Brett Johnson: Don't forget that there is actually the phone versions as well, and we've got the... So you can actually have conversations with it as well, in much the same way that we're talking, which is phenomenal.

Tom Arbuthnot: Yeah, that is huge. I do that a lot on the desktop as well, the Windows H, just chat away. And honestly, that can help your prompting a lot because when you can ramble on a bit and be like, "I'm thinking this, I might do that," and it will Then organize your thoughts and say, "Okay, based on this, here's what I would do" This is.

Brett Johnson: What I'm doing all the time. My boss knows this. I go off, take the dog for a walk. I offload my thoughts into a transcription. Copilot then makes sense of it, and sometimes it's actually creating the Cowork demos that I'm doing for customers, right? I say, "Right, I'm about to meet this customer. These are the people I'm gonna meet. Let's have a conversation and figure out what would be the best demo." And then it will write the prompt that it, what, that I should be putting into Cowork. It's fabulous.

Tom Arbuthnot: Hey, and welcome back. This week we have Brett Johnson, a great friend and second time on the show, and basically I peppered him with questions to decode for us what is a Frontier Firm, what's going on with the Anthropic models, Work IQ, Cowork, and also helping us explain Agent 365, which just went GA. Many thanks to Brett for, putting up with me asking him questions and putting him on the spot. He's always a great conversation and, really great insights from him. And also many thanks to Luware, who are the sponsor of this podcast. Really appreciate.

Brett Johnson: All their support of the community. On with the show. Hey.

Tom Arbuthnot: Everybody. Welcome back to the show. Excited to have Brett on again. We had an awesome conversation in person. Must A month, more than a month ago now at the AI tour, right?... The AI in the workplace. Yeah. And we were like, "Can't talk about this yet. Can't talk about that yet." But now we can. So, Brett, welcome back. For those that haven't listened to the last one, can you give us a little bit about, yourself and your role?

Brett Johnson: Yeah. Thanks, Tom, and I appreciate being invited back again. So, Brett Johnson, Microsoft UK. My official title is a Copilot Global Black Belt. So what that means in English is that I support our field teams on anything and everything to do with Microsoft 365 Copilot, which of course morphs into Agents and the Venn diagram of Microsoft 365, and Azure, and Copilot Studio, and all of that stuff. But I also have got tentacles back into Microsoft Corp Engineering, Technical Program Management, so there's a two-way street. So yeah, a, pretty strategic.

Tom Arbuthnot: Role. It's fun. Yeah, a long history in our space as well, that we won't go into this time, but We've known each.

Brett Johnson: A long time, Tom.

Tom Arbuthnot: Yeah. It's been a while. It's been a while. It's interesting you say that actually about your role now because it's really brought the worlds together. So obviously the kind of Dynamics part of the business and the M365 part of the business have come together. Right. Yeah. And you've got Copilot Studio, and you've got that, but then AI Foundry and the Azure side as well. So AI is really kind of interlinked. We used to talk back in the day about the three Clouds. It really is a bit of everything these days.

Brett Johnson: Yeah, and you know what? Actually, when I'm having conversation with customers about, perhaps sovereignty, we'll get onto some of that later no doubt, it's really important in order to make sure that we're very clear on what actually are we talking about. Is it the Microsoft 365 Cloud, which of course by default includes the Copilot capabilities, but as and when it connects into Azure, which is ultimately where the AI models are, we need to, we need to be cognizant of that. And so what is true in one Cloud perhaps is not true for another, so we just need to make sure that we understand all of that.

Tom Arbuthnot: Yeah. I'm excited to decode that, and we're definitely gonna talk sovereignty. Yeah, we are. But first thing, I wanna get into Frontier. Like, Frontier firms as a concept is one thing, and then the Frontier program on M365 and getting early access things is a different thing, and we see lots of talking of both. Can you decode that for us?

Brett Johnson: Yeah. No, I will. And I will have a question back actually about what, from a partner point of view and your role within the MVP status, what Frontier means, because I don't actually know what that would look like on your side of the fence. I don't know if it... But I'll answer the question first, and then I'Yeah I'll put it back to you. So look, the Frontier Program is something that's spun up probably about six months, maybe even a bit longer now. Things are happening at a rapid pace, everything... Just losing track of time, frankly. But I guess think of the Frontier Program as a preview program of the latest innovation that's coming from the AI world. Whether that's something that Microsoft has hooked up as a result of the capabilities in OpenAI's, LLMs or Anthropic's or whatever, and more on that in due course, of course. But so within the Microsoft tenant, there is the ability to turn on, and subsequently for you, the organization, to become a Frontier firm. And essentially what that means is that the capabilities across Copilot that are the new innovative things will be available for you, not necessarily for everybody. You can choose and select who gets this. But those innova- those new innovations will land in the tenant, giving you an experience for you to feed back and, suggest where things are going Mm and where they're not, and basically get an early preview. But the difference here is with.

Tom Arbuthnot: But kind of like a, like a like what we would've called a preview.

Brett Johnson: Program before. But there is also a Preview Program. There's the public preview of stuff, and the, and the, and the fundamental difference is that the public preview is stuff that is gonna come and land, right? But there may be, and I've seen it a couple of times, not very often, where the Frontier program, the capabilities that have been tested augment into something else, right? So it. But I encourage you to do that. Schedule.

Tom Arbuthnot: Schedule tasks on Copilot, for example, went back and forward on Frontier. It's Yeah, that is true... Are we gonna go with this model or are we gonna go with that model?

Brett Johnson: But there's also, and again, depending on the audience, there's actually a Frontier program outside of just the technical capabilities of features and product, bells and whistles, but there is a program where, individuals can actually be part of a wider community, and again, depending on the customer, frankly, whether or not there's opportunities to speak to the product engineering and the product marketing teams as well. So it's a, it's a pretty broad program, but good. And what about your exposure to it, Tom? What do you get as, well, from your side of the fence?

Tom Arbuthnot: Yeah. So with MVP, we've got access to some extra access in terms of Frontier. My tenant's in Frontier so all our users are in Frontier. We opt in like any other organization as MVPs to go in Frontier. Yeah. But then we fortunately have the little bit of MVP access, so which is not necessarily early feature stuff above Frontier, but a bit more explaining about what's going on and why.

Brett Johnson: So we probably have a similar experience, right, I imagine. You know?

Tom Arbuthnot: Yeah. I back and forth, but I think the Microsoft, the internal people are slightly ahead of us, which is always fun to hear what's coming and then see what we've got versus what you've got. Yeah, true. True. But, yeah, I mean, the funny thing is it's all moving so fast, as you said, that, like, it... Keeping up is a real challenge and getting time to experiment with everything.

Brett Johnson: Yeah. Yeah. That is true.

Tom Arbuthnot: So, like, Frontier, that explains it nicely and then, like, Frontier non-technically, so you'll see this in the Microsoft blogs, like they talk about Frontier firm in the concept of making use of these capabilities. So to be a forward-looking AI firm, you don't necessarily have to be in the Frontier program. That's true. It's just kind of a mix of terminology. Yeah. But many of them are because they want to be getting early access to the latest stuff and see what's going on.

Brett Johnson: It's being at the tip of the spear, right? It's having those capabilities at work, and you being able to utilize perhaps what you're experiencing in your personal lives and the AI experiences that you have from outside of your Microsoft work experience, being able to have that experience inside of work in that secure governed environment.

Tom Arbuthnot: Right? Yeah. And I would say it's a massive benefit. Things like Cowork now, which have come to Frontier, like that is a... We'll talk about that in a bit. Yeah. And that's a huge step forward in functionality, and you want at least I think some people in your organization understanding that before it goes to the masses of your organization, and Frontier gives you that really early access of, "Oh, how can we start thinking about this? How can we start thinking about adoption and training?" That kind of thing.

Brett Johnson: It helps you get your head around it, right? Because, we announced, and I know we'll come onto this as well, so these are little snippets for later in the conversation, is Agent 365. So it's the 1st of May today. We GA'd Agent 365 and E7, right? And we found out... Well, you found out about that a little while ago, but under the covers, it was there, ready to go, essentially, but we needed to make sure that you customers are at least educated about what it is in order for you to be ready for when it lands, for you to then make decisions as to what it means to you, and whether or not it's something you wanna turn on.

Tom Arbuthnot: Yeah. And it's nice to see Microsoft, taking so much feedback, and frankly, the whole world is moving so fast, this space at the moment, that, like, getting it into customers' hands and seeing what lands and what doesn't and what should be different, it's gonna be an agile world moving forward in this space, I think.

Brett Johnson: Even more so than it is today.

Tom Arbuthnot: Yeah. Right. So, we've nailed Frontier, I think. Let's talk Anthropic because-... Satya has said a few times, "We're a multi-model company. We wanna provide the best and the options to our customers." And that was said for a long time, but for the longest time in Copilot world, really we were tied to the OpenAI models for most use cases. That's changed quite radically in the last few months.

Brett Johnson: It has. It has. So, the capabilities that Anthropic has enabled have really brought to life additional capabilities across the Microsoft stack, right? So the OpenAI models are within the service boundary, or bound by enterprise data protection, the security, the governance, all of the security controls and labeling, all of that stuff, which is fantastic. I... And they're great at what they do. But what we're beginning to see is, I'm sure everybody is aware, is that there are multiple LLM providers, and they're kind of leapfrogging one another with capability. Now, if we, if we rewind a couple of years, it was all OpenAI, GPT model, this, that, and the other. And then Anthropic released Claude, and then the capabilities there came, and everybody got quite excited about vibe coding. We should have a conversation about that as well. And, the models, the models are different, and they there are some areas where they really exceed, in other areas where they somewhat outperform the other. And the Anthropic.

Tom Arbuthnot: Yeah, or tool use is what Anthropic seem to nail with their model and their harness. And then, and that maps so well to our world of graph and APIs and application integration. As you say then, GPT 5.5 dropped just recently as we record, It did Sam's already rumoring 6, and it's like, it like... And it's, it really is a this one, then this one. I guess it's nice in the Microsoft world to have the option to not be tied to just one horse.

Brett Johnson: Well, and we've got... So we've ultimately got multiple harnesses now, right? So we have the harness in for OpenAI and for Anthropic, and then depending on the use case and what's required, then the models will drive the appropriate behavior. And so for those of you that have experienced Cowork, that's fundamentally being driven by the Anthropic models, it's astonishing. Connecting that into the work item Yeah.

Tom Arbuthnot: It's a massive, it's a massive step forward in terms of functionality. Well, A lot of the things that I felt like we wanted in, M365 Copilot, like just send an email for me, just book a meeting for me, are things that I think Yeah particularly in Agentic world are expectations of what an Agent should be able to do. It's brought it to life. Cowork can do that stuff.

Brett Johnson: Yeah. It's brought it to life. But it's not just Cowork, right? So you now have the option. Again, it depends on where you're at from a customer point of view. Have you turned on Anthropic? Because there are controls there. And I think it was last week, actually, so it was the last week of April, we enabled the capability for you to turn on Anthropic, but also limit it to a subset of users, right? Which is welcomed, and hopefully, that will be the experience moving.

Tom Arbuthnot: Forward. So talk us through, Brett, why, 'cause this is a tricky conversation. Like, OpenAI models all run in Azure. There was nothing to turn on or off really. Like, it was a.

Brett Johnson: And when we say in Azure, what we mean is within the service boundary, right? Within the Microsoft. First party stuff.

Tom Arbuthnot: Yeah. Keep me honest with the terminology. Yeah, that's fine. Keep going.

Brett Johnson: That's.

Tom Arbuthnot: A good point. First party, yeah. Yeah. So whereas Anthropic is different. So explain how it's different.

Brett Johnson: Yeah, currently different. So look, the... Just, when was it? Probably around October time last year, there was the announcement that we were gonna Involve more models in Microsoft 365 Copilot, which was all very exciting. And so we enabled the control that would connect to Anthropic. Now, that actually is an external connection, so it's not running within Microsoft's service boundary. It's obviously fully secure. It's a stateless operation. It... But it was running and to a degree right now, running within Anthropic services in the US.

Tom Arbuthnot: Yeah. So Microsoft went to Anthropic, checked them out, checked the data was being processed correctly and said, "We trust you guys to process, but now we have to ask our customers on a case by case, do they also-".

Brett Johnson: Yeah. So "... Want to trust you?" That was, that was part one, and then on January I think it was the 6th, might have been the 7th, Anthropic were became an official sub-processor, right? So essentially all the, all of the DPIAs, all of the security infrastructure, the legal agreements of how it all works, and Microsoft making sure that the terms and conditions of how technology is processed is adhered to essentially. So there was, legal documentation. They became an official sub-processor. And look Anthropic are many sub-processors or one of many sub-processors that actually make Microsoft 365 Copilot work. So this is not unique. And then today, currently, the Anthropic services are in the US, and then there is a big ask for the Anthropic services to be, available within the EU. And that is coming. We, the plan currently is for it to be within this calendar year. I anticipate that we'll see it sooner rather than later because of the, because of the amount of asks that are coming- Yeah. It's a big ask... From.

Tom Arbuthnot: Customers in order for that. So obviously Brett and I are both in the UK, lots of exposure to global customers obviously, but lots Yeah Europe and lots of, lots of European customers and UK, not technically in EU, but in physically in Europe, like people for various reasons want their data processed locally. So this is not data at rest, this is just data processing. Processing. But today it would bounce out to the US and back.

Brett Johnson: Look, the plan is for, to have Anthropic services within our service boundary, just like the OpenAI services do today.

Tom Arbuthnot: Yeah. Yeah. That's really exciting 'cause that just, makes some of the conversation for legal teams inside of customers easier because it's all the same as it was. Even if they, think it's safe and they're happy with it from a security point of view, just the steps to go through. But I'm thinking particularly legal customers, like we've agreed with our customers this thing and that thing. It's just tricky.

Brett Johnson: And I, and it's timely, Tom, actually, 'cause I was, I was... I actually had a conversation with a customer just yesterday on exactly this, is that their... The agreement with their customer on this and, we... Yeah. So yeah, it's very timely. And look, Anthropic won't be the, won't be the last. There's the, there's more on the horizon, right? For those of you that are familiar with the admin console, you'll see that xAI is there, not as a sub-processor yet. But there's, exciting times ahead.

Tom Arbuthnot: It really is. And it's interesting, if we look at the AI foundry, which is different, so that's where you can choose to consume various models from Microsoft. There are. Hundreds of models on there, including Microsoft first-party ones, but tons of third parties as well.

Brett Johnson: Yeah. And that's running specifically within our European data centers. So we do have, and again, for those listening in the UK, we do have OpenAI models within the UK from a sovereign point of view. But the, there is a, there is an ask to grow that from an Anthropic point of view. But I think that we've gotta walk before we can run, so we'll see the Anthropic services within the EU first.

Tom Arbuthnot: I think the demand, I mean, the Anthropic have been on absolute tearing in terms of capabilities, and it just seems to be a really nice match with the M365 world. I mean, well let's go into Cowork a little bit more. So Cowork, as I understand it, is essentially, works closely with Anthropic, that similar model, whereas Anthropic made Cowork work on a desktop accessing files, a Microsoft version of Cowork is in a container in the Correct respecting all the Microsoft security boundaries auditable, controllable.

Brett Johnson: Correct. Everything that you've just said. So all of the, there, all of, all of the prompting, the conversation that you have with it ultimately is stored within the Exchange user's mailbox as prompt versus Purview, depending on what it's doing. So you've got, you've got the full governance, but it really comes alive with the Work IQ, right? And so what's Work IQ? So essentially, if you think about, again, if you're familiar with the Microsoft ecosystem, we've had something called the Graph, right? Which was essentially the breadcrumb of how you interact with your mailbox and Teams and things like this. And then when we start looking and adding memory, right, and customization to the AI models, as in your settings within the Microsoft 365 app, to be clear, right? And then we've got this inference layer, right? So it kinda knows what's going on at the time of you interacting with it. And so it's really starts to bring things to life. And I'll tell you a story. I was, I was working with a colleague, and we were, we're just planning for a, for a meeting, that was happening.. It was actually on Thursday, so it was earlier this week. And we were... I was talking to Cowork, do a bit of research and do this, that, and the other. And it kinda then said, "Well, but you're in the meeting with so and so right now." And I was like, "I am. So can you pull the transcript from that and add that into the, into the thought process?" Which it did do. And I just, it was kind of things like that where you really start to see and feel Work IQ come to life. And if I'm beginning to see that, then you're gonna see it very shortly, if not already.

Tom Arbuthnot: Yeah. Well, Work IQ, this is one of the rare, if I may say so, times the Microsoft marketing team have got a really good name and terminology. Yeah. It is. Like, it perfectly explains that we're taking all the knowledge of what's going on in the organization. So you... For me, obviously a lot of Teams stuff. So Teams meeting transcripts, your chat, your emails, your SharePoint, your OneDrive, all, again, respecting what am I allowed to access. But it's not just, like, it's a layer or an MCP over Graph or a connector. There's a lot of work being done at that Work IQ layer to optimize, cross-reference, help the AI use that data.

Brett Johnson: Yeah, no, it's pretty phenomenal stuff. So have you, have you used it much, I'm assuming? Because it's not quite available yet, right? So to... People might not on... That are listening to this, but customers may not think, "What are they talking about? We don't have access to this yet." But if you're on Frontier, i.e., if you turn it on, then this is a, this is a, this is a feature capability that you could get access to today, right?

Tom Arbuthnot: Yeah. It's really impressive. So like I... We've got it on. But the interesting thing with Work IQ is you can access it through your Copilot experience. So when you're asking Copilot things of... That would live on the other workloads, it's probably Work IQ, and it's increasingly Work IQ. But also there's an MCP connector. So I've connected the MCP to various scenarios I use. Yeah. So like I have, an automation that shoots me a daily summary, like what emails have I got outstanding that I should look at. And it's using the semantics of like my email is a nightmare, right? I get a lot of inbound. So it's like actually just pick out for me what I've missed that is important- Yeah and needs some... And it's very good at inferring out of the 50 emails, like actually jump to these three, watch out for this meeting, you've got this tomorrow. Like that level of inference and, anything I've done on Teams as well, meetings, prep calls, it can pull insights out of those as well. But.

Brett Johnson: It, but it obviously is way more than that, right? So there was a demo... There's a demo that I did at Cowork at an event I did for public sector in the UK, a couple of weeks ago, where I actually got Copilot to manufacture a bunch of, freedom of information requests pretending to come from, the public to said, central government departments. And I think we got... I got, I created what? 10 FOI letters and I, and I, and I essentially dragged the Word documents that were not real, but into Cowork. And it was a bit a bit of a long prompt, but it kind of just said, "Look, here's some FOI requests. I need you to, just rate them, understand the severity or the impact of what's being asked, create this Excel spreadsheet that summarizes it all. Then go and write the replies just to say thank you for sending the FOIs in. We're gonna work on them as per the SLA," blah. It went then and processed them Right, to actually write the official responses, put that into another Excel spreadsheet, as well as create a PowerPoint that was actually- Mm for the individual to then go and review and present to their, to their team, package it all up into an email ready to be sent, and it did it in, minutes. It was incredible what it did, right?

Tom Arbuthnot: Yeah. All in PowerPoint... That document creation, the first time I've started using it for PowerPoint creation, and PowerPoint creation I feel like is the, Feels like the AGI test these days. Like, once AI can do decent PowerPoint, everything- Well... Else is gonna be easy, 'I know. I know... That's been the hardest thing to nail. But Cowork is the best so far at generating decks, and it's doing that in the container in the Cloud, and then when it, when those docs are ready, putting them into your OneDrive- Correct so it's all secure end-to-end as well.

Brett Johnson: Yeah. And so, yeah, the input, the input assets and the artifacts that come back out, you're absolutely right. There's an input and an output, directory within OneDrive, and of course, the labeling of the documents and the security wrapper is all there as well. It's all there. Yeah, it's pretty exciting stuff, actually. But.

Tom Arbuthnot: It's even, like, it's things like it can now scan my diary and move meetings and send emails and, like to schedule tasks as well. Like, it really has jumped from a thing I prompt in, classic Copilot to do a thing into a. Coworker. It's, again, well-named. Anthropic take the credit for this name, but Cowork is a very good name.

Brett Johnson: But don't forget you can also talk to it, right? So when a we're, as we're talking today, Tom, I guess a lot of people may be thinking about the Microsoft 365 app or through the browser. But they, don't forget that there is actually the phone versions as well, and we've got the recent, iPhone CarPlay, Android is coming. So you can actually have conversations with it as well, in much the same way that we're talking, which is phenomenal. Yeah.

Tom Arbuthnot: That is huge. I do that a lot on the desktop as well. The Windows H just chat away. Yeah. Like, and honestly, that can help your prompting a lot because when you're prompting by typing, you're liable to do the shortest possible thing. When you can ramble on a bit and be like, "I'm thinking this, I might do that. I like... How about this? Remember that?" Like, you can just literally, you can ramble away a bit, and it will then organize- And... Your thoughts and say, "Okay, based on this, here's what I would do." It's.

Brett Johnson: What I'm doing all the time. My boss knows this. I go off, take the dog for a walk. I offload my thoughts into a transcription. Copilot then makes sense of it, and sometimes it's actually creating the Cowork demos that I'm doing for customers, right? I say, "Right, I'm about to meet this customer. These are the people I'm gonna meet. Let's, let-" Mm. "... Let's have a conversation and figure out what would be the best demo." And then it will write the prompt that it, what, that I should be putting into Cowork. It's fabulous. Love it.

Tom Arbuthnot: Yeah, I love that. I do use voice a lot. In fact, I probably, I probably tip the point where I'm more voice than typing. Certainly with AI, it's always voice first. Yeah. Well, I, When I can. I say the mobile app it's great as well.

Brett Johnson: I must admit, that one of the things, outside of work, I'm doing a bit of vibe coding and, there's the odd YouTube, channel that I'll go and look at just to keep up to date with things. And actually being able to just talk, interact, plan what it is I'm gonna go do, and then it goes and executes on it. It's what... It's super. It really is.

Tom Arbuthnot: Nice. Well, tell us what you're doing with the vibe coding. So I was at Microsoft a few weeks ago, and I feel like 50% of the sessions with product teams were, like, either Copilot CLI or Claude Code, and they were like, "Watch this, watch that. We've got MCPs for all this cover." Yeah. Do you know what? That space is super interesting.

Brett Johnson: Do you know what, Tom? It is, it, I appreciate you asking the question because actually this is really helping me learn, because we all have to learn, right? It's not something that we just know, right? You have to take time out to actually fiddle and get Yeah explore and be curious, right? And about a year ago, I got into telescopes and astrophotography. It's always been something that, a little itch that's I've had for all of my life, and I thought, "Do you know what? I'm just gonna go and I'm gonna go and do this." Anyway.

Tom Arbuthnot: Feels very on brand for you, Brett.

Brett Johnson: Yeah. No, thanks. I went down a rat hole, and I ended up... And for those of you that, are into this or photography and sorts, you'll know that you have to take long exposures, and you end up with multiple files. Like, and you can end up with hundreds of files. And I was going through this routine thinking, "Crap, there's gotta be an easier way of how I organize my life." So anyway, I was chatting to Claude and I said, "Right, here's my problem. This is what I'm gonna do." It was actually on a flight to New York. It produced a technical spec, which was 24 pages long Right? And then I gave it to VS Code that ultimately I then just said, "Look, here's the spec. Can we, can we, can we make this thing? But I don't really know how to make it, so can you help me make it?" Which it did. And the net result now, I plug in the telescope, it recognizes what it is, it pulls the files off, it organizes the files, it deletes the files off the telescope, and then imports this into this log, into this application that explains what the object is, all of this stuff. And yeah, it's really cool. It's a personal project. Maybe one day I'll widen the net and others can use it, but for the moment it's keeping my life in check. But the point is, Tom, going through that, I've learnt a tremendous amount outside of my day job into Yeah... Curiosity of building things and using different models and getting exposure to different extensions and add-ons and all of that stuff. It's been good fun.

Tom Arbuthnot: It's amazing. It's such an exciting time to be doing things like that. I mean, we're only a small organization, so we're only 10 people, but, like, things that you wouldn't have been able to do before, you can now be like, we do a lot of content, as the newsletter, the podcast- Yeah like content pipelines, organization, research, like all of that stuff. And you hit a key point there, which is I don't know how to do it. Like, this is something we're trying to instill internally is like just say to the AI, "I wanna achieve this. I don't really know what the... Am I even asking the right thing? Ask me." Go... But that PRD thing of creating a spec, like getting over the hump of like I can even say, "I vaguely have this idea. What would the best thing to be? Ask me questions." Yeah. And it'll come back at you being like A or B, this or that. Like, it's amazing.

Brett Johnson: There's.. Microsoft is about to release the new Work Trend Index for 2026. And for those of you who are like, "What's that?" So if you go and have a look at Work Lab, so aka.ms/worklab is where there's a whole bunch of research and the Work Trend Index for Copilot, will land shortly. But a couple of years ago, there was one specifically about the habit. And so I feel as though I'm completely wired in now, right? So everything that I do, whether that's in my personal life or at work, how can, how can AI Copilot help me, right? So but you need to build the habit, and you need to be in this kind of mindset. And so it was really interesting reflecting back on this Work Trend Index report from a few years ago, just how real it is, because you need to think like that. You need to build the habit. I think the next generation, it will be by default, but maybe for a number of us, it's something, it's a muscle that we're having to, having to strengthen and to learn.

Tom Arbuthnot: Yeah. No, I think that definitely, that definitely resonates, and it is. And it's, once it, once it starts to click, you're like, "Well, I'll..." Like, it's almost I'll try that first and work backwards and then you're like, "Oh, it can automate this. It can do that. It can help me with this.".

Brett Johnson: I hope I didn't bore you with my astrophotography,.

Tom Arbuthnot: No, it's awesome. It's nice to hear a little bit of, yeah, personal... And I agree with that personal project, particularly if you're a listener and you're in an enterprise, like, there's gonna be gates, there's gonna be control. Quite right, there should be. If you can find something out of work that you can vibe code on and mess about, you can take that understanding back to the workplace. Yeah. That's gonna set you in good shape, I think. And.

Brett Johnson: And I think to just double down on that, and maybe to just to close it out, if there is a, if there is a, an action or some form of activity that you do, you can then transpose that into some instruction set, whether that's a declarative Agent or Copilot Studio or Cowork, and there's these skills in Cowork that are coming as well, right? And the skill.md files, which will then consistently do what you've asked it to do. But because you know what it is that you do, and you can put that into practice, then it can really transform how it's done, and that's this is the change that we're going through.

Tom Arbuthnot: Nice. Last thing I want you to decode for us, Brett, is Agent 365. Sure. So as we, as we record, it's GA.

Brett Johnson: It.

Tom Arbuthnot: Is. As of today. It's been in preview for a while. Yeah. So what is it? How do I use it? What's it for? Like, where does it fit in this story?

Brett Johnson: Yeah, no, it's... Thank you for asking, right? So basically, we know that there is gonna be a large number of Agents that are potentially going to be created across the organization, and there are many canvases in order for that to happen, right? So there's Agent Builder within the Microsoft 365 app. There is obviously Copilot Studio for those individuals that have access to that and want to then perhaps do more. There is, we've you've already mentioned about AI Foundry. But also Agents can be made in other third-party systems as well. So there is likely to be an explosion in some way, shape, or form of Agents that will assist this digital transformation of stuff. There Nervousness around this explosion about how they are managed, how they are governed, what's the life cycle management of those Agents in Yeah.

Tom Arbuthnot: And va-and valid nervousness. We've seen this before Yeah. I do disagree... In Power Apps and Power Automate and like Teams sprawl. Like this is the next potential sprawl.

Brett Johnson: Yes. That is, that is, that is very true. So what Agent 365 will provide you is the, essentially the observability and then the subsequent controls in order for you to manage Agents, give it this life cycle, be able to turn them off, be able to manage them, be able to gate them, right? So as and when Agents get created by individuals, that they want to be able to have them published or not, and, control the sharing and all of those types of things. So basically, it's a way of providing security and governance for the Agentic world that we're about to move into, in addition and as to how you govern and manage people. So Entra ID is typically Azure Active Directory in old money, is where we would go and manage all of those things. There's a new concept called the Agent ID. And so as and when Agents become into the mix, you can choose to give them an Agent ID 'cause there's a foundational layer, and then there is, the Agent 365 capabilities that will turn on and give you enhanced observability and things like that. So what's interesting though, Tom, is that it's not just a capability that you see within the Microsoft admin portal, right? Because it wraps around Microsoft Purview and Windows Defender. So the controls that we have to keep Entra ID objects safe and secure, we're utilizing those same controls that you're already aware of to keep your Agents safe and secure.

Tom Arbuthnot: Yeah. Microsoft have a pretty good, foundation in, like, thinking about people and security and RBAC and access and logging and control and Purview and Defender as you play into that story. So take that story and think about it in Agentic terms, I guess.

Brett Johnson: So yeah, and look, that... Agent 365 is available now, right? It's been a long time coming, frankly, and it's quite exciting. I was looking at it today. There was a load of capabilities that landed in my demo tenant literally just last night in preparation for general availability. And it's really quite exciting, and it's really nice to kind of see it come together, to be honest. Yeah, it's, it really is exciting. And you.

Tom Arbuthnot: You mentioned they're third parties, but I think that's a key part of the story, is like being realistic and pragmatic, like different organizations are gonna have mixes of platforms. Of course. We see all the big providers having an Agentic option, obviously. So how does that third-party piece come into play?

Brett Johnson: Yeah, so good question. So via, and currently the Agent 365 SDK, so the software development kit, it's essentially you wrap those third-party Agents via the Agent 365 SDK, and then those third-party external Agents that doing whatever can appear within Entra ID and then subsequently managed accordingly. So, there... We'll see more on this as we go on the journey, right? There's the... We've enabled what we've enabled today, and then there's a whole bunch of capabilities. There's a new NDA roadmap that landed at the same time, right, Tom? And there's the, there's loads more. So what you see today.

Tom Arbuthnot: Yeah, we saw on just keeps adding... We saw on the public roadmap some kind of clues towards like, direct integration of registries and things. Yeah, that's right. So, And... Stay tuned for that, I guess.

Brett Johnson: Yeah, and I think one of the, one of the exciting things, I'm a very visual person, so Agent 365 provides this map capability, so you can see the world as to where Agents have been created. And, I think you shared with me a story not so long ago about, Microsoft's view of Agent 365 and the small roadmap.

Tom Arbuthnot: Yeah. I think it's, I think, I think the number's public, which is there's well over half a million Agents live in Microsoft 365. We were on campus again a few weeks ago, and we got on in, NDA session on it, and it was like the scale is amazing. And if you think about, obviously, Microsoft are very forward-leaning in this. Everybody's very intented to get into it, but, like, that similar kind of number of Agents per employee, we can argue over what the number is, but there's gonna be some number, and they're going to need to be managed.

Brett Johnson: Yeah. Well, exactly, and the... You could... There's all sorts of places that Agents can be created. The SharePoint Agents, we necessarily haven't called out. So you've got the map.

Tom Arbuthnot: Being.

Brett Johnson: Visualized where those Agents have been created and also the third-party places as well. And we're gonna And.

Tom Arbuthnot: And the other thing that was interesting was the Agent-to-Agent scenario- We've got it, yeah which is obviously gonna come to life as well. Like, obviously we just got the SharePoint. You talked about skills earlier. We got SharePoint skills publicly launched, I think it's public preview, launched, a couple of weeks ago. But, like, we're going to end up with scenarios where this Agent is working with this Agent and what happens between them and how do we audit that? 'Cause now there's no human between those two scenarios. We still want to know what's going on. We can.

Brett Johnson: Still put a human in the loop, of course, right? Don't You.

Tom Arbuthnot: Could do. You could do. I don't want to slow it down though.

Brett Johnson: But no, you're right, and the map and some of the other, and the observability that we've got makes it simpler to kind of understand really quickly what's interacting with what and what's going on. Right.

Tom Arbuthnot: Awesome.

Brett Johnson: It's fun.

Tom Arbuthnot: Brett, thanks for covering so many topics in such a short period of time. Hopefully we've given people some greater insight into those different- Well.

Brett Johnson: I hope, I hope it's been useful to you... Parts of the company. We've discussed a lot, Tom, and I genuinely, I, really appreciate being invited back. Obviously doing something right.

Tom Arbuthnot: Yeah. No, yeah, it was a great... It's, I feel like you, that I've thrown a lot at you, and you've decoded it well, so I appreciate that. Thanks for sharing all of that. And, yeah, maybe we'll do a, we'll do another one sometime and get more into your vibe coding and what you've done there.

Brett Johnson: Well, maybe. Let's do, let's do a part three in due course. Thank you, Tom.

Tom Arbuthnot: Awesome. Thanks, Brett. Appreciate it.