Reimagining Efficiency and Innovation with Taiki Yoshida: A Conversation on AI and Co-Pilot

Mark Smith: 0:33
Welcome to the Power 365 show. We're an interview staff at Microsoft across the Power Platform and Dynamics 365 Technology Stack. I hope you'll find this podcast educational and inspire you to do more with this great technology. Now let's get on with the show. In this episode, we'll be focusing on Microsoft Co-Pilot and its capabilities. What's changed? What's happening? What's new? Today's guest is from Redmond Washington in the US. He works at Microsoft as a principal program manager. He regularly creates apps and automations with Power Platform, even as free time For his family. He started his Microsoft career in 2018 to build the Power Platform business in Japan. You can find links to his bio, social media, etc. In the show notes for this episode. Welcome to the show, taiki Hi.

Taiki Yoshida: 1:22
Mike, nice to meet you.

Mark Smith: 1:23
You too. It's good to have you back on the show. I think it was show 290. 290 was the last time you're on, which is about over 200 episodes ago. Wow, I'm in the high 400s at the moment. It's good to have you back? Yeah, I'm honored to be here again. Tell me life's getting back to normal for you now. I take it with COVID squarely in the rear vision mirror. What does your week-to-week, day-to-day look like when you're not at Microsoft?

Taiki Yoshida: 1:52
That's an interesting one. Right now in Redmond Washington it's sunny and it's a beautiful weather outside. I tend to spend time barbecuing and having a lot of friends and folks. Definitely, the COVID has died down, so it's not a lot more time spent outside and inside for sure.

Mark Smith: 2:11
Nice, do you get back to Japan much?

Taiki Yoshida: 2:14
I do. Actually, my family is now back in Japan, so I actually fly now probably two or three months each time back and forth. And all that Do you miss home? I do, yes, definitely, especially Japanese food. Maybe I shouldn't tell my wife this. But, yeah, definitely the food. Of course, my wife and the kids too, but I do miss them a lot too, for sure.

Mark Smith: 2:38
Yeah, yeah, yeah, but Japanese food is just so unique, right? All the Japanese food I've eaten is like a work of art and it's amazing, and so, yeah, how do you Do you make it yourself when you're not at home in Japan, like how do you deal with the gap? Or in Seattle? Is there options for you that are high quality?

Taiki Yoshida: 3:02
Yeah, luckily we do have a lot of Japanese stores or Asian supermarkets around in Seattle and Redmond as well, so I tend to kind of buy the ingredients there and then I cook a lot of stuff actually at home myself.

Mark Smith: 3:16
Nice, nice.

Taiki Yoshida: 3:18
Yeah, maybe not sushi, very good.

Mark Smith: 3:19
So not sushi.

Taiki Yoshida: 3:23
Do you not?

Mark Smith: 3:23
roll them in all their self.

Taiki Yoshida: 3:24
No, I mean, I know that they have it, but no, I don't make sushi to myself, for sure.

Mark Smith: 3:30
But do you, if you had the choice of beef or, as opposed to, fish-based diet, do you reckon you'd lean more to the beef side of things, or lamb, or more fish?

Taiki Yoshida: 3:45
Oh, that's a good question. I actually equally like both. To be really honest, I actually came from a place called Kobe where I think it's famous for Kobe beef. For me personally, over the years it has gone a little bit too much for me in terms of the fatiness and everything. My stomach can't cope it. So I suppose in that sense I should prefer fish more.

Mark Smith: 4:07
But yeah, yeah, so tell us. What are you up to these days in Microsoft? What team do you work in? What's your general focus?

Taiki Yoshida: 4:15
I think in the last episode I was talking about how I joined this power cat team, where we focused directly with customers within the engineering team. I'm still part of that same team, but my focus has shifted from directly engaging with customers to finding out, most recently, about how the co-pilots and AI, and generative AI in particular, is enabling people and the customers across the globe. So that's where I'm trying to figure out what are the customers are doing, how are they using it? What are the best practices that we potentially could get out of those kind of stories as well? So that's my personal mission right now.

Mark Smith: 4:55
And so what are you seeing? Like you know, I think it's IDC is coined to turn AI everywhere. We've obviously heard the announcements around the M365 add-on SKU, which is gives you co-pilot in the Office suite of products. I'm starting to have customers ask does that automatically give it to us in the power platform and is there a license cost for the power platform? But more than those type of things, because obviously customers are starting to talk about it now, you know they've heard of co-pilots and the concept of a human in the loop. You know assistance in the use of AI, but are you seeing and without mentioning customer's names or anything like that, but in different industries are you starting to see the early signs of adoption across the power platform in the context of AI outside of AI builder, but more in this co-pilot space, what type of use cases are you starting to see bubble up? I suppose?

Taiki Yoshida: 5:58
I think I would say there are two main different impacts that's been provided with co-pilot and generative AI capabilities in the power platform space. First one I can definitely showcase to you as an example and actually can provide real names to this, now it's gone public is the career transformation. That's definitely the earliest signals that we see from co-pilot space. So one example I can give you here is this lady called Miyu Takai, who happens to be in Japan. So she was a lady in Japan who used to do a fax ordering entries into an Excel spreadsheet. So she would receive fax orders that would come out in a paper and then she would go and enter that detail into an Excel spreadsheet. That was her beginning of her career. So she had no experience in around the programming skills as such. It was very far from IT and was like this general affairs person who was doing that. Over the years she finally met Microsoft 365 to begin with and started playing around with SharePoint because the company allowed her to do it, because there wasn't really any governance or any of that stuff in place yet. So she was playing around and then she started to find out about Power Automate and Power Apps and eventually, although she had no prior experiences. She would now be able to learn all these new equations, for example in Power Apps, like the split function, where she would go and she had no idea what split function does. She didn't do the next Excel either, but she would then be able to use the earlier days of Power Apps I think it was called the Power Apps Ideas where it would come up automatically using GPT behind the scenes and actually create an equation that's suitable for it and that allowed her to upscale herself and to become now she actually transformed into a new career and become a Power Platform engineer in a Power Platform partner. So she transformed herself and upskilled herself through generative AI capabilities, even before it was called CokePilot, and now she's able to do this on her own and she's training customers. She's building some new apps based on CokePilot capabilities to become a prototype and learn new skills from it. So that was one example. The other example I'd like to showcase to you, which only literally came out last week, so it's very new about this lady called Ashley Francis and she was based in Seattle and she's a CPA, a Certified Public Accountant. Again, she's been like 20 years in industry without prior experiences in again, in IT. She used to do a lot of her stuff in Excel she's without even a macro or a VPA as such, but she found back 2022, in October, so this is only last year. She finally found out about Power Automate in the Power Platform and she started to start using things like Bing AI capabilities and find out all these different potential automation ideas and eventually she used to describe it to design it capabilities in Power Automate to actually create an entire flow that you can use to build cloud flows right, and that was one of her first ways of finding out how to create Power Automate flows. And eventually, in a month after she had this findings about Power Automate, she built her first flow entirely on her own using GPT, talking to chat GPT or using Bing AI and kind of using it to its best potential from both of these capabilities and then build a flow herself and eventually now she's used even AI builder to extract different sets of data directly from these tax forms directly related to her business. And she wasn't obviously having any prior experiences in data science or writing pythons or anything, so she's seen all these other folks who is showcasing again in the accounting space where they were using Python scripts and other systems to extract these kinds of data, but she was able to do it all without any of those Python or any of the background knowledge in AI to do all of these automations, which I feel is fascinating. So her processes of doing, for example, I think, was sending a letter to an engagement letter to customers, which used to take like 30 minutes per each customer, per each engagement, and that's, you know, time spent a number of customers could be a huge amount of time, right, but with this, now that she learned Power Automate through Copilot, kb, through, you know, generative AI, she was now able to do this entire process in less than an hour for every engagement and every customer. So you know, that's, you know, I think in a sense it's upskilling the people and to make these career transformations to become now AI-infused CPA person. You know it's like the latest. You know you wouldn't be expecting that to happen in just a space of less than a year, but it's happening because of the Copilot and Generative AI capabilities. That's one area.

Mark Smith: 11:25
When we look at the Power Platform and that it's a platform for all developers, so whether they start at the one end of the spectrum where it's around personal productivity and, you know, creating some automations or an app for me in the business, kind of like what Excel has been for a lot of people in the past and then we go through that developer scale and we get people that are building apps for other people. They've developed their skill sets, they know how to create automations, they know how to use connectors, they know how to use Canvas and maybe Model Driven, and we go up, of course, to Procode developers that are doing extending on the platform. When we look at Copilot and we put those two lines side by side, where do you see it kind of support? Who's it supporting at the moment? If you look at those spectrum of personal productivity, fusion teams and then ultimately ProDevs working with the Power Platform, where do you see it kind of its sweet spot? Where in your mind?

Taiki Yoshida: 12:24
So in particular about the co-pilot. I think it enables more of the citizen developer spectrum when, I would say, in a power platform and the low code capabilities not just specifically power platform but any, you know, even other competitors and other products out there where we define it as a low code, used to kind of tend to be those who are a little bit more on the Excel savvy side of people right, who would be able to, yes, at least maybe understand if equations or a V lookup, you know that really helps the customer to actually be able to, you know, do use these capabilities to an extent, whereas now with the co-pilot that really load the bar on entry, now you have people who is right, it would write just word documents Maybe, and in the kind of a same, similar sense, as long as the people can actually just write what they want to have, it ultimately kind of figures out right, using the behind-the-scenes generally abr capabilities, they'll ultimately kind of understand what the people wants and then convert it into kind of these configurations, but in perhaps cases into an application or empowered me to be a flow. So it definitely has lowered that bar and they're kind of the criteria that you needed to learn for you that we look ups and if, equations. And now people can, even though they might not use it directly into the production environment, right away from these co-pilot, it will at least help the people to understand how you can configure it, how to Make these settings, to then next time, when you want to make it yourself, it'll teach them at least really a clear example or business use case where people can just go and look and say, oh, this is how you can build this app or this is how you can build this flow. You know that really helps them, enable People to, you know, develop these stuff more so in that sense, yeah, that's the co-pilot directly has definitely Enable people to and a citizen, develop a side of things, to be more open doors to you know, more people. I would say yeah the co-pilot experience.

Mark Smith: 14:21
It's a take. Let's take power apps to start with before we look at power automate. When Charles Lamana I was on a his first presentation of it to the MVP community, and as he was presenting on the main stage I don't know how many months ago now I jumped into our MVP tenant and it had already been deployed. So well he did that. I quickly created an app For managing the seeds in my garden. Okay, right, I know it's something totally because I'm always I Hate sample apps that are sample apps or demo data. That is dead. It says you know, test one, test two, I like it to be real and tangible, and very quickly, you know, it created me a one-table app and and it populated it with sample data. That was correct. Sample data like pumpkin seeds, like corn seeds, like it. You know it had that understood enough, and from me, of course, just typing in what I wanted and of course, then I saw the second part of chat GPT within an app which is about querying your data. Right, which is it's really two, two parts of that story. Now, that's amazing simple one, one table app in your mind. Where do you see that movement happening up from the citizen developer into either the fusion team and even you know they say in the stat jumps around 40% of code and in GitHub is is Large language model assisted right about 40%. I've heard people even go higher than that. Yeah, outside of Microsoft Say, say that, that code and that. Where do you think, in a realistic timeline, how far along that developer pool do you think that we will go with things like co-pilot, where they are truly? You know that we have, we've truly really talk, gone to the developer model of Explain it and it will build it we, how do you? I mean, we've seen a lot of traction, we've had a lot from Microsoft around co-pilot, of course, since around that November timeframe last year. But where do you really see, because we're coming up, november's, not far around the corner, will be a year in. Are we gonna see kind of like Moore's law, this doubling in and I'm talking about now inside Microsoft? Do do we see that kind of we're just gonna leapfrog, leapfrog, leapfrog in in when AI is gonna go and assist us in the power platform?

Taiki Yoshida: 16:52
Yeah, that's an awesome question and I think, in terms of enabling pro developers, I think there's two aspects to it. One from the co-pilot perspective. I've seen developers which I can't disclose names yet because I haven't got the right permissions, but I've seen pro developers use the co-pilot capabilities first off in Power Apps, where they already had been using Power Apps to develop things faster and they would typically spend maybe a day or so to build these apps and they actually claimed, using co-pilot, it took now them two hours just to build an app that they would have taken about six hours or a day to build. So even for pro developers, it's contributing to build the first MVP of an app because it instantly creates an app that's working, although it may be just one table app, but it's already starting to have the right sample data, have the right details, have the right fields and everything already packed as a first MVP that's working right off, just by clicking Generate and that's it. So that's the first aspect of it and more in a pro developer in a sense of actually coding and things. I also heard another customer which actually wasn't a pro developer, but I would say he has converted into a pro developer by using the capabilities with Generative AI where he was building this. He had this requirement to load lots of data into Dataverse and that was like 60,000 records that he wanted to import. But if he did this with Power Automate, obviously there's issues around throttling and that could potentially get hit. If you do it from an UI it kind of gets tricky with like number of records to the limit. So the guidance we have in our docs is to use potentially Procode. It says write a custom code and then use that to import data. But what he did was that he had no experience with C-Sharp before but from using the Generative AI and eventually to GitHub, enterprise and Copilot, he managed to write a plugin, a C-Sharp plugin. Obviously that's a proper Procode and he developed it in four hours and then it was actually used to import the 60,000 records. I mean, he wasn't a pro developer, but that is pretty much a pro developer right, and so that happened and at one point he was going to go and quote to some external partner to kind of develop, get someone to Procode for him, and that was quoting like 50 hours, I think also. But yeah, he managed to do it in four. So I think that again, to convert even into a Procode. Also, generative AI is definitely helping there to have more people become pro developers in a sense, who wasn't maybe the pro developer, who was maybe a local developer before, but now we're having this sense of you know, because by being a local developer you already have like the basic knowledge and kind of understanding the concepts and the logic and, although you wouldn't call it code but programming, you already kind of have that in your mind. So having then an extra layer of GitHub co-pilot to help you even accelerate that to become a C-Sharp developer, it converts you into a Procode. So definitely it definitely helps the Procodes to do it even more efficiently and also helps people to shift from being a local developer to a Procode developer as well. So that's kind of like the two things that I've seen so far in that front.

Mark Smith: 20:41
What's? Have you had any general feedback? You know, outside of the use cases of you know a couple of things and it's interesting because it wasn't I don't know that it was really mentioned in it, but recently there's been I think it was a Forrester report came out that now shows the power platform is a low code leader like, not not just a leader, but it is the valeta right. It's the top of the leader quadrant and and we've seen service now jump in behind we've seen Appian and Mendex plummet. Really, you know, to some degree. Is the fact that Microsoft is so, you know, iterating at such a phenomenal rate? Do you think that is impacting the draw to the platform more and more that people can see? Hey, you know, if we look at the and I expand on it a bit if we look at the cloud wars, there's only three people standing at the end of the wars. Right, there was Azure, there was AWS and there was Google cloud platform. Really, those are considered the top three. Now we're in the AI that you know, the era of AI, which is happening in this a phenomenal rate, and there's a lot of predictions that there's only gonna be a couple of big AI providers that are at the core of large language models and things like that and, of course, microsoft's in the driving seat To all intents and purposes. What do you think the impact that is having, or what are you seeing as the impact that's having, on Customers switching to the power platform, choosing it, even Swapping out a lot of the point solutions that they're seeing in market, whether there be automation, point solutions or at point solutions. What? What's your observations being?

Taiki Yoshida: 22:21
definitely. I think the generative AI capabilities and the co-pilots that we announced has Provided a big impact towards how the customers are making decisions over which platforms to go for, which you know products to go for. Obviously, you know, with the recent macro economic new changes as well, now the customers are asking us to kind of Work, do more with less right, and I think that's something that that phrase has been particularly used with Satya's Keynotes as well as several times now, and you know having these co-pilot and generally I came with these definitely directly relates to that. Do more with less Part of things. So you know the example I showed you about that. You know people being able to produce Plugins right in four hours. That kind of thing is definitely, you know I Do more of less kind of a concept as well. So definitely these co-pilot and generally I keep these. You know. I know it's kind of a broad Answer here, but it's definitely gonna help your customers and is definitely helping out as as well to have the customers being influenced to have us as like their choice of products and services for sure. Yeah.

Mark Smith: 23:33
What are you doing personally, then, around your AI upskilling and I'm not talking just about what you're doing because of you know, in the role you're in, with your focus on Microsoft co-pilot and the capabilities of it, but outside of that, you know you're in tech, I'm in tech and I'm going. This is a quantum leap, and how do I make it Part of every part of my life? Right, I'm a all-in type person. What are you doing to kind of really embrace AI and and lean into it?

Taiki Yoshida: 24:09
Yeah, that's a good, great question. And what I've been personally be doing so far is I've been playing around with not just the GPT but other AI models as well. I've been trying to kind of use both the AI capabilities in both personal side and work as well. So Some of the things I tried was stable diffusion, which is one of the AI models you can run on your PC to draw something that I think people is familiar with. Dali was, I think, one of the major ones. Stable diffusion is very similar to that and I've been playing around on how I can make my own AI Portrait model myself using stable yes, and it has been fascinating to see that you can just do that in your own personal computer. Right, you know, and that was one thing I've tried. The other thing I've also been working on was Before, you know, as you're opening, I was a thing and I tried also using chat GPT as well and, yes, and that's definitely one of the earlier days. And then I started to kind of try out different AI capabilities like, for example, adobe Photoshop Peter now has Generated filling capabilities, yes, all my great yeah picture. So I've been trying out very different types of AI to be familiar and to be able to kind of talk, you know, maybe in a general level, on you know, what are you? I keep these. That's out there. And what I'm particularly focused and interested on on a personal and business Level is how the AI is being applied To the daily lives and the daily work. More so than that, your model itself. There's lots of people who is, you know, already a Specialist focusing on that. Actually, I models itself. I'm not a data scientist or a Python developer, to be really Transparent, so my focus is kind of more towards how do we enable that to our lives, how do we enable that to our, you know, our business and work and the daily processes and things, and that's kind of my more of our interest there, actually.

Mark Smith: 26:15
Have you outside of Adobe's image generation, which I've played with it, have you? You know, and I and I have a subscription to GPT for and or open AI and I also to what's the other one. I have it to mid-journey Mm-hmm, and I've looked at stable diffusion and of, and then I'm like almost justifying why I need to get even a larger computer Right to do that processing and and the larger GPU. I think stable diffusion is amazing as an I think mid-journey is is accessible to everybody, but stable diffusion I think can go a lot further, but it's kind of nuanced in that it's still, I feel you've got to have a, a technical mind, much more Than a mid-journey in the my experience. What have you thought around the areas of Personal assistance? You know we've had Google Home, we've had Alexa, we've had Cortana, we've had Siri. All of them are dumb in my mind as in yeah, you know, on my desk here. I have the Google system and after Three, maybe four years of having that, I have commands that I say unique commands, and it does things like lights up my studio and things. You know that at least once a week. It doesn't understand my command, even though it's been prompted with the same phrase every day for four years, and I'm like there's obviously no learning, right? If it was a biometric scan of my voice and all the nuances, whether I set it with a cold, or whether I set it in summer or winter or set it as I just passed the door, it would pick up the nuances and still build out the model and go yeah, that's a 92% accurate ease. Asking for that Not, oh, do a Google search on the phrase there, which is what it does as an alternative, right? What are your thoughts around that space? Like, surely one of the owners of those although I was surprised to see Cortana's just been discontinued on the Windows 11 platform just in the last couple of weeks Do you think that? Or what's your expectations on whether one of those players are going to apply an LLM to it and even further and really start to make it the absolute king of the hill when it comes to its powerfulness?

Taiki Yoshida: 28:57
Yeah, I think that's a very good question and I'm also surprised, personally, that the LLM model hasn't been put into any of the personal assistants at the moment. I mean, microsoft, in a sense, power Virtual Agents is kind of like a personal agent, right? Is it a personal assistant? And we've already put GPT Azure opening I should be more specific there that it can be used to automatically generate answers. That's relevant to the question. So in that sense, I think it's a matter of time that someone the big players in Google or Facebook or Google or Apple, or we'll be doing that for sure. One thing I think that's difficult in a sense is that if you're talking about business, you have a particular case, you have a particular intent, you have a kind of particular domain about it, and that's why PVA, power Virtual Agents with Azure Open AI for most recent updates, I think it was on August 7, where they released an announcement about bringing own data as you're opening our capabilities is now can be infused with Power Virtual Agents, and that's really kind of a domain. Specific is intense, specific as well, whereas personal assistants it could be anything. It could be literally from like booking an Uber to what's the weather for tomorrow, which makes it much more hard into kind of, I suppose, grounding the models to be in that particular mode. And one of the things about generative AI models is that you have to have metaprumps. Basically, metaprumps are a way of making sure that the behavior of responses are made in one particular way. You have to have those metaprumps in place beforehand to provide an answer. That's kind of the user's intent. Now, if you try and do that with something like a personal assistant, you have to have loads of different metaprumps in a way to be able to answer that. And I think that's one of the reasons why it's kind of difficult, because generative AI in general any models is not an AI that was built from previous AI. For example, if you have a recommendation engine, that's a particular set of data set in particular set of data that was used to learn for the AI models to become AI models, whereas LLM large language models they don't actually teach the models what is right and what is wrong. The purpose of the AI is to generate answers, generate contents, and that's what the purpose is, which makes it much more harder to control, and that's why we have, like at Microsoft, these grounding and we have these metaprumps in place in our products beforehand to make sure that it's behaving in a way that's intended to behave. And that makes it really hard, and I know I'm getting a little bit kind of technical and deeper here, but I think that's one of the reasons why it makes it really hard for personal assistants to be using these large language models.

Mark Smith: 32:09
I would say Final question, because we're out of time where are you going to invest your time personally, do you think, over the next 12 months in your learning development? What is Taiki going to do based on your observations of the world you're living in at the moment?

Taiki Yoshida: 32:26
So I think, as a technical person, one of the key things that would be differentiating the top performers and those who aren't is to definitely use the AI capability generally AI capabilities to do more in pretty much every aspect. So for me, to understand and learn all these different examples of how people are using a genitive AI capabilities is something that I'm definitely going to pursue in my career as well in the next 12 months. How can I make myself efficient? How can I do more using these capabilities as well? So that's definitely something that I'm going to be particularly focused on too, and even in work one view as well I'm going to be trying to create content, create different topics and assets for other people to be able to use these capabilities to and infuse it in the business. So I don't know if you've seen some of the blog posts recently, but one of my most two recent blog posts I wrote it about I just talked about today here in Apocos as well, about Mia's story, about Ashley's story, and they both have kind of clear examples of how people is using it, and that's kind of my passion in both personal side and business side of things, I would say.

Mark Smith: 33:50
Hey, thanks for listening. I'm your host business application MVP, mark Smith, otherwise known as the NZ365 guy. If there's a guest you'd like to see on the show from Microsoft, please message me on LinkedIn. If you want to be a supporter of the show, please check out buymerecoffeecom forward slash. Nz365 guy, how will you create on the power platform today? Ciao?