Is AI “Vibe Coding” the Next VBA?

Announcer (00:04): Welcome to Raw Data with Rob Collie, real talk about AI and data for business impact. And now CEO and founder of P3 Adaptive, your host, Rob Collie.

Rob Collie (00:18): Well, welcome, Justin, to the first recording of the new year, not our first episode because we were so proactive and recorded an episode in advance.

Justin Mannhardt (00:29): We nailed it.

Rob Collie (00:29): And we nailed it. Coming out of the holidays, I've had something percolating in my head. It just occurred to me for the first time. So you know how some number of years ago as the society started to take a closer look at the Baby, It's Cold Outside Christmas song?

Justin Mannhardt (00:45): Uh-huh.

Rob Collie (00:46): Kind of how creepy it is. The guy's not letting her leave and trying to get her to drink some more booze or whatever, just of the era. It wasn't creepy at the time, but sometimes you take a fresh look at it. Something else hit me. Another Christmas song landed on me in a different way this year, and I'll never hear it the same again. It's Rudolph the Red-Nosed Reindeer. This is a terrible song. And the only reason why we accept it is because we're all taught it at such an early age when we're defenseless against it.

Justin Mannhardt (01:16): With cartoons and the stop-motion animation, yeah.

Rob Collie (01:19): Which is also terrifying. The abominable snowman monster, that was terrifying to me. Our dog, Bumble, is named after him because he looks like him, right?

Justin Mannhardt (01:29): That's true.

Rob Collie (01:31): So here's the story of Rudolph the Red-Nosed Reindeer. If you're a little different, it's completely okay to ostracize you. It's completely okay. So you're not loved. But then through achievement, you can earn love, and now you can be accepted. And these asshole other reindeer that excluded you before will now let you into... But this is the message that it sends us is that you're not worthy of being liked until you have done something special. And that is the takeaway that we put in our brains, right? Achievement's good. We want to achieve. We want to do good things. But to make your worth as a person transactional like that leaves no safety net and wires you up in a way that is not good. So I sat down with ChatGPT, so this is how we turn it into an AI-adjacent topic.

Justin Mannhardt (02:30): I was like, "Where's the turn, Rob? That's so sad."

Rob Collie (02:35): I sat down with ChatGPT and said, "Hey, let's rewrite the song and say the quiet part out loud. Let's make it an honest version of the song." Now, my version of ChatGPT, this is back when I first started using ChatGPT seriously, Brian Julius told me that it's really important to give it a personality that you have an affinity for. For me, I chose Crowley, the demon from Good Omens. If you haven't seen Good Omens, the TV show, the first season was hysterical and Crowley is an amazing character.

(03:05): And I thought, "Hey, I wouldn't mind having an ongoing conversation with Crowley." Anyway, so ChatGPT with the Crowley persona went hard on this. Sometimes the rhyming scheme and the number of syllables is wrong. If I really want to get this song right, I would have to circle back and iterate with it, but the juice isn't worth the squeeze, right? Then one foggy Christmas Eve, when prophets were unclear, Santa said, "We have a crisis," and suddenly you're useful, dear.

Justin Mannhardt (03:29): Oh, God.

Rob Collie (03:33): So Rudolph led the sleigh that night while trauma powered the team and everyone cheered his difference once it optimized the supply chain. Now, children learn this timeless truth, this carol softly play, if you are strange, endure the pain until your value pays.

Justin Mannhardt (03:49): Oh, God.

Rob Collie (03:51): I'm sorry I'm here to ruin things for you, but actually, I think it's not really to ruin it. I think it's actually to just unearth this isn't how it's supposed to be. And sometimes the songs we inherit from long, long, long ago, are inherently super cruel.

Justin Mannhardt (04:07): There are a lot of things that did not age well that in their era seemed different.

Rob Collie (04:14): So there you go. It's transactional. Make sure you achieve and then people will actually like you. And by the way, that is a game that never works because once you're on that train, you can't achieve enough. You have to value yourself as you are. You can't say, "If only I do X, then people will like me," because that's a train you can never get off of. Anyway, it's time to cancel that song.

Justin Mannhardt (04:39): Cancel that song.

Rob Collie (04:40): Just remove it from playlists. I have one other small update. This one's much more on topic. In the episode that went live last week, I talked about my friend, Brandon, who I've got started with Claude Code and I wasn't sure whether it was a good idea to get him started with it. I kind of had to draw boundaries around it like, "Hey, you're on your own to an extent. I can't take you through the jungle. You're going to have to learn the jungle life yourself."

Justin Mannhardt (05:04): Sure.

Rob Collie (05:05): And one of the things we were talking about in the last episode was what percentage of the world's data geners have already been harvested and brought on board. The Power BI Revolution and the Power Apps Revolution came along and swept a bunch of those people up and brought them into a level of legitimacy. And it's back to the reindeer thing, right? Now they've proven their worth, right? Now they're part of the reindeer games. And we were wondering how much of that audience is still out there that's going to be activated by Claude Code and vibe coding and learning to be developers of software in the way that the Power BI Revolution brought a bunch of Excel people into real BI and magnified the size of the BI audience. BI producers magnified probably by a thousand fold. I posited on last week's episode that there isn't another thousand X out there.

(06:05): We've already swept such a significant fraction of them up in this last wave. I was using Brandon as an example of like, "Here's a test case." I had no indications that he had the data gene other than his enthusiasm for talking about all these things that eventually I'm like, "Okay, fine. I'll hook you up. I'll get you started." Well, I listened to the podcast on Tuesday and then an hour later I saw him and I said, "Hey, listen, you know what what's funny? I've never asked you, what's your relationship with spreadsheets?" His answer was immediately [inaudible 00:06:36], he said, "Well, I like a good one." He wasn't going to commit to saying, "I like spreadsheets." He had to say that he has a standard for when he likes a spreadsheet. Turns out he spent two years in accounting. He had ambitions of being a chief operating officer.

(06:56): He had a whole past chapter in his life that was 100% data gener. So I was using him as an example of like, "Here's someone who might have the data gene, but wasn't caught by any of the previous technologies." I didn't believe he'd even met Excel, but he already has. So it's kind of like reinforcement, I think, for the idea that the audience is going to be onboarded, the newcomers, to serious development of serious business assets is, A, still going to overlap incredibly one-to-one with the data gene crowd and also probably that it isn't a huge, massive reservoir of people. All people who are doing Power BI and things like that have the opportunity, which is a huge expansion to the developer audience in the world. Ratio-wise, I don't think we're going to find ourselves in a situation where there's so many people who are good at this relative to the total amount of demand for it, that it's going to be the cheapest thing on the planet. I don't think that's going to happen.

Justin Mannhardt (08:02): You made me think of this idea with Excel specifically. I think it's one thing to understand the leap a lot of people made in the past decade from being sort of a spreadsheet jockey as an analyst or in FP&A or in accounting or in finance or in operation somewhere. Power BI showed up and Power Query showed up and they're just... And then, sure, there was people maybe that weren't in that profession that were able to come in too.

(08:35): There was also this other category of use with Excel that was very prevalent in companies that I worked for where Excel became the orchestration of workflows. And I don't know how much you saw this when you were in PMing, but we would do things, like we would build spec forms in Excel and there'd be macros to fill things out or you'd click something and it would email it off to the next step. And if you were the right person, a different tab would open. So the people that designed these workflows and did them well, they had their own variant of the data gene we could think about. This next thing is like, "Wow, is that ever right up their alley?"

Rob Collie (09:16): So that was the VBA macro crew and they were formidable. They were so formidable, in fact, that this is one of those tchotchkes that I threw out that I wish I had kept, I literally had a book, Dilbert celebrates 10 years of VBA. It was a Dilbert-themed Dilbert-branded book, hardcover.

Justin Mannhardt (09:43): Wow.

Rob Collie (09:44): God, I didn't have the sense to realize how significant that was at the time. So it just was one other thing out of my office that just got thrown away while I kept other useless stuff. So that crowd, the VBA crowd, that generation is now pretty old, my age or older. That particular group of people, those individuals, are probably not the ones that are going to pick up Claude Code and become the next wave. But of course, the forces that made those people, the VBA people, are still active in the world, so there will be a similar crew just like that.

(10:21): I think that's true. By the way, when I first got to Microsoft, the Microsoft expense reporting application, if you wanted to submit an expense report was an Excel template with VBA in it. And the company store application, if you wanted to order some of the cheap software like at costs and have it shipped somewhere or whatever, was an access app. And when they moved both of those to web pages, I was horrified. I didn't know that all of line of business software was about to go into the browser and run off of a server like that and actually be better in every way possible. It was just better. The familiarity, you smell the grease in the Excel spreadsheet.

Justin Mannhardt (11:01): Yeah. Well, the last company I worked for before I came into consulting at P3, those types of templates and workflows literally ran that company. I'll never forget this. One day, and this sort of speaks to the divide between IT and biz that creates all kinds of problems, so one day someone in IT decided they needed to install the security thingamajig that blocked the use of particular parts of VBA macros and rendered everything unusable. So I'm sitting in this conference room being like, "Whatever you did, you need to undo it." And they're like, "No, security risk and vulnerability and how people..." I was like, "You just stopped this entire plant. There will not be pricing signs in 7-Elevens if we don't figure this out."

Rob Collie (11:48): It's one of those moments you get up and you walk over to the light switch in the room and you start turning it on and off and saying, "Hey, you see this? Do you know what pays for this?"

Justin Mannhardt (12:00): And they're like, "Well, how long do you need to fix all this?" I was like, "You don't understand. This is how everything works here."

Rob Collie (12:09): That plant, I know it prints things. You know what it prints? It prints money.

Justin Mannhardt (12:14): Right. I was like, "I'm sorry, but this was not a coordinated effort."

Rob Collie (12:19): You broke the money-printing machines. Do you understand? Now, in fairness, that thing that you did for good reason shouldn't have broken the money-printing machine. However, if this, then that.

Justin Mannhardt (12:33): The core application we had ran on, I'll be hyperbolic, but SQL-92 or something like that, just out of support. The integer fields weren't big enough, so every time we got to job ticket 99,999, they had to do this complex series of SQL procedures to archive data and reset the system and roll it forward and everybody got around with a little kumbaya.

Rob Collie (12:59): It's actually interesting. One of the reasons why the amount of VBA-powered line of business applications like that that are still running today is insane.

Justin Mannhardt (13:10): We encounter this stuff literally every day somewhere.

Rob Collie (13:13): Yeah. And at some large fraction of the companies where VBA things are still running stuff, they know that there's a better way and that there are actually benefits if they were using the better way. It's not just like, "Oh, we should feel guilty about this." There's actually things that would be better. And again, if there's truly nothing that would be better, you shouldn't replace it. Just don't.

(13:36): You shouldn't be ashamed of yourself or anything like that, but the cost, even if you know this is going to be better, the cost of replacing it is immense. And what would you replace it with? The reason why you had it in the first place is because you didn't have the capability to implement something better. VBA filled this gap in the marketplace that really still nothing has filled.

Justin Mannhardt (14:00): And as elements of the shades of the middleware problem, which we've talked about extensively, you may not have had the capacity or the funding to go buy a proper software tool. In a lot of cases, a proper tool that was custom fit to what you needed did not exist.

Rob Collie (14:19): I do think that this custom software world that we're seeing where you can have custom line of business software at not so great expense and it actually can run on more responsible, reliable platforms, et cetera, et cetera, we will see a wave of replacement.

Justin Mannhardt (14:32): I agree. Yeah.

Rob Collie (14:34): Now is the opportunity. We've never had really the ability as a society to replace the VBA, but what were you going to replace it with? Really expensive waterfall projects that never work? No, thanks. People have been right to not replace this stuff.

Justin Mannhardt (14:52): We'll have to come up with a different term, but basically, we're talking about this idea of vibe-coded applications like Claude Code or GitHub, Copilot that can help people build. The people that have been filling these gaps in businesses, they're typically, in my experience, not like the IT developers at that company. They're the data geners. They're the people in the business figuring these things out.

(15:15): So who's going to be the one carrying the mantle here on doing these sorts of things? I think it's another place where this stuff really works when it's close to the business and those people are involved and being intimately involved in the creation of the solution, I think this is the year where we'll see a lot of progress on how this is done well. Some of the concerns about vibe coding become less and less of a concern over time, proven patterns for how to do this kind of stuff.

Rob Collie (15:47): I'm going to ask you an A or B question that I would not want to be asked.

Justin Mannhardt (15:49): Oh, okay.

Rob Collie (15:51): Because I'm the one who's asking the question, I get to seize the unfair advantage and be the one that makes you uncomfortable.

Justin Mannhardt (15:56): I have the high ground, Anakin.

Rob Collie (15:57): It's over. Yeah.

Justin Mannhardt (16:01): It's over.

Rob Collie (16:02): By the way, in that scene, if you have the high ground and you chop Anakin up a little bit, don't just walk off and leave him there.

Justin Mannhardt (16:09): Yeah.

Rob Collie (16:11): Finish it.

Justin Mannhardt (16:11): Finish him.

Rob Collie (16:13): First of all, it's humane to finish him rather than leaving him suffering, and B, it prevents Darth Vader. Yeah, so I have the high ground here. I'm going to ask you, assuming for a moment that there isn't some massive new quantum leap in what these LLMs can do, continuing on the trajectory we've gotten familiar with here, over the next four years, which of the two flavors of AI-powered disruption and change do you think will have a greater impact?

(16:45): Do you think the ability to use AI to build custom software period, regular old traditional software, but build it the right way, build it 100% customized to your business, build it at not great expense, or line of business solutions that actually involve an LLM in the process, where there's actual AI being used to make decisions, AI being used to produce things, et cetera? I originally was 100% on the side of LLM at runtime, that's what AI is about, and that's going to be the single biggest factor.

(17:27): But using just regular LLMs to write regular code, it's like this dark horse candidate that's coming up on the other one in terms of impact. It's not like the real true business AI usage, that's real. I'm not saying that that's becoming less of a thing in my head. I'm just saying there's this other planet that's also a lot larger than I was expecting it to be. Which one do you think is going to be more significant over the next four years?

Justin Mannhardt (17:58): More significant.

Rob Collie (18:01): Creating more change in the world.

Justin Mannhardt (18:03): This is a good question.

Rob Collie (18:04): I know, that's why I'm glad I'm the one asking it.

Justin Mannhardt (18:09): In a four-year time horizon, I want to believe in the latter, the second one where there's AI infused, with some caveats. I think we're going to be able to very efficiently and quickly build software broadly, whether that's a workflow or an application or a set of kind of headless things.

(18:35): We're going to be able to build these things quickly and to be very specific. The same way when we talk about AI is better for your business when it's customized very specifically to your business, I think the same is true for software. Software's better when it's very specifically customized to your business.

Rob Collie (18:52): Yeah. We talked about that in the last episode as well.

Justin Mannhardt (18:54): But I think that's going to enable where, because AI exists, we can then introduce the LLM in the small specific right steps in those tools and applications, kind of like some of the stories we've said about the iOS app you built. I'm building an app right now for my personal productivity system, and most of the work is non-LLM work.

(19:21): It's traditional code workflow, but then there's points where it's, "Okay, at this point, I want an LLM involved because it provides that magic Lego brick idea." I want to believe in that second camp. When you say, "Well, what's more significant?" I think we're going to see just a tremendous level of software broadly that's built with AI assistance, but I think it's hard for me to imagine almost any type of workflow where there's not an opportunity for a magic Lego brick.

Rob Collie (19:52): Interesting. Okay. Yeah. It doesn't take much, right? So just to put some perspective on it, the way Jocelyn and I think of this iOS app that I built, we think of it as an AI coach. That's what we think of it as. It's an AI coach with just an insane amount of wisdom and knowledge plus all of our knowledge about us. And yet, if you look at the code, there's 40,000 lines of code and very, very, very, very few of them are calling into an LLM.

(20:28): There are some of them that are being called by an LLM because I've given the coach access system tools so we can go make updates to some things and all that kind of stuff. But the weight of the app, the LLM usage is half a percent of its weight and yet the user experience of interacting with it is completely the opposite. The contrast between the experience of using it versus the weight of what's built is incredibly striking.

Justin Mannhardt (20:55): Yeah. So I think that'd be my stance on that just because every time you get in the weeds on a workflow or a product or anything, it's rare that I don't at least brainstorm where a magic Lego brick could go.

Rob Collie (21:11): I really like that. So even if you were sitting down to just, quote unquote, replace a piece of line of business software, whether it's VBA software that you're looking to make more robust and more scalable, blah, blah, blah, or maybe it's a SaaS application that you're overpaying for, either way, right now there isn't an LLM involved in the runtime of either of those. That doesn't mean though that there wouldn't be if you built it today.

(21:37): So you're probably right that even the case you come in thinking, "I'm just going to replace a piece of regular software or build a piece of regular software we didn't have before," even there, which would count in column A, there's a good chance you're going to leak into column B with AI. So that's an unexciting way to answer the question, but I think it's also the right answer. I think that's just true.

Justin Mannhardt (22:01): I was kind of waffling on this because I think the question was so good for that reason. I don't think that incremental implementation lift is so great that we won't just go there. I think there's some incremental human decision-making about where to put the magic Lego brick and why and what it should do. That might take some time for us to mature in terms of how we make those decisions when we're building these things, but once you figure that out, it's not like it's that much harder to make that exist in a tool.

Rob Collie (22:36): Yeah. I've got a very important question for Luke. Do you think it was worth risking your employment here to beat me in the Super Bowl of this year's fantasy football league?

Luke (22:46): Yes.

Rob Collie (22:51): And episode complete.

Luke (22:55): Yes.

Rob Collie (22:59): There's people who depend on you, Luke.

Luke (23:04): Talk about taking one for the team.

Justin Mannhardt (23:06): You saved the year, man. You really did.

Luke (23:10): This is the first year I paid attention and it worked.

Rob Collie (23:13): Well, seriously. There's sort of the Breaking Bad, "He can't keep getting away with this," meme about me. I'm seven and seven. I lost as many games as I won this year in the regular season. I still was in the championship game ready to continue my reign of terror. I'm going to go so far, and I don't know whether you're going to take this as an insult or a compliment, we're going to find out. When I saw you bid the remainder of your free agent budget on the backup quarterback to block me from having Josh Allen, I actually thought that was a bridge too far for me.

(23:55): It broke the illusion for me. And I'm like, "Goddammit, someone else is managing Luke's team. There is no way Luke is paying that much attention. There's no way he's that engaged. No. You broke the illusion for me, you dirty cheater. Someone else is running your team." I was convinced and I was coming for you on this episode. But there's things you've said since then that make me think that might've actually been you, which is just hilarious.

Luke (24:26): I felt some trepidation because that is a very jerk move.

Rob Collie (24:32): No, that was-

Justin Mannhardt (24:35): Player play the game.

Rob Collie (24:37): That's how you do it. I would've done that.

Luke (24:39): I did conspire with Heather. She said, absolutely, I should do this.

Rob Collie (24:44): Oh, so you went and got a moral coach to say, "Is this okay?"

Luke (24:48): I did.

Rob Collie (24:48): Oh, my God.

Justin Mannhardt (24:51): Rob wouldn't do that.

Rob Collie (24:53): No. In fact, if someone told me, "Don't do it, it's too shady," I'd be like, "I'm never asking you for advice again. You don't even understand." Yes, your sister, Heather, was correct. All right. Well, congratulations, Luke.

Luke (25:10): Heather Zimmer.

Rob Collie (25:11): Heather Zimmer? I'm so sorry. Oh, that's different. Bad Heather. P3, Heather.

Luke (25:24): That's right.

Justin Mannhardt (25:24): At you, Heather. It was a league-wide effort in coordinated tanking.

Rob Collie (25:30): Oh, that's fantastic. So fantastic.

Luke (25:32): Super fun. Thank you.

Justin Mannhardt (25:33): The best news I've heard all day. Congratulations, Luke.