experience.

I think everybody knows who you guys are in the community pretty well but just quickly, do you wanna introduce yourselves and then we'll get stuck into... Alev promised to decode the Microsoft AI strategy just before we came on, so no pressure.

Alev Tamer: Hi, everyone. My name is Alev Tamer. I am a Senior Partner Solution Architect in what we called a region SIMA. So I am responsible of my South Europe partners, and as you know, I am all about Copilot, AI, and agents at work. Chris?

Chris Wheeler: Yeah. Hi, everyone. So I'm Chris. I am a Copilot Solution Engineer aligned to UK healthcare, primarily NHS. And yeah, like Alev, really in the weeds of Copilot, trying to be a good pre-sales resource where I can.

Tom Arbuthnot: Awesome. So I'm gonna avoid jumping straight into Autopilots and Scout and OpenClaw, because we did say from the last show we did, we'd talk about multi-model and Researcher and Critique. So let's start there. Chris, maybe I can throw to you first. Like, start from the top. What is this idea of multi-model and having multiple models work at the same time?

Chris Wheeler: Yeah. So I think we're kind of shifting from this idea that Microsoft isn't just the OpenAI guys. We're not just backing the one horse. I think where we've been doing a lot of work in realizing that, you know, Microsoft's very much integrated into loads of things. We love open source.

We love partners. We love third-party solutions. And obviously when it comes to models, it's no different, and I think now that we've changed that dynamic, it's opened up a load of really great opportunities for us. So where we've kind of looked at building in more models, we give people a lot of really good choice, not just with different models, but the different types of models as well.

So you see you can still use like GPT-5.5, 5.4, 5.3, and yeah, just effectively it's just great to have different choices where we know that Anthropic's very strong in the market. We know that GPT's always been there from the very get-go, and the idea around it is that just that flexibility gives more accuracy, more creativity, more outputs into the way that we need to deliver our work.

So the idea around having... It's the first, it's a working together thing.

Tom Arbuthnot: That's really interesting to me. It's like, yes, you've got a model selector, so I can select model A or model B or model C, and that's cool. And I can see a future where we have reasoning budgets, and like actually do I wanna use the crazy high model or do I wanna use the Microsoft model or whatever.

But it's when you put them together to say, model A come up with the information and model B validate it and critique it or vice versa, that's something that's quite unique to Microsoft in the knowledge work space, because typically you're betting on one of the horses, and that would be your model stack.

Whereas here you can bring together multiple models to work in tandem.

Chris Wheeler: Yeah. And you're still in control. You know, you still select whichever one you prefer, as with all the other things, you know, with choosing the different GPT models. But ultimately the critique is really interesting because actually it's almost like an assistant checking another assistant's work as well.

And it's another assistant which has a different training set. So it's great when you have those two models actually checking each other's work, because actually some things it might say, "Well, actually this is a hallucination. This didn't pick up on it, but I have." Or, "Actually, that's pretty sound. I'd reword it slightly differently," or different things like that.

And we've seen this in coding, right?

Tom Arbuthnot: So it feels like to me like coding is kind of six months ahead of knowledge work quite often. And like coding in Claude Code, we got the plugin for Codex, and then Codex did a plugin. Like, multimodal and everything's open code. You can flip models and in Copilot CLI, you can obviously choose different models as well. So they for a long while on the dev side have got the idea of like, well, I'll use this model for this scenario and this for this scenario, and this can check my work. Yeah. Same principle coming to knowledge work.

Chris Wheeler: Yeah. And we're already seeing it in things like agent building as well. So typically I spend a lot of time in Copilot Studio, and usually when you have a certain topic, that topic may be delivered by, you know, DeepSeek or maybe like a GPT model, for example. Yeah. And obviously when you break it down, you may have a model performing better with different tasks.

It depends on obviously the subject, of course, and what department or what subject you're talking about. But yeah, I do like that flexibility. Obviously with that there comes an element of, okay, well, we need to manage that effectively security-wise, governance, you know. What does that look like?

We're still quite early in that element of, okay, well, how complex does this need to be? Do we need to give this complexity to our customers? Can we make it easy for them to understand? And obviously, yeah, when you have something like Copilot, we do make it easy. When you've got more agent options, there's a lot more, obviously with low-code, pro code, options there.

So yeah, I do like that.

Tom Arbuthnot: Nice.

Alev Tamer: And also there's a new one -- have you heard about DRACO Benchmark? It's all about deep research, completeness, accuracy, and objectivity. So the DRACO Benchmark, the Research Critic has the highest point comparing to other deep research. For example, when you use the OpenAI's deep research, Mini-03, you can see that the benchmark is like 31.7, and within the research with Critic it's 57.4%.

So it's the highest comparing to others.

Tom Arbuthnot: Interesting. Do you think part of that is because they're able to use both models to get the outcome?

Alev Tamer: It's more accurate, and like Chris said, it actually uses like two different models at the same time, and Microsoft is the only company which can provide these kind of capabilities.

Tom Arbuthnot: Nice. So, do we go next to Cowork or do we go to, what we were talking about as ClawPilot but now talking about as Autopilots and Scout? Where do we go next?

Alev Tamer: So I do use ClawPilot more than Cowork now.

Tom Arbuthnot: Okay. Well, let's -- I'll let you choose. Maybe we should start with defining how the two are different because Cowork is essentially similar to Anthropic's Cowork, but running in Microsoft Cloud in a container, so it's running in the cloud, and it's a multi-model.

At the moment it's using Anthropic models, and it's running in the cloud connecting to Graph, doing things for you. It can generate PowerPoints, it can generate files. So how is the kind of AutoPilots and Scout and OpenClaw different?

Alev Tamer: So if you think about Copilot Cowork, it only uses Anthropic's Claude model. And as Microsoft, we promise to use the latest and the greatest models coming from different companies. In ClawPilot, we can select different models coming from different companies like both OpenAI as well as Anthropic and Microsoft. We are actually including some of the Microsoft AI models for flash coding or code flash in ClawPilot.

So you use a model to work with. In the same session or the same chat, you'll need to just keep using the same model. If you have a different job to do, you can just create a new chat and then use different models for different work. So as Microsoft, the AI strategy is actually providing an AI platform which is intelligent, trusted, and scalable.

And we all know that Microsoft is actually using AI for the last 30 years, and we are talking about a 50 years of technology company. So it's literally AI built on top of work.

Tom Arbuthnot: Yeah, sorry. That's an interesting kind of distinction. So ClawPilot, you can do multi-model, and ClawPilot is essentially built on OpenClaw, but then you've done some things internally to kind of bulk it and secure it and make it more Microsoft aligned.

So is it essentially consuming the models from Foundry? Is that how it's working?

Alev Tamer: Not as much as Foundry. Foundry has thousands of models, but we can provide the main models from OpenAI as well as Anthropic's and some Microsoft models. But the magic in the background is Work IQ. It's personalized. It knows me better than sometimes maybe Chris.

Tom Arbuthnot: I don't know. I feel like that's a challenge. That's another pub quiz we need to build.

Chris Wheeler: Yeah. Maybe one for the next podcast. Yeah, let's do that.

Alev Tamer: So I used to -- my performance is based on my partner's performance, and 60% of my performance... because we went through performance review when I used ClawPilot to produce my performance review. I need to go and search for my partner's market offers. And I don't know if you had or experienced this, but searching something in Marketplace is sometimes impossible. So what I did, I was like, "This is a marketplace," and it can open the links because it uses your credentials and you need to create a GitHub.

Because it knows about my job, I just ask it to, "Please, this is my partner. Let's go through the marketplace and create me a deck with all of my partner's offers and make it align with Frontier partner or Frontier firm concept." My manager -- And is it at that point...

Tom Arbuthnot: At that point, is Copilot like computer use driving the browser or is it running a browser in the background?

Alev Tamer: It opens a browser. Yeah. And then it starts running the job. So Cowork is not able to do it at the moment. Yeah. But I don't know if anyone loves just preparing for a performance interview, but I used to either just procrastinate about it and spend hours and hours of work.

It was done in 20 minutes. Now my manager wants me to show this to my team because he was amazed by the results. The PowerPoint presentation, I'm like, "Oh my God, this is the thing. This is what Copilot meant to be."

Tom Arbuthnot: What model did you pick when you did that, out of interest?

Alev Tamer: I did use Claude models.

Tom Arbuthnot: Interesting. And Chris, you were saying, because you got quite big into Cowork, didn't you? You did quite a few things with it. Yeah. What were the things that you've picked Cowork for, and what were the things that you've picked... so what have you picked for what?

Chris Wheeler: I'm still a big fan of fire and forget. Not fire and forget, sorry, just fire and just leave for a little bit and come back to it.

Yeah. I think where you give Cowork a certain task like... I'll give you a good example. Just recently we've got a NHS event coming up next week called Confed, and I'm running the M365 Copilot stand, and what I don't have enough of is good demo packs. So what I do is I give Cowork the task of building me a lot of dummy data, so Excel, PowerPoint, Word documents.

I ask it to give me the full-blown demo pack with all the prompts, and I then give it, I think, oh, I gave it like 18 different personas, like CIO, CDIO, CEO, CFO, all of those. Yeah. And I wanna say, "Look, build me out all of these demo packs." And literally just, it obviously takes a good couple of hours for it to go through every single one.

You see obviously all the outputs and all the folders being created. But literally something like that, I just absolutely love. The same thing is when I start building like GitHub repository packages as well. So I've been doing a lot of work with... Ha, it's quite funny. I'm a big fan of Azure Static Web Apps.

And I've used it for personal stuff. So obviously I've got my diet and my fitness app that runs off of Azure, which is a hobby, it's free, it's good. And every week I just use Cowork to update that based on my scales, based on my preferences.

Alev Tamer: You should show that.

Chris Wheeler: Yeah. Yeah, if you want to try it, I'll be there. I mean, if people want to, it's public. You just go onto www.christrainingplan.com, and you could see, literally you could see all my recipes, all my training. At the moment I've got a knee injury, so it's giving me a bit more of a relaxed training plan.

But no, the recipes -- You just, yeah, you're just caveating it.

Tom Arbuthnot: When people look at it, Chris, they'll be like, "You should be working a bit harder." Please just... You know, it does usually have gym sessions and cycling and all this sort of stuff.

Chris Wheeler: Unfortunately I did a 10K last weekend and I've hurt my knee, so it's told me to take it easy. So, but anyway, the recipes are really good. Like, it's so simple. It's perfect. And my wife can access the app just in case I'm away and she still wants to cook some food. It's just brilliant.

I also use it for Monday Night Football. I have like an Excel spreadsheet with all of the players and a couple of stats, which I've just given like what's their shooting, what's their defending, and it creates a team sheet, and it produces that on a static web app for me, obviously linked to GitHub.

So anyway, personal stuff, brilliant. Work stuff, even better. So I've basically created a fake website for one of my NHS customers, and I've just built a Copilot Studio agent that's basically just been -- It's just used the website as a knowledge source. But what you do is you embed that into the website and you show it working.

So something like that. I just asked Cowork, "Hey, by the way, look at Torbay Hospital's website. Recreate it." Obviously, it doesn't copy it exactly because it's copyright infringement there, but it just does a really good job building a really nice looking NHS website with the embedded code.

I don't have to worry about coding it anymore. It gives me all the GitHub repositories, all the images, everything, all the readme documents, all that type of stuff, and it's just saves me so much time, and I really do like it. But no, the main one is the multiple documents, the demo packs. It would've taken me absolutely ages to try and do that before.

And now -- And you're firing that off to a cloud container as well.

Tom Arbuthnot: And I'm firing it off. So just to help people understand, it's like, off you go, off my machine. You can close your laptop. Leave, yeah. Yeah, exactly. Yeah. You hold onto that when I come back maybe tomorrow, maybe when I remember, you know.

Chris Wheeler: And this is the reason why there's that joke going around with the Claw Pilot and the OpenClaw stuff, where somebody selling those little plastic arm things to just hold the laptop up just a little bit because if you close your laptop, it just doesn't work.

Yeah, yeah. They're like coding on the beach. Yeah. And obviously with Cowork, you just fire it off, it's in the cloud, shut my laptop, go off, come back, it's all done. So there's some slight differences. Obviously there's a benefit of having cloud rather than local. There's some benefits of having local than cloud, so yeah.

But I mean, in terms of my work, it's just great. I mean, we've got some of our colleagues, haven't we, Alev, that just love building HTML dashboards. We've got loads where we basically get like a really good account plan every week. It builds a HTML file for us.

We've got one of my other colleagues, Stuart Nichols, he's just made a really good sort of M365 licensing and Copilot strategy dashboard that he then shares with customers. And what they do is this basically collects all the information and gives you a foolproof plan on how to successfully deploy Copilot, and it goes through things like security, governance, licensing, adoption, change management, all that type of stuff.

He's obviously a customer success person, so obviously he's really into the whole post-sales and really optimizing the experience there. But there's some really cool things happening with Cowork. Dashboards is definitely one of them. And it's available on mobile now. Yeah, available on mobile, yeah.

Alev Tamer: The other thing is skills. Now we are actually talking about agent skills. For example, I actually created a skill for me to create the demo. So what I'm basically doing, I just do like forward slash, and I call the skills to create the demo. It just asks questions about what kind of industry, which customers do you have specific? Is it like a medium, big, or enterprise? And then it asks about what kind of roles. Is it the business decision makers or IT? And then it just creates things, and then I actually ask it to create a HTML dashboard so I can actually prepare for my... And it has like timers, everything, the prompts.

Chris Wheeler: Just to add to that, I think the skills thing, for those that don't know what that is, it's effectively like a set of instructions again. So it's basically like a file that's embedded. It has like the very similar instructions that you have with agents. Say, "Look, this is what I like doing, this is how I like it done." And obviously it's really great for repetitive tasks. So like if Alev keeps building a HTML dashboard or like demo packs or presentations, it knows what to do because of that skill.

Whereas obviously an agent obviously has that in the instructions, but we're not... Yeah, so they're very, very similar in that regard.

Tom Arbuthnot: Yeah. They sound more complicated because the industry has been kind of flitting around skills and plugins and connectors and like it's nice to see Microsoft and most industry have standardized on skills now, which essentially is really just markdown files, like with sets of instructions.

Like SharePoint have added them, obviously Cowork's added them, Copilot as well, and Claw will obviously, whatever variation of the Claw you use would use those as well. So it's nice that you can use those skills in different ways, though those tools work differently. So the ability to do things are slightly different when you're on a desktop running an agent versus running Cowork in the cloud, as an example.

Alev Tamer: One of the things I do use for my job, because it's nearly impossible just to grasp all of the changes and announcements coming from Microsoft -- especially in the build time. I just give it specific emails that will receive some updates, because we have too many different notifications, newsletters, emails, and I make it to create a filterable dashboard for what is upcoming, what's internal, what's under NDA, and it actually creates and updates it every 5:00 AM in the morning.

That's the automated part. So this is, we talked about in the previous podcast, but this is how we can become a frontier firm. Human-led, agent-operated. Because it asks me, "Would you like me to do this?" I can either just ask it to do it once or do it all the time. Be careful with do it all the time. Some of my friends ended up with missing files and emails.

Tom Arbuthnot: So where are you both landing on agent on the desktop versus Cowork? Or am I mischaracterizing that as a versity? Do you think in reality there might be different tools for different use cases?

Chris Wheeler: Yeah, I'd agree with that. I also think that there are certain considerations, like obviously when you talk about more local based AI, you know, with runtime for example, I've got no idea how a system admin or even like a desktop admin would react to that sort of thing to say, "Hey, I'm running this."

Tom Arbuthnot: A lot of the Build keynote was around how Microsoft are gonna use execution containers and layers of policies. Like, there's gonna be a big conversation in enterprise for sure about this agent is impersonating Alev and has everything Alev... like, how do I log it? How do I trust it?

And then, yeah, how does Agent 365 understand what it's doing?

Chris Wheeler: The accountability piece as well, like Alev scheduled this PowerShell script. Was it Alev or was it her agent? Yes. Like those sort of things. So there needs to be some sort of consideration there. I think we'll probably spend a bit of time trying to explain that it's still really safe.

Obviously Microsoft's all about trust and security. The other thing as well is I think with things like Cowork as well, we're still kind of a little bit behind just because obviously the way that Anthropic works, and we're just kind of spending a lot of time explaining where Anthropic sits in our architecture as well.

And then obviously I'm still... I mean, in my role, especially in healthcare, we're still kind of learning to walk. Like, the dream is to have, you know, human-led agent orchestrated, like, down the line. I think we're still getting into the situation where we're still looking at, right, okay, who's gonna build this agent?

Who's gonna use it? And even better, who needs to manage it? Obviously we're still having those Agent 365 conversations. We're still building really good agents. Like, we've kind of moved from knowledge retrieval agents. We're now getting AI doers now, so we're looking at things like tools and all that sort of stuff.

And then we're broaching on automation. There are like a few hurdles to jump through to get there. But like, we're still in that realization stage, and obviously with that comes the ROI as well. NHS is going through a very sensitive time right now, so it's kind of being delicate with that.

You know, AI's not there to replace people, it's there to help, even though there are certain things happening within the NHS which is unavoidable. So yeah, we're kind of not there. That is the dream. I'd love to start talking about that. I think realistically we're still kind of, you know, turning on Agent Builder, getting people used to Copilot Studio, integrated with Azure as well, and all these sort of things.

Tom Arbuthnot: It's just a lot going on, isn't there? Because as we talk, it's interesting because the Autopilots and the Cowork are largely, as they stand, extra tools to do stuff for me, the things. Whereas the agent, Copilot Studio and Agent Builder type approach is more like I could build a agent for my team, for my workflow that stands independent of me.

Yeah. And I imagine those lines will blur as we get to share Autopilots and things, but it's...

Chris Wheeler: Yeah. And not just for yourself. I think the main focus of Microsoft is to try and scale out as successfully as possible. So the idea being is that even though things like Claw Pilot, Open Claw, Cowork for example is really good for the smaller numbers.

Obviously M365 is gonna be more broader. Copilot Chat obviously for everyone. And Copilot Studio, building agents that could just be standardized, templated. So there are some things happening with NHS England that's gonna do that for all the 1.8 million people on the tenant, right? And for us it'll be really good.

I mean, from a consumption perspective. And then obviously the more niche specialized instances where you do need local runtime, where you do need more advanced specialties, that will be really pinpoint key areas where we will see really good outcomes there. So yeah, I mean, from a sales perspective, it's kind of teetering more along the more broader demographic.

But you know, we still like to see these really cool tools happening. Like you may have seen that new Windows device that they were launching as well, so yeah, there's a load of cool things happening. I'm really keen to see what happens in the future when that becomes GA.

Tom Arbuthnot: Yeah, I guess then it's interesting where we'll get with AutoPilots because the part of the conversation was AutoPilots could have their own Entra identity, so suddenly actually am I building team members? Yeah. Do I build them in the... I was about to say the old way. Like Copilot Studio and Foundry or in Claw-type model. Like, I think it's just exciting, but there's not a clear one way to do things. It's like here's a menu of options, and it's nice to see Microsoft really keeping up with the rest of the industry in like, okay, here's an option.

We'll look at this, experiment with this. Here's an option. Like we were talking before we recorded on the pod, like, it's a bit of a challenge keeping up with everything that's going on. But I'd almost rather be in that situation than looking over the fence and being like, "Oh, we don't have a client-side version," or, "We don't have a Cowork," you know?

So it's nice to have the options, and Microsoft, I think, are being very pragmatic with, "Oh, Anthropic are doing something really good over here. We should integrate that model." We talked about new models that can... the Mistral got announced on Foundry. There's more stuff coming there. Actually, maybe that's a good topic to get into because a lot of this stuff at the moment depends, some of it depends exclusively on Anthropic.

Some of it depends on you having the Anthropic models enabled. It all depends on Frontier. How do we help decode that for people?

Chris Wheeler: So the Anthropic stuff's interesting because Cowork will eventually have GPT 5.5 or whatever the latest version will be when it comes out. So yeah, we are -- But that will unblock it for people that are blocked.

Tom Arbuthnot: Or I don't want my data outside Europe at the moment. For anyone that doesn't want Anthropic because of data processing or politics, whatever reason they've got towards Anthropic...

Chris Wheeler: We don't just wanna say, "Well, it's take it or leave it," you know? Yeah. It's not like a be all, end all.

But yeah, we wanna give that variety, but obviously just be aware that if you do limit it, you will limit your experiences, so you will not have all the benefits of Researcher with Critique or Counsel and all the other stuff. So but yeah, I think in terms of just dissecting that and really reassuring customers, there's just a few things to consider because ultimately Anthropic is still US-based.

That hopefully should be changing fairly soon. The idea being is that other M365 services already follows that same processing mechanism. If you look up the data protection addendum, you'll notice that there's a segment that says other M365 services will process outside of country. So there seems to be just a bit more of an awareness of that.

So like, when you talk about things like spelling and grammar, PowerPoint designer, Excel advanced analytics, all that type of stuff, that processing already follows the same processing like what we do with Anthropic. So I think it's just a reassurance that, as you've seen, nothing bad has happened from having all these additional services. It still provides a really great... there hasn't been any snooping or interception of any data or anything like that at all because it's built around our secure service boundaries. I think people have got lost.

Tom Arbuthnot: As well, because the first rev was it's not on Microsoft and we're pushing it to Azure. The second rev was it's on Microsoft, but now it's in the US, and hopefully the... probably not the third, there's probably more variances, but the third will be now we can process on Microsoft in Europe as well. And there's just been a lot going on to help people understand that.

Chris Wheeler: No, Europe's an interesting topic because obviously you've got the EU data boundary. The UK's not part of that, so it's a very interesting conversation where even though there'll be an EU presence, doesn't mean the data will be ring-fenced there. So there just needs to be that acceptance that data could go to different places, but it is all completely secure.

Tom Arbuthnot: And processing versus at rest as well is an interesting one, isn't it?

Chris Wheeler: Yeah. No, at rest is absolutely fine. But it's just the processing that causes a few concerns. And understandably, you've got your own policies where it says NHS may say, "Look, you know, our patient data, we try to keep it within the EU."

So it's just kind of reassuring them that even though it may go off to other places just for processing, it's all volatile, it's all stateless.

Tom Arbuthnot: Yeah, so it's not... It's just in memory. It's not committed to disk or whatever it may be.

Alev Tamer: For the European partners, for example, they are okay with using Anthropic because they really get the benefits out of it. And this is why actually we just provide them some kind of flex routing. So if they would like to not just use it in the EU, and if they're okay with the US, they can actually use flex routing.

Tom Arbuthnot: Yeah, to burst out when there's not enough compute. Alev, I think you wanted to show us some stuff. Do you wanna do a bit of a demo?

Alev Tamer: Of course. Share my screen. This is ClawPilot, and most probably it will change its name as Scout. So the main difference is you'll be able to select models from different companies, including Microsoft AI.

Whenever I need computer use, I just use ClawPilot. And Claude Opus 4.8 is now available because, as we promised, we'll make the latest and the greatest model available from different companies. So one of the things that we worked with, actually Chris, is, so we have our roles and definitions. So if I do it live, by the way, it will take some more time.

But whenever you try to create your prompt and use the prompt coach, because if you tell it, like, "I would like to do this with ClawPilot, please create me a prompt," it will create you a really good prompt. So what I would like to see is, I want it to go through my emails, my communications, and tell me what I can actually, what is aligned with my role, what kind of improvements I can get, and what kind of gaps.

And it just gives me really, really good readable results. Because I would like it to create like 1,500 words, I can just ask it more. This is, I can share with my manager. It's very easy, and I can just ask it to run this by the end of every Friday.

Chris Wheeler: It's an interesting question, Tom, because I think there's a lot of people at Microsoft that always question, "Am I doing the right thing here? Am I on the right tracks? Am I doing a good job?" I keep asking myself that. God, it's giving me anxiety sometimes. But it's just great to have this to say, "Look, we have access to your roles, your expectations." It has access to Work IQ as well, so I do like it. And then like, that thing that Alev said about getting your manager involved, it's really great to align that because the support from the manager to then...

So I use this a little bit differently. Obviously, I try and say, "Look, you know, what do I need to do to get to the next level to get that promotion?" Or "How do I get into a manager role?" Or obviously that's there in our role agreement.

Tom Arbuthnot: Basically, is the Work IQ that's really, like, as Alev said at the start of the pod, the secret sauce here? Because theoretically, you could go to another AI product and say here's my role description, what should I be doing? Yeah. But the next step here is what have I done that's appropriate? What haven't I done? Yeah. Because it has your meetings, it has your contacts, it has your emails.

Chris Wheeler: Yeah. I mean, you can still use like another tool, and you can feed it in like a meeting transcript, but it's very siloed. It's very singular. I think with WorkIQ it just knows everything that you've done, the people, it's got identity, data, emails, all that type of stuff. Yeah. So I think in terms of just asking it what you've done recently, it knows exactly what you've done recently. And obviously with the semantic indexing and everything around that as well, like the broader landscape of finding that information is just yeah, it just improves it tenfold.

Alev Tamer: It's just very nice. With Copilot Cowork, I love creating dashboards, like Chris said. So I can just create a dashboard for a customer that I would like to learn more about.

Chris Wheeler: Like Pets at Home.

Alev Tamer: Yes, it's nice. So I can just ask the customer name, the objectives. This is created by Prompt Coach anyways, so I can give the customer name, Pets at Home, they are kind.

Workplace file and last seven days, all categories, I can just ask it to search everything because it's been trained with billions of data as well, and it has access to Microsoft and My Work data. Look at this dashboard. I can see if it's business supplier, supplier chain. I can filter by geography.

Tom Arbuthnot: Yeah. It's interesting this idea that like one of one UIs or interfaces, like for my specific use case, the way I like to work, like I'm very bad with big blocks of text, for example. Like, I can't really process them. So I'm always like, I want bullets, I want options, I want minimal... ask me one question at a time, don't give me 10 questions.

Things like that. Like really a customized experience to how you want to work.

Alev Tamer: So this is one other thing. If you would like to be in control, you'll have the selections. I can either activate it or I can activate and make it run every day. So whenever it needs a permission from you, for example, dealing with your OneDrive files, dealing with your emails, it will ask you, "Would you like me to ask you or would you like me to do it automatically?"

So this is where the human-led agent operated approach comes up.

Tom Arbuthnot: And Copilot -- And this is where I'm excited about Cowork because it's running in the container in the cloud. Like, I think a lot of the hype is around the Claw model at the moment, but actually being able to fire it off on my mobile and it go away and do the work and come back to it, and actually I can work on multiple machines and it can all sync out because it's in the cloud, which I know is kind of a work in progress.

Like, I think that is really powerful. But I'm still curious as these two stories evolve, where we end up with like, what can Claw do or also models do that Cowork can't or vice versa? Because they're potentially quite similar.

Alev Tamer: Yes. You just need to try and see which one is exactly helping you a lot more with a specific job.

So this is what I do with ClawPilot versus Cowork. So this was one of the example trying to prepare for a podcast with Tom and Chris. So it gives like punchy, energetic, calm, credible, witty, and it selected Chris for witty. I don't know why. And you can get ready for your meetings, your conversations, everything.

And rehearsal dashboard. So I do this with my demo with our partners or customers. So I can see, Tom, you are the punchy one. I don't know why I am the calm one. I am the calm one.

Chris Wheeler: I am the witty one.

Alev Tamer: Yes, Chris is the witty one.

I don't know if you heard, but there is a book called "Open to Work." It was LinkedIn CEO and Chief Economical Opportunities Officer. They write this book, and it's all about how AI is going to change our work. I would definitely recommend to read it because it will change our work.

Tom Arbuthnot: Amazing. Well, thanks so much both of you for decoding some of that. I think a lot of this stuff is hot off the press as we record, so it's great to get opinions on it.

And yeah, we'll see. You can now... If you're in Frontier, you can get Cowork now, provided you've got the Anthropic models enabled, and the Scouts, which is the first inflection of Autopilot, which is essentially the branded variant of what the kind of Copilot Claw or is going to be for customers is available in Frontier as well.

But you have to jump through a few hoops of some additional security configuration, if I got it right, to make sure that you're all aligned as you should be.

Awesome. Well, I was about to say, I don't know what we cover next, but these things move so fast, we'll probably be back again in at least a couple of months. Chris, hope your NHS thing goes well. That sounds like a really good solution.

Chris Wheeler: Yeah. Absolutely. No, we've got enough demos to cover for any persona.

So yeah, we've got some cool agents to show as well. That doesn't just stop. There's been loads of cool things. Been vibing quite a lot with Power Apps. Yeah. Absolutely love that. Absolutely love vibing.

Alev Tamer: Nice. And also there are some things, some action coming to Outlook and some other applications that is just makes your life easier.

Tom Arbuthnot: Okay. Well, that's our next topic then, sounds like. Good. Awesome. Well, Chris, Alev, thanks so much for joining us again. And yeah, if anybody's got any questions, put them in the comments below, and we'll see you next time. Thanks a lot.

Chris Wheeler: Sounds good. Cheers.

Alev Tamer: Bye.