Hi, I'm Paul Ford. And I'm Rich Ciani. And this is Reckless, R-E-Q-L-E-S-S, the podcast about how AI is changing the world of software. It's brought to you by the... amazing software AI empowered business app development platform aboard launching soon, launching soon. Damn it. We're about to see a good demo right after this meeting. Aren't you ready? It's worth sharing with everyone that if you've wanted to see our faces, we're not sure why.
But you can now. Our full video of us staring into cameras is on YouTube. So you can find us there. We're also going to make some YouTube shorts that are going to go... Just go viral. We're kind of limping in the light here. By the time you hear this podcast, it might be like four hours before it's up on YouTube. But what happened is the people who are really good at marketing have taken us aside and said...
Listen, YouTube's the big platform, and I hate it. You don't mind. You don't mind people looking at your big shiny head. I don't want to do it. You know, it's not that I don't mind. I just don't see the people. I can talk to a room of 2,000 because it could be three. I feel like a small but overweight dog. Standing in front of a jury of 100,000 people who are just very disappointed with me because I peed on the floor. You're at the Westminster Dog Show. That's what I am.
That's how I feel. All right. Let's get a little more attitude here, Paul. All right. Well, welcome. Let's play that theme music. All right. Good, good, good. So Rich, you were going to ask me what I've been working on. Well, I think it's worth to take a minute and talk about how we work so differently with this stuff these days. Okay, this stuff. When you say this stuff, you mean... Just grappling with how to...
understand AI, leverage AI, and go to market with the thing that people want, right? We're both kind of aiming for the same thing. You tend to open a development environment and play. You just start dancing. You're freestyle jazz. Jazz music. My way into the tech industry was I wanted to make publishing platforms. I built them on my own 20 years ago.
And then I realized what content management was. And then I turned that into a consulting job. And then I went and built content management tools for magazines. And so that's my way. My way of learning is just to jump in. Yeah, yeah. And by the way, I like to jump in too. But what I tend to do is step back out and think about sort of the business context. So I'm not in there like...
you know, wearing an engineer's hat like you are. I think it's worth saying. I'm just not. And I think that benefits us. I did go in because I wanted to see where the edges were, but then I came back out. How is it going? Are you having a good time? Well, I mean, honestly, our business has gotten really busy, so I have a little less time to mess with this stuff than I did. But, you know, there's another aspect. And actually, before we move on from that subject.
You did some research recently and you found something important, which is how much of the software development process is actually writing code? Less than 50%. 45%. According to some, I remember, I can't remember where you found them. It was a meta study. It was actually a pretty big study. It was like 40 to 50. So I just took the number 45. But it's, I mean, less than half. What's the rest? That's a big deal. Project management.
Product management, design, UI, UX design. If it's games, model design. actors uh motion graph uh 3d motion capture of like you know special effects and stuff like gaming it's even smaller it's interestingly because it's like movie production if it's a serious game so um DevOps, data modeling, data architect, security. But actually, to put the last part of the conversation into context, right?
I go in and I start with code and it does, now look, it's 45%. It's the biggest chunk. It is the biggest chunk. It's really amazing that a bot can, you can give it words and it will produce anything at all. Like it'll make code. Oh my God, we're in the future. But when software is like that plus... 20 other roles plus a lot of processes and then users actually using it and it iterates and changes from there and i don't like that's where we're working right now and i think like
It's complicated. It's complex and there's a lot of different skills, right? It's sort of like when it draws you a picture of like a bear and you're like, wow, drew me a picture of a bear. It's not really a good picture of a bear. It's wild that it's there, but it just kind of sucks. Right now, ChatGPT rolled out a new version, and everybody can make Studio Ghibli-style images. Like, oh, my God, it's Totoro, but it's Andrew Cuomo. Oh, whoa. And Cuomo-ro. And, okay.
You know, but the thing is, is they all suck and they're really generic. And you've seen by the minute you've seen one, you've seen them all like they're really interesting for about five minutes. You're like, that's wacky. And then by minute 10, you're like, uh-huh.
Yeah. It gets pretty predictable pretty fast. It's actually fascinating. We're getting really tired of the joke. Yeah, yeah, yeah. I think everybody's getting a little tired of the joke, except for Sam Altman, who's really into it. So... So anyway, okay, that's another giant conversation. So what I've been working on recently, remember we had Clay Shirky on the podcast? Yeah. Love Clay. Clay can just go. Gotta have him back on.
Right in there with us. I love Clay. And so Clay has this database that he works with. It's called iPads, which is a terrible name for a data product. And every time I type it in. Claude has stopped me because it thinks it might be pedophilia content, but it's called iPads. I'm wrestling with that. iPads is a database. that colleges have to put all the data into the box about gender breakdown and what courses offered. Oh, I see. So it's a statistical database that like gathers demographic data.
across universities for one giant blob of information so we can tell number of faculty number like not just demographic like a whole lot of information links to all the mission statements got it this is a shared this is an openly available data set That is funded by, I'm guessing, public sector money. At least so far, right? And so, like, and if you're somebody like...
Clay, you're really interested in this data set because it helps you. You're at a really big university and it helps you understand where higher ed is headed. Sure. Think bigger thoughts. But. The data has a web interface, and it's all very downloadable and freely available, and it's filled out. It's mandatory that colleges fill this out. I guess probably that's how you keep your accreditation. So it's a really good data set, but it literally was started.
in 1981 and it you can feel it like it's the column names are very arbitrary they're only eight characters wide like as far as you and it's it's only available in microsoft access formats i think like 700 megs so It's not a huge amount of data, but it is a blob of data. It's very well documented. There is an Excel file that describes every field and what's in it.
But, you know, it's like really hard to work with as raw data. You can use their web interface. But if you just kind of wanted to do your own thing, mapping. Put a lot of time in. Scrub the data. normalize lookup so it's not looking at codes, all of it. All that. So I've now worked on this for maybe like 12 hours. And I've made, frankly, immense progress. Like I've made six weeks of... When you say working on this, you're trying to come up with a clean, usable...
interface to traverse the information. Very good question. I'm going to get really nerdy for a minute and then you ask me questions to translate it into civilian talk. Okay. Okay. Okay. I'm taking the access database and I'm using Claude. Okay, so it's Microsoft Access Database. Yes. And I am feeding it. And so I used Claude to write a tool that uses some open source tools to turn that into a SQLite database.
which is much more tractable. I can say, so that's a different format, but it's much easier to script and work with from the command line. And then from there, I've been using Claude to look at samples of the data, look at the data dictionary. and create a more usable version of this data set. A more human readable version.
But still very programmary. Like everything is named with underscores, but it's legible. Like a civilian would recognize it. Okay. And you did this because you're just a good friend. You're just a good guy. This is an ugly migration problem. It's just ugly enough. It's a good way to test out.
the the capabilities of these incredible new tools well and it's also our we're building a product that accelerates software development but what happens and we were just talking about this what happens after the software exists Got to get the data in. Yeah. Nobody shows up. Very, very few people show up empty-handed when it comes to business software. Yep.
Sometimes they do, but very often. And I think that's another aspect of these tools. It's like they kind of don't have a migration story. Yeah. They're just like, oh, you're going to build a new app. And it's like, well, what's going to go in the app? Stuff. Yeah. Okay. So.
I'm now actually at a block because the data dictionary is so large that in order to make the tables readable, because there's hundreds and hundreds and hundreds of columns, I actually can't get Claude to look at the whole thing because it's too big. So there's a context. So now I'm working on a way to I have to get it to write code.
I'm going to make a database of the column names, have Claude write code to rename all the column names based on a query to the database. Like, we're in Kooky Town, but... I do feel that given probably another 12 hours, I'll get to a really good place. Hmm. So you nerded out for a minute here. You know what's really nice about this? I had Clay give me a couple of prompts or a couple of queries.
And a lot of the things that we'd like to do with this, like writing the SQL queries around, like, you know. geographic location is hard. I've done it before. And here, if I give it the schema, I'm going to be able to get it to write the queries and I'll be able to do things with this database that really, really would have been like hours and hours of work in minutes. And that...
Like, there's a good target here. Anyway, so that's where I am right now. And you know what that makes me... You know what that is if we're going to draw a larger lesson from it? You ready? Yeah. AI's getting a little boring. That's where you were taking me. Because I was thinking to myself, I just started to see listeners dropping off as you were talking through this. Like a bad CNN poll during a debate. And I was wondering where you were going with this. And I think...
I think you're right. What about all the spectacular headlines about robots that can dance? If we take a breath, first of all, we've got to be clear.
boring is good in technology i remember once we were working on a project and there was a client who was not a real technically aware person and i went to a meeting with this person and i said something along the lines of i promise you the platform we're building is really good and boring and she felt i was referring to the design and the interface and that i was
promising to deliver crap. And what I was actually promising to deliver was stability and long-term maintainability. It's a little language difference between the two worlds. And she reamed me out and I had to go to Cipriani and get yelled at. at for an hour and a half because that's what you get to do in client services. Well, this is how you grow as a client services person, Paul. Never said that again. I'll tell you that. All right. But boring is good.
ChatGPT just rolled out Studio Ghibli and photorealistic bears and whatever the hell, right? But if you really take a step back compared to when that first version of all this stuff showed up and it just... felt like the computer had come alive in a new way like it did when it went from windowing interface or mobile in your hand like just this new modality i've never seen before it's all incremental now
And they keep promising AGI. Honeymoon's over? Is that kind of what you're saying? We're adding new features to a database. It's very comprehensible. It's getting more and more comprehensible. It's less like... Boy, we don't you know, because when this first came out, everybody was like, wow, we don't really know how it works. And we kind of do now like it's layers. And yeah, there is opacity in the way that it scrunches the data. But
You can learn how this works. You can figure out all the pieces. And increasingly, you can build predictable systems, not necessarily on top of LLMs. It's really hard to make them perfectly predictable. You can do the same kind of heuristic processes that you do when you're bringing in real world data.
And, you know, analyzing plain text and speech, you can do those to get stuff into a classic computer model where people are moving cards around on a screen or assigning tasks to each other or all of that stuff. There's all this G-Wiz and all this novelty, but I'm telling you, let me ask you, and we can pressure test this really easily. Fast forward six months. Fast forward a year. Is Sam Altman saying...
Hey, amazing news. ChatGPT can now replace your operating system. You don't need Windows anymore. No. He's not. Why not? Because... Well, I think it goes back to – well, first off, I said that very confidently. I don't know anything. I don't know what's cooking on the 46th floor of OpenAI. But – I want them. It's actually vegan bean burgers. Yeah. I mean, obviously. Right. Yeah. I mean, look, I think when you say it's getting really boring.
What you're really saying is eventually these things peter out and you're kind of on your own with a toolbox. And the same old challenges kick in. And what you end up finding is. And also at that point, Richard, like the maximum amount of market value has been extracted from the novelty. Yes. Like, IPOs are happening, and now we're all left. It's like waking up after the honeymoon. It's just like, oh, I guess we're married. I don't want to discount the fact that when you did hit the wall...
You got to the wall in like three days instead of five months. Like the fact that you even got to the wall. Now you're at the wall. The thing that if there's one thing to take away and if there's one thing our graphics design team can use. for the cover of this podcast, it's this. Right, because we have to make YouTube covers now. It's like a whole thing. You can't dip back into the magic bucket. It's weird. When you dip back in...
It's sort of like glitter. It's like glitter you buy from Michael's art supplies. The magic is real. But when you go really far off to the edges of your journey and then you're like, I need more magic. I got to get this across the line. It falls. And you're left in a very lonely place. Well, no, I wrote about it. Essentially called Microsoft Access Studio. I wrote about it in the newsletter. And I'm like, this part right here is just programming.
Exactly. Exactly. Now, let me ask you. Now, here's the thing. Here's where it's really good. And a lot of people have given this advice. Simon Willison has given this advice, which is when you are in no man's land, we should name this place that AI got you. And then left you there. Well, it's sort of the DMZ between human computer and... Yeah, and there's... You're going to be okay for a while because for some reason AI dropped a power bar in your pocket.
So you can kind of keep going and you're kind of in this weird place. And when you're in that weird place, the way to use AI is to. build you little micro tools to make little steps forward. If you go back in and say you got it wrong, you didn't finish the job, go back and finish the job. The thing about AI is like, you're right, I got it wrong.
Let's try that again. But this time it'll be in closure. Like it'll just, yeah. Closure is a language by the way, for people like this is, this is vibe coding. Vibe coding is like, ah, you know, I'll hit it with the same hammer. See what happens. Yeah. And the truth is you're technical. You understand the smaller steps you should take. You know which little functions and little components you need to bring to the forefront to get moving again.
That is not trivial. You're talking about this like, hey, you know, I'm kind of vibing through it. You're not. You're 30 years of very thoughtful technical thinking. that is now being brought to the forefront, essentially, the wheel has been handed back over to you. That is just the reality of it. Is that boring?
I think it's interesting that we're having this conversation while everyone else is talking about how they built businesses in 45 days with AI and how AI will eventually break into our homes and take our children. Like that is...
That is the stranger conversation. And the truth is. I'm a little tired of that conversation. I mean, I just, I think, and I'm not alone. I think people are getting a little bored. You're hearing it less, by the way. I think you're hearing that conversation a little bit less now. What I think is really. Interesting. You always say technology is about skipping steps, right? Trot that one out, old Richard.
and what you just said was really interesting because i think it's like hey it lets you skip a bunch of steps it uses a plain language interface produces a bunch of stuff and lets you skip a bunch of steps but you need to actually get better at identifying what steps It doesn't let you skip. And then you've got to kind of craft, get in there, and you have to figure out how to get from A to B. That's right. Let me ask you a question. Okay.
You asked me a future state question. Let me ask you a future state. Will you be able to type in, what is it called? Pedobase? iPads. Okay, iPads. Will I be able to type this prompt? Sorry, and by the way, it's the integrated post-secondary education data system. So it just couldn't be worse. That's really cool. It's got a nice website. They're good. You can go access it. No, they're doing their thing. All right. Ready? Here's the prompt. Build me a web-based iPads.
browser that masks away all the technical aspects of it and glossary mapping and all that. And makes it human readable, excuse me. And human usable for non-technical people. And like, that's my prompt. That's your prompt. And, oh, okay. Okay, in a year. I don't think so. Two years.
Oh, that's a forget the two year question. I'll tell you what. Here's what actually has to happen for that. I don't I don't think you get these one shot solutions without a lot of subsystems. So when you describe that, what I immediately go to is one like Claude or. Anthropic or OpenAI or Microsoft or whoever, you can't get that out of one company, out of one solution. I don't think we're there. So they have to go acquire all those little layers because...
It's about layers. It's about traversing through different sort of ways of understanding stuff and extracting. Yeah, but even gluing those layers together is not trivial. It's not like you can throw them all in the box. What is the outside chance in two years? It could be like... Maybe through acquisitions, a set of problems like that could fall to a prompt. I would not. I will absolutely place very big bets. And, in fact, that's what I'm doing with you right now. I am placing a big bet that.
development can be accelerated and in the hands of many, many more people, 10x, 5x, whatever, some x, sometimes 100x. And then you can build tools and use those tools much more quickly than before.
I absolutely believe some of that's happening. We've proved out some of it. We've worked with people doing things for them, and I'm 100% convinced that that's coming. But the thing you just described... I almost think it's a meaningless question, not because what you said was meaningless, but because there's so many little abstract things and little bits you have to tinker with in order to even understand what that means, what you just said.
you will get an opinion it will write you a report or generate a bunch of code in two years but the odds that it's actually what you need or want or that it's the right thing for your user or so on and so forth are really really low it might work
But it might not, it's just, there's no, there's very little chance that it'll be the thing you want unless you go back and iterate on the prompt and get it just right. And again, now you start to, that's going to be what programming is. Do you, so, okay. All right. So you're not seeing AI really eat away at like.
You don't see how it can bring it all together is what I'm hearing. There's so many little problems along the way. It has to venture out and... and deal with them all right i think a more subtle point and then you know we should probably we should probably wind this one up but that's not how humans are when humans can't wrap anything up ever right if
If you had told me five years ago, I'll be able to tell a computer to make me a Studio Ghibli image from any photograph, I'd be like, ah, that sounds really like an impressive product. and now there's a glut of that trash everywhere and everybody's angry about it or excited about it or whatever but it kind of doesn't mean anything right like the next thing will be can it make me you know somebody's gonna make like Totoro but it'll be Goodfellas yeah
right and then you know joe pesci will be in it and we'll be watching that we'll be like oh my god there is no end game with this stuff there's no like one shot solution because the minute i come up with the thing you just described somebody's going to look at it, including me, and go, wow, it'd be really interesting if I could put this kind of query in. I don't really like the way the mapping works. And then the product isn't done anymore.
Because it's a human artifact. You don't trust done and the understanding of done. Nothing has ever been done. The only thing that gets something done, what is the one technology that completes a task? What? A deadline. Yes. A commitment to someone else. Yeah. Yeah. It's another kind of deadline. Yeah. Yeah. It could be a customer. It could be a boss. It could be whatever.
I got a busy early next week. Okay. But I should finish this thing and share it with clay and put it on GitHub and just sort of like make it a project and be done with it. Yeah. Okay. So I'm going to, I'm going to let you do something I never let you do. Oh, no. I'm going to let you give me a deadline. All right. Tuesday. I just said next week early is busy. You know I'm actually presenting something to you I've been working on on Tuesday. Oh, yeah, that's right. Friday.
Friday's good. Friday's really good. Friday. Friday, I have a working SQLite implementation that I can walk you through. Man, we could have a podcast on deadlines, but... Let's spare the masses. Because for all the good of AI, it's actually not getting it done. There you go. Neither am I. Yeah.
All right. I needed that. All right. From Friday. Let's see how I do. It was nice to see your face on video, Paul Ford. Oh, my God. And we're going to have to learn. Like, I have to learn to look at the camera and I'm staring. You have to. I've seen. I've watched a bunch of podcasts. Nobody looks at the camera.
They're all like looking around. I'd like to make something good that doesn't look like two Fliberty gibbets hanging out in their apartment. But that's actually what this is right now. All right, well. Maybe we'll just add some videos to it and call it a day Richard is great talking to you in nine minutes. We get to go to another meeting
Always great. Give us five stars. Give us thumbs ups. If it's thumbs, give us thumbs. If it's stars, give us stars. Yeah, you know what? It's time. We're on all the podcasting platforms and we're on YouTube. And this was just faces talking. talking heads but eventually we'll have a lot it'll be a lot more visual and interactive we want to we want to dial it up it'll it's gonna be fun
Well, we'll even have more guests. You'll see their heads. And check us out at aboard.com. Sign up for the newsletter. You'll get notifications and early invites on the very cool future cut of the platform. Aboard.com. Yeah, that's true. We'd love to get some beta testers nearby. All right. Have a wonderful week. Bye. Bye.