Abram, how are you, mate? Back again. Hey, always a pleasure, Daniel. Nice seeing you here this week. What's going on? How's your week been? What's been going on in your world? Uh, crazy busy. Uh, as it is every April, uh, we always complain about it and we do nothing about it. Uh, that's as it goes. Yeah. Uh, yeah. Well, well, for me anyway, I have had uh a lot of chats this week, and we're going to chat about a couple of things in this this episode, but a lot of chats with with organizations uh getting started with uh with AI and Copilot in particular. Um now, what people are starting to feel is that they're falling behind a little bit. We've got all of this talk about uh all these these new technologies and these new things coming out. We've got we're going to chat about agent swarms in a minute as well, which a lot of you watching and listening may not have even heard about yet, but we're going to have a chat about that. So, people are starting to get this uh this idea that they're falling behind at a rapid rate of knots, but look, the reality is that people uh you know, every it's still early, extremely early in this space. So, um those watching and and listening um don't be put off. You're not falling behind. uh there are a lot of things coming out but how do you how do you sort of what's your take on that? Um with things moving so quickly. Well, yeah, it's been really astounding how fast the industry, the tech and even Microsoft is moving. You know, we've got the saying in our organization uh that a week in business chat is a month anywhere else. Uh because we are you know of course we were all working on something else before this and we are moving so fast like we make decisions we go uh it's just absolutely astounding how fast we can move and yet uh the open- source uh and the startups are moving even faster right this week we had multiple releases you know from open AI for the 4.1 4 mini nano 03 3 uh IO4 mini um I probably missed one or two in there not to mention the other labs and all the things that they are doing. So it is moving fast. Yes. Uh that is totally true and also it is still very early and it's going to keep going fast. Yeah, the good news is you know as much news and hype and everything that comes out right we can see the trajectories we can see the trends we know the things that are coming out and the way they are going to improve things. Sometimes we get a little bit surprised but we can see these things coming and we can plan for them uh and get them into their products uh make them you know safe and available for enterprises and other businesses to use. the topic question without notice. Is it I know we're both sort of ingrained in this AI uh space every day, but is it a fad? Is it is it going to go away? And is it you know we we've had when it first when AI you know the conversations first uh you know arised and all that sort of stuff everybody it was moving really really really quickly and everybody was talking about no it's just a fat it's just going to it'll pass over. It'll blow over. there's no point sort of, you know, going too far down the track. What do you what do you what's your take? What are your thoughts? I do not think that this is blockchain 2.0. Uh that, you know, I I never got into NFT and crypto and blockchain. Um, of course, Bitcoin prices don't seem to be a fad with however many trillions of dollars are in Bitcoin, but we don't actually have a much utility out of it yet other than, you know, betting on the price. I This one is not a fan, right? You know, and we can point to a few other tech trends and say, you know, sure, you know, it didn't matter that much, but we can also point to mobile mobile phones, smartphones. Yeah. the internet, the PC, um, you know, they they weren't fads. You know, they took some years to really find their footing. That's absolutely true. But this wave of AI is not going anywhere. I agree. I don't think so either. Um so hey I want to want to chat to you about uh an area that's probably it's rapidly changing at the moment and and it's a it's probably the biggest area uh the advancing area at the moment is around this this concept of agent swarms. So having multiple autonomous types of agents working together in a more I guess coordinated and distributed type of way and really forming your your digital employee type of ecosystem. Um we've spoken before about how you know how good agent uh I'll say agent architecture is anyway in terms of having specific agents for specific tasks. So handling um one type of thing. So in this concept of agent swarms, each type of agents really typically specialized in a for a particular car a task and then they can call on each other. How are you seeing first of all how powerful is that? Second of all, how are you seeing the potential of that in organizations? Are we going to start to see I can see you know single person organizations now having not actual physical employees but entire companies being built um with these type of agent swarms. You know I think you know the the term swarms you know really seems to come out of like science fiction I think and we imagine you know some superhero movie where like the villain or the hero has robots all over. Uh, my kids watch Sonic, Sonic 2, Sonic 3. Um, right. And Dr. Robotnik has got all these, you know, robot swarms, you know, doing his bidding. And I think we kind of latched on to that mental image of these things in a way that's fun, but, you know, also probably not all that helpful. On the other hand, like we we can see where this trend is going. You know, like I said, you know, it's any particular new release of, you know, OpenAI's swarm prototype or Google agent to agent uh or other things like individual features can be, you know, interesting and a surprise and get a lot of hype. But we can see this trend for sure of moving towards these more specialized agents working together on your behalf. I do think, you know, the technology is moving that direction. You know, as we iterate and find AI's footing a little bit better, we're finding what works better. And this is something that we've absolutely learned in Microsoft 365 C-Pilot and we've seen, you know, the lessons apply to other AI products as well. Uh, where they start off, you know, they're just model knowledge and they're great for that. Then they add web search and then they start adding a few more tools. Uh and then they kind of hit a wall for how much how many different things that they can do. It turns out it is incredibly difficult to have our current technology of a language model and you know language model orchestrator you know for these transformer models to coherently consistently accurately relevantly use tools. Uh, it only works for a handful. You know, I'm hearing, you know, for a long time it was only about five. It seems like maybe the industry is now getting to about 20, which is encouraging. And if you look at uh 04 Mini, for instance, that was just that just came out. It's actually able to use several tools, you know, quite consistently and to to great effect. But what it can't do is a thousand things. It doesn't have a thousand tools. You know, an example I usually use in Microsoft 365 Copilot is, you know, add a comment to a Word document. That's relatively common, but it's also not in the top 10 things, not in the top 20 things that I do at work. And there's a thousand or 10,000 capabilities behind that. And so, we imagine AI just able to just do all of these things for us. And it doesn't really scale that way. What's really great about agents and particularly agents working together in a swarm or you know if you want to call it a crew or whatever you want group is what autogen calls it. Lots of terms surely surely surely as a product manager you got to come up with some type of trademark word for a Microsofty co-piloty type of thing. Yeah, we'll see. I don't know if a whole plane of co-pilots uh really is the correct limit. Um, yeah, fighter jets. They fly in form. Formation. There we go. A formation of agents. Maybe that will be it. Get something. Maybe something out of maybe something out of Top Gun or something. We get Get some opportunity. Squadron. Squadron sounds cool. But the great thing about these is each individual agent is awesome at its task. It's and it's tools and it's got the instructions and the capabilities to use, you know, a handful of tools that are necessary for that role, persona, piece of this process, business task, whatever it is. And when you have a um an orchestrating agent, let's see, top gun, this would be squad leader, wing leader if this was Star Wars, right? Iceman, Iceman, I don't know. Maverick, which leader, that one. Um, now it only needs to know about 10 or 20 agents. And if they each know about 10 or 20 skills, well, great. Now we're at 100 4 to 400 different kinds of things that it can do. Possibly it can scale farther than that. Uh, we'll see. The other really useful thing about agent to agent uh is that it each of these individual agents gets its own room and its own instructions. Like if you try to put all of this information into a system prompt about 400 different things it can do, right? It's really going to lose coherence very quickly. So exactly what shape this turns out to be if some equivalent protocol to MCP really takes off in the industry or you know it's built into the API like OpenAI does with agent handoffs or we have some sort of co-pilot squadron uh or whatever that you know the individual feature and its exact shape isn't quite obvious yet but I think there is a clear trend on this and now of course it does allow you to do much broader things because it can use those 100, 200, 400 different skills and capabilities. And now we're getting to like a useful subset of the number of things that I do. So, where we at with you you speak about these these orchestrator agents um where from a uh specifically from M365 copilot perspective, where are we at with that? Is that where are the capabilities at? Well, you know, I was pretty intentional last year when we released the original, you know, the agents product, uh, and how we built it. Um, you know, I knew at the time when we started working on it in January of 24 that we didn't yet have the technology for agent swarms, like they weren't going to work yet. Um, but I could tell already at the time that that is the trend that we were going to be moving towards. So we were int I was intentional when designing this first pro the first version of this product is yes you can click on an agent and have a conversation with it but more interesting and what I recommend anybody to use once they're familiar with an agent is to use it in one conversation with by app mentioning so you can start with co-pilot or one of your agents and then you can continue the conversation with another agent you can continue that conversation you know with a third agent so what that's doing is making you the squad leader, the swarm owner, uh what beehive keeper, um API. Anyway, okay. If you're if you're if you're watching and listening in the comments below, uh or hit us up. Give give us your idea or your thoughts. Let's let's uh let's brainstorm some ideas for our our our swarm swarm leaders. Anyway, continue on. So would uh the feature the product that we have allows you to be that swarm leader. So you have to know which agents to use but you can use them together and they're aware of all the context uh and the project and it's a the by far the most useful way to use agents right now I believe uh is setting up that way. Now meanwhile you see you know features like semantic kernel just released their version of agent swarms just a couple of weeks ago. Um, you know, we're going to have uh a lot of movement in this space coming up soon. Yeah, I know. It's it's it's interesting how you and I love the way that you explain and I use this I I've stole this off you by the way uh when I talk to people about using agents uh in the in the context of a conversation. So being able to atmention and we've spoken about this before but that single conversation thread becomes that type of almost like a project as as you you call it as well and and I guess having an understanding of what different agents do and bringing them into that that same conversation and that's how you sort of hand off different things to different agents and it it's a good good concept that you talk about where you become that orchestrator uh agent well, the human orchestrator um and and bring those in yourself, which I think is is a bit of a mindset change on on how and we'll get into this uh in a little bit as well, but it is a different way of thinking about getting a task or a project or a job done. Um but it's it's something that needs to become a habit, I guess. Um yeah, interesting. Now, there's a lot of conjecture, not a conjecture, but a lot of debate and topic and conversation about automations versus agents. So, what's the actual what's the difference? I mean, we've got agents that can do automations. We've got automations that that can do some tasks that agents do. So, what's how do you explain what the difference is between automations and agents? Well, you know, a lot of the difference is just uh it is new marketing, right? There is some of that, right? I used to work on cloud and then I worked on big data analytics and then I worked on AI and now I'm working on language models and that now agents, right? the you know there's a lot of companies will take you know this that trending term and say oh yeah our existing software is that or they'll add like very minimal kinds of things and say oh our existing software is also this thing and now you know our product is agents uh but I think there is an important and useful distinction between an automation and an agent so the one hand like we've had automations for a long time and really you know if you get all the if you go all the way back right really just sequential programming is also an automation and software engineering is largely you know creating those automations one from one end and of course it advanced right to robotic process automation uh and other kinds of things and across systems like Zapier and if this then that The difference between an automation and an agent is how we it's decided how the steps of the work are decided. That's the really key thing. So in Power Automate Zapier or if this then that you can call a language model to do a thing but you have to call it in sequence. you know, this thing happens and then it calls this language model and even potentially calls that language model with this prompt and then it sends you an email and those four steps always happen in that order. When it's an agent, instead of those four steps always happen happening in order, the agent decides whether a step is necessary and in what order to do them. So the technical definition you know that I like from the industry is LLM plus tool plus decision loop. That decision loop is what I'm talking about here where the language model it calls a tool it gets the results of the tool maybe send an email and gets success or searches the web and you know gets some of the information but then the decision loop says oh that wasn't the information that I needed actually I need to search the web again with a slightly different phrase or it made me learn about this thing and now I'm curious and want to go learn about another thing. So, this can be super powerful and I've been playing with 03 that does this natively now in chat GPT just all of the time. Uh, and it's really impressive how many things and different things it can do. I saw a scenario just yesterday. Somebody took um a picture of a menu, a Chinese restaurant menu that did not have the restaurant name on it, and asked chat GPT, find this restaurant, and it used the user information tool to get its IP get the user's IP address to find out where they're located, you know, within like a zip code. Uh and then it must have searched the web for all of these different uh dish dishes somehow. Yeah. Of course, it had to translate uh between Chinese and English. Uh, and then it looked up images of like the food at this restaurant and matched them to the menu names really and then told the user what the restaurant was, which is, you know, a dozen different tool calls using Python. Uh, really astounding stuff, which nobody would ever build an automation for that. Yeah. Right. That that would be a crazy thing to do. That would be insane. Yeah. Now, that isn't to say that you should just move from automations to agents. Like, that isn't the lesson here. Um, the downside of this flexibility is it works less often. Basically, I was going to say there's a bit of I guess you can sort of see the apprehension of of certain people and, you know, in moving to that agent type of um structure, right? Yeah, absolutely. You know, I I used to work on uh Viva Insights, you know, and a big a big task of Viva Insights and workplace analytics before it was uh what we called what business process automation, BPA. Yeah. Whole whole acronym involved. I had to learn about it as I worked on the product. Uh but there's so much work that goes into like figuring out what the process is, right? That's like 98% of the work. Uh and then it's okay. Well, now it's pretty straightforward to plug together with with one of these tools or even sequential programming. And there's a lot of business process that work that works that way and should continue to work that way. Basically, any time that you can use an automation to solve some problem like you should. The problem comes in when you know new scenarios come up or it changes. And so there's only so far that you can take automations as well because any particular process is going to change and get modified and be a little bit different next time that it runs and so on. Agents can really help with that that makes them really useful here that they can change and modify. Um but then also they can also they can just do a new thing like find the Chinese restaurant from a menu. Uh which is not something you'd ever try to build in an automation. No. No, absolutely not. I mean, we've we've moved we've moved pretty quickly. I mean, as we said before, this this space is moving very very quickly. We've kind of moved very quickly into this this topic of agents. Um, where, you know, initially we've been in this chat type of experience. It's more transactional. People are used to using Google, putting in a search search term, coming back with responses. we've moved through into more of a chat type of of scenario. Um, how important is what from your experience, how how important is learning that that conversational style when we're starting to now think and talk about uh, you know, how agents can can sort of plug into to businesses or or individual roles really. Yeah. you know there's two ways to use chat and you know generative AI you know these interfaces one is prompt engineering and you know there's lots and lots of guides on prompt engineering and you know be specific and detailed and give a persona and we've even got some of those prompt engineering guidance for agents because the instructions of an agent are also prompt engineering how important before I'm just going to interrupt for a How important. So prompt engineering was a massive thing when think when when chacha bet came out, right? There's there's going to be a whole, you know, there's going to be jobs. You're going to be a prompt engineer now. There's going to be all these jobs that appear. It seemed to have um I know not disappeared, but took a little bit of a backseat. I mean, how how important do you think, you know, structuring prompts the right way and personas and all of that sort of stuff is? How important is that now for users? Not at all important. Not at all. Yeah, if you're going to go build a product and you're trying to get to, you know, 100 million annual run rate, like you had better treat your prompt like if software, right, where it's into individual parts, there's multiple teams maintaining it, you're running complex evaluation suites, you know, there's long online AB tests like there's software engineering of prompts, but the the whole idea of users need to learn this new skill of prompt engineering, it's just like learning boolean logic. for Alta Vista in 1998. Like it's the same thing, right? That Google in 1999 says we don't need any of that. Um sorry everybody that looks fully in search terms. Um we just fixed it so that end users don't have to. But similarly, if you're building a search engine yourself, right, you'd better understand those boolean search terms, how they work, how to apply them behind the scenes. So prompt engineering, you know, it it fully moved into software engineering. And so that was one way of interacting with these chat interfaces is writing super good props. And that's still important for software developers and even agent developers and makers. You know, there's some strong things that you can do there. But users should really focus on chat and on conversation. Well, first off, it works now, so that's good. like like Google fixed you know search terms in 1999 you know GPT oh let's say probably GPT40 is probably the one that really fixed uh needing to prompt engineer and of course the competing models have as well the great thing about conversation is that everybody knows how to do it where you're having a conversation right now and you didn't have to train me in talking right um my mother did that when when I was two. Uh, you know, my brain was very plastic and I got very good at it by the time I was six. So, conversation, you know, comes naturally. I consider it a skill, right? I I don't put it on my resume. Can talk can talk good? Well, you you'd probably be surprised. I mean the pro people are probably going to to look for that and and and how um you know it it would it's probably going to be a skill being a good conversationalist and knowing how to have a conversation is probably going to be extremely important in in jobs moving forward I would think. Well, that is true. You know, that's a good point. You know, I see some people on, you know, Twitter or whatever that X, I should say, um that just spend a huge amount of time with language models. Uh and you can see some of the results that they get. You know, just today I built an agent uh that is argumentative uh and like a person online. I I built this agent. It's like a person online and it was such a good prompt that I found on, you know, from somebody's Twitter account, X account, that I had to like I had to go put this into an agent and try it out. So, first off, it drops all the capital letters. It says, "Use millennial slang." But and then occasionally, ironically, use Zoomer slang. Um, it just What the hell? What's Zuma slang? What the hell? Zoomer. Oh, um, the broccoli haircuts, right? Um, what do they even say? Uh um skiibbidity. Skibid is what my daughter says. Um right. Right. Yeah. And I can tell you what it means, but it would just it would just make you less intelligent to learn. Okay, we'll leave it we'll leave it there. So, it's true that people that are really good at these different styles of conversation can get really awesome results, but it's not any particular one way to do it, right? You know, like uh Ethan Malik, you know, is calls himself an LLM somoleier, uh which is really cool. And you know, he just comes up with really interesting things to ask, you know, and then others, you know, uh, like seem to like dive into the psyche of these language models, uh, and change how they think with some really fascinating results. Uh, but you don't have to do all of those things in order to be productive at work, right? You know, those things are fun with as toys and like explore and push the frontier, but just having a conversation and this is new because the thing looks like a search box, right? And so you think that you should type in a search term, you know, a search phrase. And even when we say you can use natural language like what we what people understand that to mean is okay search terms written in a sentence find me like you know the latest score of the Dodgers game right like it's not actually any more useful information than Dodgers game right that you can type in your Google right now. Yeah. the the the the pattern to learn it and not even learn but get used to is asking follow-up questions. Try not to create a new chat, right? I explore a little bit more right after you ask for the score of the Dodgers game. Ask, you know, who the star players are, who are the MVPs uh in that game, who scored the most runs. Describe each inning. I don't know. uh right whatever your interest is in this game try following on and asking that in a conversation because maybe I should use a work scenario for this but for any kind of thing that you're that you're using copilot or another AI for but I think that's that's a good lesson to learn and a good way to learn how to use AI or any any of the lang any of the models as as well right is is trying to get out of that that that search type of mindset but just learn and and just practice and have a conversation and go backwards and forwards and get into the habit of um having yeah actually finding out more information and seeing what happens and seeing the the different types of of words and language and conversations that get sort of get the best results. Yeah, for sure. You know, it's really awesome how easy this AI is to use and how broadly distributed it is. Certainly enterprises are rolling it out but you know any person on the planet you know if they've got an internet connection and a cell phone right they can use co-pilot they could use chatgbt and Gemini and claude um you know these frontier latest bleeding edge AI products and they didn't have to learn a new skill they didn't have to learn English right these things work in basically any language that you know is widely spoken uh and they have access to this knowledge and these new capabilities and scenario iOS and when it's when they can use it at their work, it's making them more effective without needing new skills. They need new habits, but I think it's f focusing on new habits much more than new skills. Yeah, absolutely. Um, now I think you've spoken about this topic before, but um I'm going to ask you a question. Why do you think chat is the best and the worst interface for AI? Well, we were just talking about why it's the best, which is you don't need to learn any new skills in order to use it. Uh if you can talk because now we've got, you know, voice mode uh or transcription, you know, so there's advantages to using advanced voice versus transcription walkie-talkie style. I'm using I'm using Windows H at the moment quite a bit, actually. very effective. Good. Uh last Saturday, right? I I uh you know spoke for 15 minutes with transcription uh and just and then I asked co-pilot, okay, turn it into this format. Um right, ended up with a great great resulting document. So you don't even have to type be able to type in order to use these things. So chat is fantastic. The other thing that chat is awesome for so first so first it's not a new skill that you have to learn. The second is it can do anything. Anything, right? If you can express it in language, you can ask the AI to do it and the AI will attempt to do it and often be successful. On the other hand, it's also this um what you call like the blank page problem. Yeah. where you open it up and there's a chat box and you're like, "What does this thing do? What should I use it for?" The problem is the flip side of that can do anything which is give me something like I need one thing to do in order to start. Uh it's hard to think of anything. Uh that's not how our brains work. Uh, try to come up with like a sentence right now that no person has ever stated. Challenging you, Daniel. Okay. Come up with a sentence right now. No person in the world has ever spoke words in this order. Hang on. I can Well, what was that? Zuma. What was that? What were you talking about before? Skiby or something like that. I don't know. That's where I would use AI, right? I would say I would I would say uh what is a sentence that no one has ever used and let's see what comes let's see what comes let's see what what it does here okay the luminous quarker danced gracefully under the emerald waterfall singing an ancient melody known only to the stars there you go all right who who has who has ever used that sentence before ever. Yeah. Um, luminous is not a word I generally pick up. What about What about quarker? Does that make sense? I don't know what that is. Do you know what a Do you know what a quawker is? You don't know what a quacker? I have no idea what that is. Okay. What is a quacker? And I'll I'll give you I'll give you the the exact terminology. A uh hang on. It's not a quacker. I know what a quacker is. No, no, no, no. Quaca is a small massupi native to Australia. That's why you don't know it. That's why I don't know. Right. As we were talking before, I can only name three cities in Australia. Yeah, that's right. Uh habitat. If you want to know the habitat, they're found in small areas of southwestern Australia. I'm in Melbourne, so I'm on the east coast, including some smaller islands off the coast of Western Australia, particularly Rott Nest Island and Bald Island. There you go. Little bit of little bit of Calker is dancing under the luminous waterfall. That's right. That's right. Anyway, sorry, we digress. So, uh, we were talking about, uh, the blank page problem. Yeah. And, you know, that's since it's this new habit, it's like how do you start a new habit? It's, you know, you've got to do it a few times, right? Otherwise, it's not a habit. And so you we need some way to get from the blank page to using it a few times. This is you know quite challenging and we spend a lot of time in co-pilot on what we the technical term is zero query which comes from like our search days like if you type if you click on this Google search um it'll give you some suggestions or you type in the address bar of edge it'll give you some suggestions. So we call it zero query of why don't you try asking about this thing and it's useful you know we so in co Microsoft 365 copilot we'll show a few zero query cards getting a little bit even more technical right and it'll be something like summarize all of the requests my manager had for me in this last week as a way to try to get beyond this starting problem but of course that's not chat that's not a chat interface at all I'm really you just get three or five or six options of, you know, things that you could use this AI for. And so it does get you past the blank page problem a little bit. Um, but only a little bit. You still have to get from one query to three and from three queries to 10 uh and moving into a habit of your daily life of using the AI. So that's the problem with chat is uh it's too flexible. It's not like any technology that we've ever used before. And of course it there there's the other situation which we talked about last week of it can't do some scenarios. Yeah. If you ask it to do your dishes, right, it can't. Uh right. And there's Well, it can tell you how. It can tell you how. Uh which I did not need. My mother taught me that one as well. Wow. So, if you've got kids, if you've got if you've got kids, then it's a good it's just a good reiteration to get them to actually do working on that one. Waiting for my daughter to ask co-pilot how to wash dishes. She mo she mostly asks about drawing dragons. So, getting past that blank page problem uh is really tough for chat. And that then even once you're there, there's things it can do and there's things that it can't. And that's that is tough to learn. Uh just because the it's a wide open space. We haven't constrained it at all through this interface. So it it looks like it can do anything and it will try to do anything and you can communicate anything, right? But what is anything? What is a thing? You the quawkers of the luminous waterfalls. Uh and then some of the time it will work and some of the time it won't. So that is yeah it's uh like I'm looking for an interface that is better than chat. Uh and it doesn't exist yet u for this. Yeah. It was I was listening to a um a podcast actually this morning um and it was it was talking about not specifically copilot but just AI uh adoption in uh or getting started and and making it a habit to to start to just to introduce it into your your daily workflows. But it was the um the guy I was talk listening to was uh Nathan Barry from ConvertKit or or now kit. And how he sort of started um making it a habit was he just he got a an additional monitor. So it was a vertical screen and it just he just had in this case it was Claude cuz he was he was using it for for writing and the output from Claude is is quite good with with writing. Um but it was just there constantly all the time. And you know whenever he started to to do something, it just you know because it was there visually in front of him started to then you know become um a habit to just ask the ask Claude to do some stuff. Have you got any how how have you seen people sort of integrate or or change behaviors and habits in in that that in that sort of way to to bring it into to everyday workflows? Yeah, writing is a great example, right? So, I write a fair amount, you know, as a product manager and um what I do is I use the web app uh so that I can put co-pilot on the edge. You know, you can click on a co-pilot button in Edge and it's right there. Uh, and it can interact well, it can co-pilot consumer can read whatever you've got there on the screen. Um, which is, you know, super useful. And we do the same thing in Microsoft 365, right? It's it's there on the side of Word. It's there on the side of PowerPoint. It's there on the side of Outlook. And yeah, it's it's fantastic to use this way. If you if you've got a widescreen monitor, and you should you should have a widescreen monitor, open it and pin it, pin it to the side. Uh, and you know, it it helps as a visual reminder that it's there, which is one of these hurdles, of course, is to change products to go use this AI thing. When it's there on the side, it's much more visible. And when it can interact with or at least read and interpret the things that are going on in the main part of the screen, uh it's super helpful. You know, the the great the really useful aspect of it, I think, is that it's got the context. When you're using a chat interface, you've got to like in a full page like you go to copilot.microsoft.com. Uh you've got to go open up or you well you open it up and it's standalone. Yeah. And you've got to bring in the context, right? And we we make it as easy as possible with slashmention and add files and upload files and points to this and points to that, you know, different M365 things or external things, but it's a bunch of steps in order to get there. this side pane you know we call app chat uh is a fantastic way of operating it I argue that app chat while it's not even my my project um I think it's better than than my than the full when you say chat when you say app chat is that not just in edge but you're talking about the app chat in in the other applications as well yeah yes that's right like in word or in PowerPoint as well as Edge and Teams and and everywhere else. Um I think it's because it's got all the capabilities of what we call you know mainline which is just to distinguish it, you know, the full page experience. It's got all the capabilities of the full page experience plus it's got context of what you're doing right now without you having to type it in and go click around and and add that context in. But really the app chats are where it's at often when you're working in Word, you're working in PowerPoint, what Excel, whatever it is. Copilot Excel is awesome by the way. Um, if you're not using that one, as well as, you know, you can see these features um, you know, kind of the artifacts model that Claude uses and canvas that ChatGpt uses. Um, it's a super strong way of working with these things. Yeah. something that I don't Yeah, I don't do that a lot, but something I'll I'll make a change on is is yes, that that that app app chat experience. Um because you're right, moving, you know, bringing bringing the context into the you called it main mainline or something like that. But yes, bringing that in is is just an extra I guess an extra step. And a lot of people wouldn't realize um especially the edge example is is something that is is extremely powerful for people to to start uh to start using and a really good way to learn that conversation. Uh especially when you're on a page. Um having that backwards and forwards um you know conversation is is again something that is is is powerful. Yeah. A really cool version of that uh that I've seen some people do um is some of these products including Microsoft Copilot. That's the consumer one. If you have Pro, you can turn on vision and it opens. It shows up at the bottom of the screen instead of the side of the screen, but it's the same kind of idea. and you just continue to browse the web like as you were doing whatever you're working on whatever you know whatever it is scrolling Facebook you can just say what does this mean out loud like like you're talking to yourself then copilot will look at it and then describe what it means um is that is that in the in the consumer app yes you're talking that is the consumer one not the work one um but you and you can see people doing this with Gemini as well and Chrome. Um so I I don't think we have we don't have that and I haven't seen it in any work product yet. Um yet well you can look at these trends just like I was saying before you look at these trends and of course it's going to happen in the work products from all of the players. Um now this doesn't work in all circumstances right not every computer has got a microphone. I'm not always willing to talk out loud, you know, depending what I'm doing and where I'm at. Um, but it's it's really fascinating to work with AI that way. It is. But it's been another I think that could be the pod. It's been another great episode. Good chat. Um, and unless there's anything else, we we might wrap it up. Sounds good, Daniel. Yeah, this was a lot of fun to chat about. Until next time. Now people, we are on YouTube and we are now across all of the uh popular or not popular anywhere that you get your podcasts published there on YouTube as well. Head over to return onintelligence.ai. Also pop your email address in there. We'll keep you updated with all the latest episodes direct to your inbox and you can um you can stay updated there. But once again, Abram, thanks mate. It was uh was a good one. A lot of fun talking to you, Daniel. See you.