What AI Can’t Fix (And What You Still Have to Do Yourself)

Mitch: Hey. Welcome back to Make Others Successful, a podcast where we share insights, stories, and strategies to help you build a better workplace and get more out of your Microsoft subscription. My name is Mitch and I am joining here just to start off our conversation and kinda set the stage. I was actually not in this episode. We left it just for Matt and Mike to have sort of a an old married couple chat, we'll call it, and they initially were going to record the whole thing and they they were gonna do the intro too.

It was a little bit entertaining, so we left a couple of little clips in there for you to hear them coming up with the intro and whatnot. So hope you enjoy those. As far as the conversation today goes, it is the one about AI. So it's the concept of will AI take my job? Is it going to help me?

What should I be thinking about this over the next couple years? And really just zooming out on that conversation and helping you understand where you fit in AI's journey and where AI fits in your journey. And hopefully can put some kind of concepts to and and help you think about things in a slightly different way that helps you look at it as an opportunity and less of an overwhelming unknown thing. Let's jump into it. Here's the conversation with Matt and Mike.

Mike: If it gets a different intro, I'm gonna be offended. Alright. We are we rolling?

Matt: Okay. I'm terrible at the the intro.

Mike: Yeah. I can I can do the intro? But as Mitch would say, I'm Welcome to Probably others the terrible.

Matt: Make Others Successful, a podcast where we help talk about something and something and something and something. I got the first part of it. Welcome to Make Others Successful, a podcast about that part I got.

Mike: I'm just gonna intro it. It'll be alright. Hey, everybody. Welcome to the Make Others Successful podcast, a podcast where we talk about technology and strategy to help make you successful so that you can make others successful and on down the line down the line. This is episode 47.

This one is just kinda like a fireside chat with Matt and Mike, something like that. We're gonna talk a lot a little bit about AI. The the topic today is what AI can't fix and what you might still have to do yourself.

Matt: Yeah.

Mike: Right? Alright. So let's get right into it. So right now, there's a lot of distraction with AI. Microsoft's talking about Copilot almost nonstop.

It's virtually

Matt: Everybody's talking about Copilot.

Mike: The only thing they're pushing whenever I log into any of my any other customers, tenants, and I go navigate around their world, I'm seeing new and different Copilot things every day.

Matt: We use ActiveCampaign and AI is a big, like, prominent when you log in. Hey, we've got this new AI feature. Every newsletter I see is all about the AI, like AI, AI, AI. So, are people getting distracted? What are some common things that are distracting people?

And what's the truths behind those statements and and things that people are talking about?

Mike: Sure. So there's a lot of hype. Yep. And we're gonna try and kinda dispel some of the myths involved in that hype and and talk about like where the rubber meets the road, like how we can really use AI, how we can be successful with it, how we can use it in our in our daily life.

Matt: Yep.

Mike: Cool. Makes sense. Let's jump right in. One of the first myths that's getting a lot of hype, especially lately, like literally in the last couple weeks, we're, what, middle July here, is AI is gonna take your job.

Matt: Yep. There's a lot of fear mongering. People are getting fearful themselves and behaving in weird ways because they're afraid that AI is gonna take their job. There is truth to the fact that AI is gonna help. If AI didn't help and automate things, it wouldn't be a thing.

Right? Like, it wouldn't we wouldn't be talking about it if it didn't if it wasn't able to automate some things. But the reality is AI is only so good. It can only do so much. It's really good.

So for example, in the communication collaboration space, it's really good at editing content, at refining content, at gathering some rough drafts. It doesn't replace somebody who's an expert at the content. It doesn't replace somebody who is creating original thought. Right? People who are doing that today, they're not gonna lose their jobs.

They're it's not gonna go away. Right? Mhmm. What will happen is the editing is going to be easier. They won't need to send it to an editor or they won't need to spend so much time editing and so they'll have more time to do other things.

That's the reality.

Mike: When I think about building apps and how AI can be used in building apps, is there gonna be a robot that sits at my desk and writes the code? No. Can I use an AI assist to write some code? Absolutely, I can. Is AI going to ultimately, like, figure out the purpose or the why behind the thing that needs to be built?

Probably not.

Matt: Probably not.

Mike: Right? And that's still important. That's the, like, that's why we still need a human involved in that process.

Matt: That's also one of the reasons that Microsoft named their tool what they named it, Copilot. Right? It's not something that's gonna just Yeah.

Mike: Not pilot.

Matt: Yeah. Yep.

Mike: Yeah. Actually, when we come up with our own AI product, we'll call it pilot.

Matt: Pilot. Yep. And then

Mike: everyone everyone will because this

Matt: will be the best one. It'll actually do everything.

Mike: If Matt if it's Matt and Mike, you know it's gonna be the best. Yeah. Right. Excellent. Alright.

So our next statement, our hype statement is AI requires a prompt engineer to be effective. So that doesn't mean an engineer that's on time.

Matt: No. It means, like, there were literally job postings for people who their only skill was prompting AI. They knew how to prompt AI really, really well. They didn't have necessarily the subject matter expertise or these other things, but they're they're experts at prompting AI. And, you know, my feedback on that is that is that that was a fad.

It's like, it it's is it gonna be a a category, a skill that's required to get a job? Yeah. A 100%. Mhmm. But that skill is overshadowed entirely by your expertise in the subject matter.

Right? That is gonna be the primary thing.

Mike: Do you

Matt: do you know about coding? Right? Like if we're talking coding. Mhmm. Do you know about communications?

Do you know about, you know, the topic that you're that you're trying to get a job to do? That's gonna be the overarching thing that's most important. Prompting is gonna be high up there, but this concept that you what's most important is that you know all the secret code to prompt appropriately. That's just not true.

Mike: So it's very interesting. I I connect this very strongly with, like, in our field, what we've learned over the years has been helpful to us, which is we've become very good at keyword searches using Google, Bing, whatever your search engine is. Mhmm. We've figured out how to enter the right phrases so that we find the best response in the quickest possible Yeah. Manner.

And a prompt engineer is simply someone who is adept at creating prompts that would return similar results. And one of the things that I see going on is with AI getting better, you don't need to be an engineer anymore.

Matt: It's not a specific Right.

Mike: The AI has gotten better so that those especially of us who if you're any good at all at keyword searches, for example, just think for five minutes about what the question is that you're trying to ask. And that domain knowledge that you have about whatever the field is that you're in or whatever the thing is that you're researching is the thing that's gonna enable you to build that prompt, assemble assemble that prompt or keyword search.

Matt: I would contest that learning to prompt is a more trainable skill than being able to search using Google and using some of these other things. You know, yes, you can train on that a little bit. You know, you can get but it's, I think even, this is even more trainable because a lot of it comes down to, it's a conversation. Have a conversation. Give it the context.

Have these things, right? And that's another piece of it is in search, you put in that first search and it's like a one time shot. Maybe you got what you wanted, maybe what you didn't. To refine it, have to go back and remake the search. But the engine doesn't really know that, like, you're refining an old search.

It just knows you're making a new search. Whereas this is a chat. You're having a conversation, you can ask it, and it makes it more intuitive. Like, for me, prompting is a skill that is way more trainable than other technologies out there and how to use them.

Mike: I think there's another advantage that I just recognize that prompting and use use of AI has over search engines and that is that refinement piece. Yeah. And it's something like, when we do a search, we may be looking to solve a particular problem and we actually don't know the right keyword yet.

Matt: We don't.

Mike: Don't know what a particular feature is called of a given product. Yep. And then we do a search that gets us close and we read some of the articles and we digest that. We go, that's what that thing's called. Yep.

And then we do our next search, which looks for that thing. And I think AI has the ability to look into that those results and kind of discern what it is that you're asking for. Pick those things out, and then you can have a conversation with it Yep. And get deeper faster. I think that's really, really cool.

Matt: And I think that's why this is a bit of a fad, people are worried about silly things in this regard, right? Prompting is really important. It's really, really important. There's no doubt about it in my mind that if you're a knowledge worker, if you're a coder, a developer, understanding what that means and how to prompt is really important. But what's more important is to be a subject matter expert because I can train someone to be decent at AI, and all you really need is to be decent.

Right? Mhmm. You don't need to be an expert, s tier, top of the capabilities around prompting. You just need to be able to be effective.

Mike: Yep. Very good. So that kind of takes us right into our next statement, which is AI is always wrong. So Yeah. In speaking about the expertise, the subject matter expert, right, there's this, like, don't want to use AI because it's giving me the wrong answer all the time.

Matt: Yeah. There's a bunch of different nuances to this that are very interesting. I've heard a bunch of organizations who are like, if I if AI can't be a 100 accurate, we shouldn't use it at all. Or it's always wrong so I and I can never trust it so I don't ever want to use it. Or it was wrong and led me down a bad path and so then, you know, I I'm mistrusting it, etcetera.

AI isn't always wrong. If it was, nobody would be using it. It is wrong sometimes. It's why that subject matter expertise is so important. It's why that's the most valuable thing because you need to be able to fact check, gut check the results coming back.

But quite frankly, if it was always wrong, there wouldn't be value. And there's lots of studies, tons of studies that are out there today about the value that AI is bringing. Obviously, another thing that a lot of people don't understand is, like, ChatGPT, Copilot, they're going to be wrong more often than a targeted, focused AI experience. So that's why a lot of people are talking about agent or agentic AI. Right?

The the deal is Copilot and ChatGPT are broad. They're very broad, open ended. It's up to you, the prompt person, to create a prompt that puts all the the gates and controls, etcetera, on your prompt, an agent comes with some stuff built in, right? And that stuff that's built in is going to give people that have a niche question a better experience than if they were to use that broad AI. And so doing that is gonna make the experience even better.

Right? Saying, hey, don't look at all the internet, just look at this piece of the internet or look at this piece of my intranet in the case of Copilot. Right? That can really improve those results, get you better feedback, allow you to have a much better experience. And so what AI are you talking about?

You talking about a generic AI that you put somebody that has no concept of how to prompt in front of to ask a question? Yeah, it's not surprising. You're probably gonna get something incorrect. Are you asking it something that's really complicated that's not well documented out in the world? It that's gonna be the place that it fails the most.

Right?

Mike: Right.

Matt: Where it has to make a bunch of inferences and and assessments about what this content is, you know. It's all about how you're using it and there's tons of scenarios where it's gonna work beautifully.

Mike: Right. So don't ask ChatGPT where the vacation request form is. Yeah. Right?

Matt: That won't work.

Mike: Yeah. Right. So you need a more narrowly focused thing, something that's maybe an HR agent

Matt: Yeah.

Mike: Inside of your organization, which is something that you could build and scope appropriately so that it would be useful for your end users. Yeah. Which takes us right into the next thing, which is AI and existing tools. Yeah. Right?

So there are there are a lot of tools, particularly in m three sixty five. We talked about ChatGPT. There are a lot of tools out there. The features of Copilot are now coming right into some of the tools that we're familiar with, even things like Word and Excel. So if you've ever struggled with building formulas in Excel, guess what?

Copilot is now there to aid. Among other things, Power Platform, Power Automate, Power Apps, all of these different places. Let's

Matt: talk So about is AI gonna replace all those tools?

Mike: That is somebody's question, I'm sure.

Matt: Yeah. You answer What do you think?

Mike: I don't think so. I don't think it's gonna replace the tools per se. I also don't think it's gonna replace the people that engineer the tools into what they need to do. So if you think the scenario that you were just talking about where you need to narrow the subject matter, put some boundaries around it so that you get better answers, for example, someone still needs to do that. The content that feeds that, let's Needs say to live somewhere.

HR site, for example, needs to exist somewhere. It just can't divine it out of ether.

Matt: The other thing is that AI isn't really good at generating information in a format that is best suited to communicate the information for a very specific topic in a consistent way. Yeah. And so you're still going to want that thing to direct you to the source of truth. You know, you want it to have a repository, a source that it's pulling that information from, which is still going be in SharePoint or it's gonna be a Word document, or it's gonna be a PDF, or it's gonna be a PowerPoint presentation. When you think about Microsoft's approach to it, they, what they want to do is put Copilot and AI on the front end.

So they want Copilot to be helping you create and generate and manage and, you know, leverage all of these things. But, you know, they're really still gonna be dealing with all of these different tools and platforms that are the core content repositories content So

Mike: what I just heard in that is Copilot is gonna help me create the content that I put on my Internet, author pages

Matt: Yep.

Mike: For example. And then Copilot might also help me discern, review, analyze, investigate, interrogate that same content

Matt: Yep.

Mike: In order to answer questions for my end users. Yep. And maybe just take them to the page Yep. Right, for for more detail or whatever

Matt: that is. That's their vision. That's that's what they're seeing this this going in the future.

Mike: So Intranets are still gonna be a thing. They're not gonna be replaced.

Matt: Yeah. I I I don't see them being replaced anytime soon. The the organizations that are having the best luck using AI have things like an intranet, have things have systems that have well organized knowledge sources. That's one of the core things that are required to create AI. Right?

AI has to have a repository of information that it's pulling from. And if you don't have that well organized, if you don't have that in a system that can be accessed, you're kind of in trouble. If you've just got a shared file or shared shared drive with 5,000 documents in it, in a folder structure, it's going to be a challenge for you to For sure. Leverage AI.

Mike: Let's talk a little bit about kind of the other side. We talked a little bit about communication, collaboration. Let's talk about apps and automation. And so we all we know now that there's AI assist basically in all of these tools. Yeah.

So if you need to build an expression that maybe you don't understand how to write yourself already or maybe front and center, frankly, you open up Power Apps, it's like, let's get started today. Tell me what you want to build. Yeah. And you prompt it and it like, voila, builds an app. Yeah.

So what do we what do we feel about the quality of that experience and what you get out of that?

Matt: So it's really interesting. I will take, I'll even expand it out more and talk even Excel and Word and these other things. There's, there are vastly different experiences between these different tools and how well those things work, right? It specifically on the apps and automation side, I haven't had a great experience. Like, it's not been transformative.

I can tell you custom dev, you know, I'm writing C Sharp code. I'm writing PowerShell code. I'm and I'm in Visual Studio or Visual Studio code. Amazing. Like, Copilot and AI is like, it's awesome.

Like, it can really accelerate what you're doing. It can give you ideas. You still have to be a professional developer. You still have to be able to review it. You still have to be able to put your stamp of approval and go, yep, yep, that does what I want it to do.

But it's amazing. Power Automate, Power Apps, like, you know, I have not had the same level of experience. Like, it's it's amazing how how different those experiences are.

Mike: Yeah. So I think there's, like, and I can speak a little bit to the Power Apps and the the no code, low code side of it, but I think there's a very specific, like, quantifiable difference in what you were talking about where it is used successfully and the no code, low code, and that is the volume of code that has to be written. Mhmm. So in your what you were talking about is like, I gotta write a bunch of code, like thousands of lines of code maybe. That takes a lot of time no matter how good a developer you are.

Mhmm. Yes, there are generation tools and things like that, but that's still a lot of time to review all of the code after you've got it, all that kind of stuff. And then on the no code, low code side of things, the code or the expressions that you're called to write or that you're involved in writing are usually pretty short. They're like similar to an Excel expression. Right?

Maybe two, three lines of code is pretty common at the most. And so one of the things that that I've learned in that experience is in the time it takes me to construct, to engineer a prompt that will give me the expression that I want, I could have written that expression three times over.

Matt: Yeah. Because it's not very good right now at bigger structure things. Like, it's usually the best cases that I found to use it in is in a context where you're in one expression. And in one expression, you're usually not doing something like search for this thing in my array of whatever. Right?

Like, that's not what you're doing. No.

Mike: You're gonna do And a concatenate these two

Matt: and or, well, the overall flow has a section, maybe a scope, that does what I'm talking about. Yeah. But it doesn't do a very good job of generating that. Right. Like, you just talk about generating all the actions and it just doesn't

Mike: Even not as if you think about the number of key presses it takes to write the prompt Yeah. Versus the number of key presses it if you once you're fluent in the expressions and the functions and whatever that is, like, get there. Like, maybe use Copilot to get started and learn those things. But once you understand what they are and you become fluent in those functions and Mental issues. Language, you might as well just use them and write them out.

It comes out much faster. Yeah. I think the other area that I see on the particularly on the Power Apps and Power Automate side is there's a push to just like build your whole app or start your app or create your flow initially with Copilot and let it author it for you. Yeah. And my experience there on the Power Automate side is it will build a shell, a structure overflow that's roughly there.

And then you have to go in and finish up the expressions and make it actually work, connect it to data Yep. Have it do things. And then on the app side, I've been pretty disappointed because once you understand all of the deeper details of something like the Power Platform and Dataverse and all of the features that are there, business process flows and auditing records and all that kind of stuff, there are certain things that you wanna turn on and enable and structure a certain way. And no matter how good your prompt is You won't cannot understand it to that level yet.

Matt: So that's one of the things I was gonna say is that one of the problems that I think a lot of, especially more advanced or seasoned developers have is even if it produces something that is correct, it's not the way I would do it. Like, my

Mike: Describe standard your standard.

Matt: I mean, that's the that is half the problem, is that most people don't describe it. They don't have it written down. It's not like you have a

Mike: Mhmm.

Matt: Here's my code code guideline coding guidelines, like, which we have in C Sharp Java, etcetera. Even though I say that, we have them. Absolutely. We do have a naming convention that we all prefer Yep. For actions and for whatever, right?

And it's using its own. And you can probably tell it what you want, a little bit, but now you have to add more to your prompt. The use of scopes and the use of, like, there's just so many little things that some of it's personal, like signature style stuff, some of it is standards and practices that we agree as a team to use, but it can also just make it feel not great. Yeah. Even if it is, does everything you want.

It's perfect. It actually works exactly the way you asked it to work. It could still not feel great. And I would say the same thing with the other things, like to bring it back, like, some of the other tools. I can see the same thing with Excel, and I can see the same thing with Word.

Like, there

Mike: is

Matt: a gap still missing. It's way better than what it used to be. I do agree with you that I think the apps and automation side is the worst. Like, I think they've spent a lot of time on it, but it's still, at least in its current state, not great. But Word and Excel and PowerPoint still are not the best.

When I think about creating, honestly, you know, in Versus Code, that's like,

Mike: You're it's a making everybody want to live in that world. I mean, it is

Matt: amazing. There's no other way I can describe it. Compared to, comparatively, right? Like it's not, again, it has its problems. Again, you have to be an expert.

Again, you can make huge mistakes by letting it just go run wild. But the ability for it to get really close to what you want and easily just go tweak a couple things and then it work is like

Mike: Well, what's also very interesting is the the Versus Code as an editor also, like, once you have a volume of code written and you have to Yeah. You have you have to tweak something Yeah. It has all of the built in connections to say, I Yeah. Wanna rename this and change it everywhere. Right?

And so it it, like, all that's

Matt: And can

Mike: and you can ask to

Matt: work across your entire code base, not just on one spot. How can you do that in a solution? Refactor. Where you where can you in a solution come up with Copilot and say, hey, I need to do this with my project, and I need this and this and this and this and this. Go make all my changes, and then see what the changes are so that you don't just go, oh, I hope I had worked.

And then easily be able to make tweaks as you need to. That's just there. There

Mike: are some limited limited areas and we're like going a little bit long on this right now, but there are some limited areas like in Power Apps, for example, in Canvas apps, you rename something Yeah. It will see that elsewhere and go rename it for Yeah. But it's not all the time and it's not perfect. And so there's always that challenge. Yeah.

Cool.

Matt: Cool. So tools aren't gonna go away. AI isn't just gonna replace all of our tools.

Mike: And neither are our jobs. Yeah. I knew I needed something to do tomorrow.

Matt: So we

Mike: gotta come to work. Alright. Well, I'll be here. So let's wrap this up.

Matt: Yeah. So we went through a bunch of things that I'm sure everybody's heard when they are in conversations about AI. Hopefully, we've helped kind of bring some sanity to the conversation. A lot of these things are pretty assertive, clickbaity, you know, aggressive statements. Hopefully, what we say makes sense.

You know, obviously, it's what we've experienced, what we've seen. And hopefully it helps you kind of make sense of what's going on.

Mike: Yeah. So Copilot, other AI tools, they're helpers.

Matt: Yep.

Mike: Not magic pills. Yep. Which by the way, AI's been around a long time. Like, I don't

Matt: A long time.

Mike: You've been using a Maps app on your phone. Right?

Matt: Yeah. Yep.

Mike: To like tell you where to go. That's AI, y'all. Yep. If you didn't know it. So it's been around.

It's been helping us for a long time. It's gonna continue to do so. But it's not gonna take over. It's not gonna replace humans.

Matt: Not anytime soon. Right. It's gotta get way better.

Mike: And my belief my one of my strongest beliefs is human connection, human relationship is the most valuable thing in the universe, by the way. That's Get

Matt: up by your soapbox.

Mike: You bet. That's not gonna go away.

Matt: And with that

Mike: Well, wait a minute. Before we go, in case you're you're following us and you're interested in how you might be able to use these tools, how you might understand things like Copilot and some of the other tools that we talked about a little bit better. And if you're new to this world, we have a three sixty five foundations course that's intended to help you and your team kinda get aligned with those tools and get a kind of a big picture mindset and approach when you're

Matt: Best way to start.

Mike: Good way to start. Yep. Cost effective, comprehensive, good. Yep. Alright.

Well, thanks Matt for your time today.

Matt: Was Yeah.

Mike: Pleasure talking to you as always and we'll do this again next month.

Matt: Yeah. Bye everybody.

Mitch: Hey. Thanks for tuning in to Make Others Successful. If you enjoyed this episode, we'd love to hear from you. There's a couple ways you can do that. One, we'd love for you to rate our show on your favorite podcasting app.

And then if you have feedback or topic ideas or suggestions for future episodes, head over to bulb.digital/feedback and let us know. Your input is super valuable and we'd love to hear from you. Don't forget to subscribe and share this podcast with others who care about building a better workplace. Until next time, keep making others successful. I'll see you.

Mike: I don't know what my bee if my bees are gonna make it another year. Those two hives that I broke out to try to get, cause I had new queens Yeah. Being made, I think they pooped out on me. Again? So I gave them each a new frame of eggs

Matt: Yeah.

Mike: To see if they'll do it again while I'm gone. And then in the process of doing that, I opened up one of my other hives, my big one that I was taking those out of, and that hive's getting a little bit pissy with me when I did that because three days earlier when I opened it up, then I'm digging into the nest and I pulled off the top brood box because I had been checking frames in there, pull off the top brood box and I looked and I'm like, I think that's a queen laying there dead. Oh. And I pulled her out and I looked at her, and she had another bee or two, like, attending to her because she's, like, still kicking a leg or something like that. And I like, man, I might have just killed the queen, which happens sometimes.

But I wasn't sure because she's kinda small. So I don't know for sure, but then the three days later after that happened, I got in there and they were like not happy bees, which told me that they probably didn't have a queen. I didn't see any signs of their building a new queen yet, but I think they're gonna do probably the same thing while I'm gone.