Why Most AI Projects Are Solving the Wrong Problem From Day One

Samuel

Ryan Cunningham is the corporate vice president of Copilot Studio and Power Platform at Microsoft. Agents, apps and automation across all the Microsoft stack roll up to him. We get into why most AI projects stall before they even start. Whether apps and Power Platform still matter now that you can build almost anything from a chat window. How to actually choose between pro code, low code, and vibe coding. And how Copilot Studio credit-based pricing is changing how people think about cost. Quick thanks to Casper, a Microsoft partner, for sponsoring this episode. If you want to hear how to move from having AI tools deployed to actually operationalizing them inside your business processes, check out my episode with Eric Murray from Casper, link in the description. And before we jump in, I have a small favor to ask. If you're getting value from this show, hitting subscribe, dropping a comment, or leaving a like is one of the best ways you can support me. It helps more people find these conversations, and honestly, it means the world. Thank you. Now, let's jump in. Ryan, welcome to the AI Frontier Playbook. Ryan, you're the CVP of Copilot Studio and Power Platform at Microsoft. So, agents, apps, automation, all of it rolls up to you. I talk to customer every week. We're trying to figure out where all of this is going. And today I get to put their questions to the person actually building it. So I'm really happy to have you here.

Ryan Cunningham

Hey, happy to be here having conversations. I will say I too talk to customers every week trying to figure out where all this is going. It's a rapidly moving market, rapidly changing customer expectations. It is a fun time to be alive. So happy to have the conversation today.

Samuel

So let's jump in directly. So something I hear constantly is that people are stuck before they even start. That's something that's coming out a lot while discussing with my customers: Copilot, Copilot Studio, Power Automate, Azure, M365. You kind of have no idea which door to walk through. So when a leader comes to you and says, "Okay, just tell me where to start." What are the first couple of questions you want them to answer?

Ryan Cunningham

Well, I mean I think there's two versions of this problem, right? One version of this problem is what people are really asking is not what tool should I use but how do I think about a good AI program in my company like what do I actually want to accomplish? I would say most interesting AI use cases right now have nothing to do with what technology you start at or whether you're building a chatbot or an automation. The interesting AI use cases today are what needle are you going to move? What change are you going to affect in your business? What does it mean to have dramatically more productive employees or dramatically more effective business processes? And how do you make that really specific? I met a customer several months ago who said, "Here's my AI use case today. It takes me 36 hours to generate a quote. I want it to be 2 hours by June." That is a beautiful AI use case for a couple of reasons. One, she didn't know or care what the technology solution was yet, right? She was coming with a really clear idea of ROI upfront, right? So, it wasn't go show me a fancy tool and then postrationalize what it's worth. It was I know exactly what this is worth. Help me go achieve it and get it done. Right? There was also a very clear ambition and constraint loaded into that statement. The ambition is 36 hours to two hours. You don't get that just by pushing people to work a little harder. You have to fundamentally rethink how a process works. You have to rip a lot of steps out of it. You have to automate a lot of things. That's a full process transformation by June means I can't afford to go hire an army of consultants to make me a multi-million dollar PowerPoint over a year's time. I have to get started tomorrow. Like, I have to move fast, right? And so that already creates the constraint of not overthinking the problem. Probably needing to involve technical people and process experts. That means I probably need a higher level of abstraction, right? I'm already starting to put the boundaries around the type of solution that I need. And so I think the very first conversation I have with every customer is just, do what you're really trying to get out of this? If what you're trying to get out of this is ship a chatbot because my boss told me we need to do an AI thing, go back to the drawing board. You're not ready. Like go have a better go come back with a better use case that has a clearer understanding of what you're trying to achieve. And all of the interesting ones are structured in that ambition constraint mentality. Then the question becomes okay what is the right set of situation and leadership and structure we need to put in place to make it successful and then what tools should those people be using? And that comes back to the second more actionable question which is hey I want to run an automation that automatically extracts information from incoming requests for quotes and routes them to the right people and processes them and then automatically enters now we can start talking the right tool for the job and that's where actually it's a huge asset to be in a company that has a very robust platform and toolkit that all works under the same governance structure and kind of general framework right because if I'm going to a point solution vendor that's best-in-class at one thing, then their hammer makes everything look like a nail, right? If I'm using a platform, I get the chance to use the right tool for the job and compose it in a better way. But you're only ready to have that conversation when you have a richer grounding in what it is we're trying to accomplish in the first place. Where do we need a balance of personal productivity versus long-running process automation? Where do we need a balance of very low-level control in the hands of professionals versus the agility of a very highly abstracted platform? And most interesting solutions don't just fall in one part of that graph. They start thinking about how they use the whole thing together in the right.

Samuel

Yeah. And I agree. I'm seeing a lot of customer not thinking about their ROI or not even a goal. I was having a conversation with Pam Maynard. She was saying like start by defining a bold goal like your North Star before trying to find tools and use cases and I think the framework based on what you've said is: don't try to adapt the use case to the tool like define your use case define your ROI and then use the right tool which is part of your toolbox or the platform right?

Ryan Cunningham

Yeah, and a lot of these things are people are looking for a they think a use case is a very specific concrete thing: employee onboarding or whatever it is quote response and that there are real concrete examples there. The most interesting things are really outcomes. Like a lot of ways I coach people is like go write your promotion case for next year. What do you want to be able to say, right? Work back from that. And that's not going to have a long list of bullet points of POCs that you tried. It's going to go say, "Hey, I cut quote generation from 36 hours to two hours in 6 months." That's a slam dunk promotion case. Great. What does it need to what do you need to do to make that true? Right? Because at the end of the day, nobody really cares what tool you tried or how many tokens you spent on a coding agent. They ultimately care what came out the other side. How is the business operating different today than it was before. And the cool thing about this moment is we used to have a pretty good understanding of what the constraints on our businesses were and what the constraints on software engineering was. We used to know what was practical to build or automate in a certain time frame or budget. At the end of the day all of the constraints really boil down to time, money, expertise. And so where you get creative is what if I start shifting those constraints? What if I could do something 10 times faster than I could do it before? Something that I used to take me a year, could I do it in six weeks? Right? That's a 10x transformation. Something that used to cost me a million dollars, could I do it for 100k? Something that used to cost me a whole team of experts. Can I do it with two or three people? And how do I start to take that on as an incremental challenge? Those are the really important things to ground on and then upfront what is it worth it for me to risk to make that outcome true, Right? And even if I get halfway there, what's my break-even case? And then what is my upside case? And that starts to design the parameters of a solution even before I start talking about individual technologies or token.

Samuel

Love it. At a smaller scale, that's what I'm doing with my podcast sure, doing a podcast like this on my own is a lot of work, but now using AI, I can accomplish more, but I'm still hiring some people to do some stuff because the ROI with AI or the quality wasn't there. So I just Yeah.

Ryan Cunningham

Yeah.

Samuel

And talking of tools in your toolbox, Power Platform obviously is part of our toolbox at Microsoft. And there's a question I got asked point blank, so I'll put it out to you. Is Power Platform dead? Because AI can basically generate whatever you want and end it back in the chat right now. So do we still need apps and interfaces? And does the platform still have a place?

Ryan Cunningham

I think there are actually two very different questions there. One is are apps dead in general like have user experiences gone away because we can now have different modes of text-based interaction. And then the other question is how are we going to build them going forward? Like are we still going to drag and drop and point-and-click and use deterministic interfaces or is everything going to be coding agent chat-based software generation? I think those are parallel and both important questions everybody needs to ask themselves right now. I would say on the first one, I am very bullish that user experiences are not going away anytime soon. And I think there are some things chat is great for and there are some things that truly can be fully lights out automated, but there are a lot of things still left over, right? And if I really am going to have an agent or team of agents truly run major parts of my business, process a 100,000 transactions a month, go do a 100,000 insurance claims, go manage all of my supply chain there are places where I need to manage that work. I need to interact with it. There will be exceptions where agents get blocked and can't do things. There will be times where they do the wrong thing and need to be corrected. And only some of those interfaces are good in chat. Go show me all the hundred thousand customer issues from the last three months is terrible at text. Like, go give me something to click on. Give me a grid to filter. Like we're not going to completely regress 40 years of user experience innovation just because we now have text in a new way. I think it's really much more that's very monolithic and shortsighted as a as a as a kind of insight. And I think you expand out a little further from that. I think there is the real kind of worry in a lot of our LinkedIn feeds and even in our stock market, which is all SaaS dead? Is you if I can vibe code anything in a weekend, why would I ever pay somebody else for software? And I think that honestly is actually not that interesting of a question to me. I think that question really misses why most people pay for software. The hard and expensive part has never been getting to the first prototype of something. The hard and expensive part is are you building the right thing in the first place? And what happens six weeks from now, 6 months from now, six years from now, right? Like all that custom software that every one of our customer organizations and Microsoft ourselves built over the last 30 years the.NET Core application that a vendor built for us in 2012, did it get its.NET Core upgrade on time? Yeah. Right. Are we are we on the hook to maintain it if it goes down tomorrow? Like and that's why people pay for software, right? And so I think the ability to move faster creating and automating things, the ability to not be locked in silos of point solutions, the ability to not have to have a human come to a set of rectangles and manually type into them for everything. All that is goodness and we should kill a lot of the current sort of world of SaaS and point solutions because it's just inefficient. That does not mean that all software goes away. It does not mean that all human experiences go away any anytime soon. And so then we have to ask ourselves okay so what's the most efficient way to build for that? And I and I would say look there's a lot of things that a coding agent really expands the capability for to do that was never possible before. There are also a lot of things for which the tokens may not be worth it. Like yeah just move that button 20 pixels to the left and make it green is frankly a lot cheaper to do with drag and drop than it is with typing to a reasoning agent that's going to be very resource intensive. And so I think the world is starting to realize that hey, this whole conversation about celebrating token maxing is actually kind of short-sighted. Like we need to be thinking about value maxing and we need to be thinking about how do we start taking what we've already improved and make it incrementally better. You if it ain't broke, don't fix it. But then how do we start to tackle the places that we have not been able to address and the problems we have not been able to traditionally solve? Those are the right places to bring AI into whether it's an agent whether it is an a AI step in a workflow whether it is vibe coding something directly and the beautiful thing about the platform is it's not a binary choice I in a in a point solution startup world either everything has to be vibe coded or nothing because that's all they've got right in a on a Power Platform stack we can have a much more mature conversation about what needs to move and what needs to stay what is not solved on the current stack, what can be solved with the current stack. And look, we're going to get that wrong in certain places. And for all of the reasons of a hype cycle around technology, but the more we can hold ourselves accountable as a community to just doing the right thing for customers and for the process, the better. And that's why having this sort of grounding of what value are you trying to achieve is so important going in and what are you willing to risk on it? Because if you don't have that barometer, you don't know where the tokens are worth it or not. You don't know where doing it with a traditional workflow versus an agent is going to be the right thing or not. And so you have to have that grounding if you're going to make the right choice about solving a problem.

Samuel

I think the token cost is just starting to raise in the public discussion, right? This is not something people were worried about six months ago, but now it's everywhere in the in the news on LinkedIn.

Ryan Cunningham

Well and there's a lot of macro factors driving that one is frontier models have made a major leap forward in the last six months and the architecture behind them how we run them in a hosted container or virtual machine with a set of tools is also very resource intensive and so the types of things that you can accomplish on the frontier now have dramatically expanded. The problem is the cost to do them has also dramatically increased and therefore is no longer practical to fit into simple dollar per user per month licenses at great margins and this is not just a Microsoft thing in fact the whole industry is like if you look at Claude from Anthropic, if you look at Google Gemini if you look at OpenAI all of them are moving to this world but it does mean we have to be thoughtful now like where is it worth it to use those tools because they solve problems that I couldn't solve before and where is it not worth it? And I will say there are plenty of examples on both sides of that in this moment right now particularly while that capacity is scarce and it will be scarce for a while. We're in a we're in a place where the ability to expand literal GPUs in data centers in geolocations is constrained by physics. And so that means in any place where supply is constrained and demand is high price goes up. That's just a core basic economic principle. And we will be in those conditions for a while and we as an industry need to just make thoughtful choices. Again, that's why having a robust platform of many different tools in the toolkit is really valuable for customers to think through.

Samuel

Building on that, you mentioned that now users have a bunch of choices, right? You can go no code, low code, you can go vibe coding, you can be full pro code, you can go spec driven. So I think most people still just default to whatever they already know. I'm I'm talking every week with developers who just want to continue the way they were doing with fully pro code no vibe coding seeing people who have been working on the Power Platform for years that still want to be full Power Platform. So how do you actually decide which ones will fit the problem in front of you?

Ryan Cunningham

I mean I think the worst way to make that decision is well this is what I've always done and know, so it just must be the right answer right like everybody in this moment whether you are a career computer scientist software engineer or longtime Power Platform user or first-time vibe coder your most important thing is to just get curious get hands-on get aware of what is out there It is really hard to learn how to swim without getting in the Yeah you cannot learn how to swim by reading stuff on LinkedIn. You just got to get in the water. And so, like the number one thing for a lot of people is be hands-on with this stuff. I am a daily active user of GitHub Copilot CLI and I build stuff all the time just to stay on the edge of what is possible in those places, right? And I think you just at some level you just can't outsource this stuff. You have to have a firsthand understanding and then you have to be really grounded. What are you trying to achieve? So that you can choose the right tool for the job, right? And so I think for a lot of all of us have to stretch outside of our comfort zones in a time like this. And if we're not doing that and we're just using the bias of what we've always done or the bias of what we read about and the conclusion we drew secondhand, we're going to be operating at an information asymmetry and that's ultimately going to be a disadvantage, right? So some of this is just get hands-on and then be practical, right? Just be very practical about where something is worth it and where it is not. I think there there's a lot of bad examples to this. I met a partner who I'll keep anonymous who was very proud to tell me that they had taken a bunch of customers' Power Automate RPA flows and recreated them as agents. I said, well, what was wrong with the RPA flows? Oh, nothing. We just wanted to upgrade them. Well, great. You took something that was working fine and you probably made it more expensive for that customer for at best the same level of functionality and let's be honest, possibly less reliability. So you got if you don't have a really firm understanding of the value you're getting from something and you're just doing it because you want to say that you did it, you're not actually creating any net new value in the world. At the same time, there's a lot of stuff that we have not solved yet, even in the most advanced Power Platform customers because it wasn't practical to do with the current mainstream declarative version of Power Platform. That's the stuff to go tackle. Go tackle the unsolved problems. Don't just re resolve the solved problems. That's the space to really apply AI.

Samuel

Quick break to thank our sponsor, Casper. They're a Canadian technology consulting firm and they've been helping organizations get real business outcomes from technology for 30 years. As a Microsoft partner, they work on modernizing operations, improving productivity, strengthening cyber security, and getting more value out of data through Microsoft technologies, automation, custom app development, and AI. It's really worth a look if you're trying to get more out of your Microsoft stack. Now, let's go back to the conversation. I think people don't like to be told that AI is not the answer to their use case or their challenges. But there's I mean it's not it's not doing everything right. It's very powerful but there's limitations and there's still places where you actually most places where the human is still the way to go.

Ryan Cunningham

Yeah. And look, I think this is also not binary, right? Like a lot of the things people do in their jobs today are repetitive low-value manual tasks that could be a lot better with AI. And if we can compress more of the way we spend all of our time today into more valuable tasks then we can take on a lot of new things right and so we have to have that understanding of what the upside is and that's true for developers it's true for every other operating function in an organization the goal is just be pragmatic about it right and I think that's the important piece you get credit for moving the needle it's not like if you move a needle and only 60% of that move of the needle was calling an LLM that you only get 60% of the credit. You get credit for moving the needle, right? So, you got to go understand where am I going to move the needle? How am I going to use AI to help me move it faster than I would have been able to move it before? And then how am I going to use AI in the actual operating solution itself? Those are all independent decisions and the real shift is that the constraints have moved on how fast you can create and iterate on something and how autonomously that thing can run. And you just have to be really clear-minded about both of those aspects and where you're applying the value.

Samuel

And to your point, I remember I've been doing Power Platform for actually before it was called Power Platform and I remember how tedious it was just to create fields. It would take hours one by one waiting for it to load and then you made a mistake in the schema need to delete it and redo it again. And now we can co-author with AI. I was I was creating a full table yesterday in the Dataverse. It took me 10 minutes. Which probably four years ago would have taken me at least four hours.

Ryan Cunningham

100%. And even four years ago that felt radically faster than doing it the traditional way. So, like we are always compounding from a software development perspective all we've been doing for 40 years is adding layers of abstraction so that we don't have to write low-level assembly code and in many ways coding agents and all of the AI around them is just that next layer of abstraction. And it makes a ton of sense. Now, sometimes we compile to an asset that itself is agentic or uses AI. Sometimes we compile to an asset that is deterministic and declarative and often we compile to a solution now that is a mix of the two.

Samuel

Yeah. And that's super important to think about as we're doing solution design which actually is a good segue to my next question. We want to get into making an app agentic. So going from one that sits there waiting for a click to one that can basically look at your data and do something about it on its own. So yeah, if someone's got a working app today, what's the smallest first step that starts moving it in that direction?

Ryan Cunningham

Yeah. So I think no app exists on an island, right? Every app that's been built in the Power Platform over the last eight years exists for some reason. There's some business process behind that, some goal people are trying to achieve and accomplish. I would say the very first thing is just make sure you actually understand what that goal is or what why does this thing exist? What are you trying to get done? It's not doesn't exist because your boss asked you to build an app. It exists because there's a problem you're trying to solve, a process you're trying to move forward, a decision you're trying to memorialize. Get grounded in the reason and then get really curious. Is this the best way to solve that problem? So, it's like and where do people spend the most time with the least value in this thing? Right? If what do what is the goal we're trying to achieve? And then what is the least efficient part of achieving it that then how could I make it better? Right? Do I really have to do all that manual data entry? Could I use things like form fill or form predict to fill them out automatically? Do I have to wait for a human to come to the app to do that? Could I trigger it on an event? From some other external source. Where do people get stuck in the app? Where do they have questions or need to move a process forward that they can't do today? Maybe I can bring Copilot into Power Apps to help them solve that problem without leaving context. Right? But again, it's you got to be curious about the process itself and what value looks like before where to pull the individual levers and tools. We got a lot of levers and tools. Every Power App now exposes an MCP server that any agent can use the skills of the app. Every app can bring Copilot into it to work with the data with the user. Every app has AI built into it now in certain ways. But you got to understand what do I want those things to do in order to have value. And you can do that then you can really upgrade a lot of things in place without throwing them out without having to abandon that app or recreate it. But think about it as a set of skills and a step and a process and then think about how do I make the process itself better.

Samuel

A good example of that is how we use copilot at Microsoft with Copilot for Sales or the sales agent. I think like CRM still have a place but doing research to find the right opportunity or updating an opportunity is really low value task. I mean that updating it has value, but going into the system, clicking to multiple places to get access to it doesn't have value. So being able to just go into chat and tell it what to do and then it updates the system is very very valuable.

Ryan Cunningham

I use CRM at Microsoft every day and I'm not even a seller. But I haven't been to the front door of the CRM app in months. So like and that's and it's not just chat, right? I have a my team has helped build a set of agents that automatically prepare us for every meeting. It enriches our briefing and notes with information and resources gleaned from that Dataverse MCP server from the underlying tool. We've vibe-coded in our own platform using our vibe coding tools new user experiences that are very opinionated for what do I want to see as an engineering leader when I engage with a customer because the out of the box stuff is not really optimized for that. It's optimized for sellers and sales leaders. And so all the data and logic is incredibly valuable. But the interfaces and the way we interact with it changed dramatically and that's great. And historically it would not have been practical. It would have been possible for us to do all that, but it would have been much more expensive and much slower, right? And so we just dealt with the old way. Now it's much faster and more practical to start innovating around those data sources. And that has unlocked a whole lot of productivity for us.

Samuel

And we recently announced the app MCP. I was totally amazed when I've seen that. So for those who don't know what this is, it's it kind of turns an app into something other agents can reach into like data, forms, actions, everything is callable directly from Copilot Chat and it's rendering his answer as an app in Copilot Chat. It's so impressive.

Ryan Cunningham

Yeah, it just brings the snippet of user experience that you need right in line where you're using it. And that it's super valuable. You if I don't have to go load a full screen browser and switch context, then I just need to see the five fields of the form that I want to fill out based on this last meeting or whatever it is. I should be able to stay in context and I should be able to do my work there. And then of course, it's always one click if I want to go deep and explore something in more depth. So all these things compose. I think it's much more important to go think about how do I want to compose them to make a process better and then and then that can lead to the right tool for the job. But app MCP is super cool and really neat use cases coming out of that space.

Samuel

But when you're architecting those apps do you have to think and build this architecture around the fact that someone might call it into an agent at some point? You need you need to think about it ahead of delivering the app right?

Ryan Cunningham

Yeah. Yes and no. I would say in some ways all this stuff apps as MCP servers apps as micro-experiences are actually going the other way around they're saying look somebody already encoded the details of the business process what it what it takes to qualify a lead is implicitly defined in the fields you need to fill out in a in a form and whether or not they're valid and what happens after you see right and so the business logic the validation logic and user experience you can look at it as yes I built an app that I expect a human to come to but you also can look at it as no I documented my business process in great detail I documented it implicitly in what I said was a valid write to the database what actions would trigger after that what is mandatory to fill out what is optional and so I've already done the hard work of defining the process. The question is now how do I get people through that process in a more efficient way than having to go to a full screen UX and type in all those boxes, right? Because where the human is doing the implicit part of saying, "Oh, I went to a meeting and now I'm going to go enter all this information or I got an email and now I'm going to go update this thing." The human is the API in the integration layer in that sense. And I think that's the type of labor that we can offload from people so that they can do the more interesting work of what should we do in that next meeting? How do we move the conversation forward? What is actually going to solve a customer problem is probably not manual data entry into my into my CRM system or whatever other business app is out there.

Samuel

Interesting. And the next layer to that will be autonomous agents, right? Sure. Yeah. Where is there I was about to say no human in the loop which is not true but where there's less human interaction. So once you've got agents actually running and doing things on their own. The question everyone is asking right now is like who's keeping an eye on all of this? How do you let agents move fast without it turning into a mess nobody can see or control?

Ryan Cunningham

Yeah. Look, I think this is a real conversation to have, but I also think it's not as novel or scary as some people make it out to be. And here's why. The business process outsourcing market is substantially larger by several multiples than the entire business software market today. Most customers are already running core parts of their back office, whether it's finance, accounting, HR IT support, you name it, through third party vendors that sit three, five, nine time zones away, that aren't full-time employees, that they have very little direct personal relationship with, right? And we have built systems for managing the scale of that work at arm's length. And we've built systems for auditing, for setting rules, for complying with those rules, for knowing what happens when something doesn't go the right way. And we can reuse a lot of that infrastructure, right? We understand what it means. Look at things like Agent 365, right? That didn't get all invented from scratch. That was adapted from systems built for a world of very diverse human workforces. Right? How do I do data sensitivity labeling? How do I do Defender type of protections? How do I do auditing and role-based access control and apply it to an agent which has some characteristics of a human user? They have an identity, they log in, they do activity but also has characteristics of software. They operate at much higher scale. They do things over a much broader set of resources sometimes. But it's not a 180 degree shift of a lot of things created from scratch. It's a 10 degree shift of composing a lot of these existing mature capabilities and optimize them for a world of agents as opposed to a world of people and I think that obviously in a world where those two assets work together very very very carefully and so I think that's the really cool part about doing this in Microsoft is again we're not just a point solution startup or lab that's making all this stuff up on the fly we're we're really understanding how businesses tick and operate today and applying a lot of this new thinking into those mature systems.

Samuel

And actually, it brings me to a question about costs because we're spinning up a lot of agents. And Copilot Studio runs on credits, right? So, there's one shared pool for the whole tenant and then it will be burned down by how much your agents are doing rather than how many people are using them, which is a big change in how people thought about costs at Microsoft. People were used to buy a license and that's it. Now we're introducing this credit consumption concept as part of the Power Platform. So let's say someone spins up an agent that will loop a few times and like suddenly there's a bill for tens of thousands of credits and nobody knows exactly where it went. So before anyone builds anything, how should they figure out what a use case is actually going to cost?

Ryan Cunningham

So I would say some of this is technology like we are actively building a lot better things like cost controls and calculators and monitors and stuff like that. And that's all very important work. I see some of this is also just skill and maturity though. Like this whole idea that resources have variable costs and more usage of them will drive a larger bill feels very new and scary to people that have worked in SaaS-based business applications and productivity software. It is absolutely the mainstream world of all cloud software and has been for two decades. So this is not actually a new concept even for Microsoft. We are a tens of billions of dollars of revenue in this exact frame today of businesses that are 100% consumptive and always have been, right? And so, and that's the Azure cloud and all other clouds, right? I think some of this is we there are actually some skills and muscles to build around sort of how not just how do I predict the future with my crystal ball but how do I set budgets? How do I understand what something is worth and what envelope I want to operate to? How do I set good operational rigor over that? Where am I willing to take a risk and what people am I willing to trust to take a risk to try things out and then how do I optimize and create rigor around my spend right and so these two things I think have to evolve in parallel yes the tools themselves need to expose great data and great policies and rules and things like that and they are already climbing those hills but also every Microsoft employee and every customer that has not had to think about this before you got to get in the pool you got to start swimming you got to start thinking about how do I solve these problems? And again, it all comes back to that very first conversation. If you don't have a clear idea of what you're trying to achieve at the end of the day, then it's really hard to know what it's worth to cost it, right? And I would I would much rather come from the perspective of here's what I'm trying to achieve and here's what I'm willing to spend. Now, let's go think about how we design a solution to fit that boundary and create a good governance structure as we operate it to keep it within that boundary. That I am, hey, I'm going to go blindly give capability to a bunch of people and then hope that they do the right thing with it and retroactively come in and try to think about how much will it cost in a year or two or five. That's always going to be a hard problem even when the calculators are twice as good as they are today, right? And so I think you just have to have that clear mindset of ROI up front. That will help a lot in terms of setting constraints on how we operate.

Samuel

Yeah, I often tell my customer if you're worried that much about the cost is maybe your use case is not the right one. Because sure if your use case is right, you will save enough not to have to worry about the cost or improvement in quality of life or retention of your employees or whatever else. But you again it all drills down to or come back to the fact that you need to identify your use case and your goal before starting to build which is not the case of most proofs of concept I'm seeing right now.

Ryan Cunningham

What I'm willing to spend, invest, or risk in cutting customer acquisition cost in half is very different than what I'm willing to spend, invest, or risk in making lunch ordering easier in a certain location. Right? So just having a good rudder and understanding of where we're steering with these initiatives helps a lot downstream in some of these ROI conversations.

Samuel

I think there's a big shift because in the past we were going to a partner asking for a solution that will be built on D365 or Power Apps then they will come back with a quotation and then we know give or take how much it will cost. Now we're going in a world where we have a use case and we're asking a partner to or we're doing it in-house and we don't exactly know where it will land. We don't yet know how we will do it and how much it will cost. So it's a big big change in framing and thinking about how we deploy those tools.

Ryan Cunningham

Right. Yeah. But I think what you're talking about is actually a much broader thing, right? Which is it used to be the hard and expensive part was building or implementing the thing. There are ERP deployments from the SAPs of the world that can take seven years or a decade to implement. So if you're going to take that one, that's an extreme case, but even traditional custom software development, forget Power Platform very expensive, right? And so the onus was really frontloaded on making sure you have the right design and a very rigid plan because it's very expensive to do it and you better be sure you did it right. Now, that is the fundamental constraint that's shifting. If it is an order of magnitude easier to implement something, then actually the risk of implementing something that is slightly wrong goes down dramatically. And the and the value is that you actually have a much better idea of how something is performing after you've created it than when you try to draw pictures about it and predict the future of it. Right? And so for a lot of this, it's how do you compress your time to wrong? Assume that the first version of anything you ship, whether it's an app or an agent or implementation of a CRM system, you're going to have something wrong in there, the faster and cheaper you can get to knowing exactly what that is. And then the faster and cheaper you can get on an improvement train to make it better is the faster you're going to get to really high ROI. And AI data scientists and researchers call that hill climbing, right? How do I get into market and understand my performance against a benchmark and know that the very first time I measure it is probably the worst it will ever be and then how do I start tweaking levers, giving better instructions, upgrading different parameters and climb that hill to higher quality. There's a meta version of that in all of our business process solutions, right? And this is the agile methodology we've been preaching for a generation, but actually finally possible in the type of time cycles that I think a lot of people have have always dreamed of. And so I think this is the really important thing for us to really take to heart and help customers kind of navigate through as as we go.

Samuel

Ryan, it was a very good conversation and honestly, I have so many more questions, but we're almost at the end of our time. So before we wrap up, there's something I ask everyone on the show. If you forget about the platform and the products for a second, can you tell me about you? Like what's one thing you do with AI that makes you more productive every single day?

Ryan Cunningham

Oh man. I do a lot with AI. I am a heavy user of our products and actually I'm a heavy user of competitive products too because again you got to be in the pool, you got to be in the water. And I use these things personally and professionally. I would say one of my favorite things I've been doing recently especially as I've been ramping up in Copilot Studio, I'm relatively new leading the Copilot Studio team is I have a series of agents that now prepare regular and sometimes daily briefings for me in the form of an automated podcast that's personalized to me.

Samuel

Oh, that's great.

Ryan Cunningham

And when I commute in the morning, I have a set of effectively they're just they end up as just Word documents, but that are optimized to be read aloud as I'm driving into work in the morning. And it's been a very rapid way for me to deeply understand issues that otherwise, would have taken a lot of other manual effort. And so, it's been fun to experiment with how do I get lots of skills and tools in there beyond Work IQ is definitely important. Getting access to some of our core telemetry signal and product signal, getting access to public sources of information and private sources of information about what people are saying and doing with the product. And then and then having very specific sets of instructions of what do I want to know out of that has generated hours of incredibly personalized training material just for me has been really really valuable. So I think just on a personal level I think those types of use cases are super interesting and again historically that would have been very cost prohibitive to go have created and now this is something I can I can do in a few hours and move on from. So it's pretty cool actually would have been almost impossible.

Samuel

I just hope that AI won't replace podcasters but let's see.

Ryan Cunningham

Oh no. Let me tell you this. They're not anywhere close to as interesting as a real human conversation, and I would never share them with anybody, but they're very functional in closing a gap and solving...

Samuel

Yeah, that's what I think as well. Yeah 100% last question Ryan you sit right at the front of where all of these things are heading right so you see it before most of us do like looking out over the next few years or even 10 years. How do you think AI and agents are going to reshape the way we build and run a business?

Ryan Cunningham

Yeah. Oh man, I yeah, anybody who makes a prediction about the future that's longer than three or six months is either lying or trying to sell you something. So I think the world is moving fast. But I would say look, the future is already here. It's just not evenly distributed in the classic sci-fi quote, right? Like our own team has dramatically changed how we build software and how we operate in the last six months. And I think every other part of the industry is going to go through that similar shift. My goal with Power Platform and Copilot Studio specifically is to build the abstraction layer that makes it much easier for every other business function, every other customer to make that same operational shift. And I would say the products are never going to be perfect. They're the worst today they will ever be. But a whole lot of transformation is already possible using them and we can attest to it firsthand. And so it's definitely time to get hands-on and get into it if folks are not yet. And that's also the best way that the products improve is through that feedback cycle of just being really hands-on with them.

Samuel

Thanks a ton. Ryan, honestly, it's been an honor to have you on the show. I've been following you for so long. I've been in the Power Platform and D365 ecosystem for more than 15 years now. So I've been following you for some time. It was a very fun and insightful discussion. So thanks a lot for joining us today.

Ryan Cunningham

Thanks for doing what you do and thank you to all your listeners for being part of this community. This is a really special asset in technology and it's really great to have conversations like this. Thank you.

Samuel

Have a great day.

Ryan Cunningham

Cool. Take care.

Samuel

Take care.

Ryan Cunningham

Cheers.

Samuel

Ryan's story about the partner who rebuilt a working RPA flow as an agent and made it more expensive without changing what it actually did is the whole conversation in one sentence. AI is not automatically the upgrade. It only counts if it moves something that matters. Three things I want you to walk away with from this episode. First, define your ambition and your constraint before you touch a single tool. The customer who wanted quote generation to go from 36 hours to two hours by June knew exactly what she wanted and by when. That clarity is what turned a vague idea into something buildable in the first place. Second, get hands-on with the tools yourself instead of forming an opinion from LinkedIn. Ryan still builds things daily with GitHub Copilot CLI just to stay current. If you are only reading about vibe coding or agents, you are working off someone else's secondhand take. Third, set budget and risk tolerance before you scale an agent, not after. Copilot Studio credits work like Azure consumption, not a flat license. Know what an outcome is worth to you before you find out what it costs. If you want more conversation like this one, subscribe to the AI Frontier Playbook whenever you listen and sign up for the AI Frontier Playbook newsletter to stay sharp between episodes. Thank you so much for listening and I'll see you in the next one.