AI Finally Got Real - Ep 001

Welcome to Drive by AI. This is my first attempt at recording a podcast on the morning commute, so expect some background noise.

I've been pretty critical of AI hype for the last year, and most of the 10x productivity claims still don't stack up. But something genuinely shifted recently. The models got smarter. Coding agents like Claude Code matured. The work coming out actually holds up to our standards at Vokke now.

In this episode, I cover:

• What changed with AI in software development
• Why I'm still not buying every productivity claim on LinkedIn
• How AI is reshaping work for regulated industries like finance, defence, and healthcare
• Why software is the canary in the coal mine for every other knowledge industry
• Why human review still matters when adopting AI in your business

Honest take from a founder who builds high-assurance AI systems for serious operators. 

If you're listening, let me know if the format works.

Transcript

SPEAKER_00 0:02

Alrighty, this is my first driving podcast. Uh so I'm testing things out and seeing how it goes. So you're gonna hear some some noise in the background. Uh, but we'll see what happens. So I guess first of all, why am I doing this? You know, I I found that it has been quite hard to consistently produce content. Uh it's something I really value doing. I think when I focus on it, I'm okay at it. But people seem to like what I put out there. But, you know, I run a business, I'm a dad, I've got a lot going on in my life. And sticking to a schedule to produce content all the time is not the easiest thing to do. So I think it's something that easily gets put off for for client work or for other things that feel like a higher priority. Yet I'm pretty aware that over time, if this is something that I'd committed to, say, for the last five years doing regularly, my life would be in a better position than it is today. So I I see the importance of it and I see why I need to do it. But yeah, it's a hard thing to commit to. So I've seen a few videos people putting up where they are just kind of squeezing it in where they can, and you know, they'll be eating dinner and recording a video, they'll be getting ready, putting on their makeup, recording a video, going for a walk and recording an audio podcast. And it's made me realize that there's really no excuse, right? At the moment in my life, I drive to work most mornings, I'm sitting in the car. I'll often use that time to make some phone calls, I'll listen to a podcast, something like that. But it's really often people aren't even picking up because it's too early. Like I can make that time more productive, and it's actually a pretty good time to record, I think. Um so and look, whether I put this out or not, whether anyone listens or not, I'm not sure. But I think it's a good way to capture some thoughts. Potentially I can put out the raw audio as a podcast and then use this as a feed into other content, LinkedIn posts, blog posts, things like that. I've often got a lot of ideas bubbling around in my mind, and I don't think I've got a shortage of things that I want to say. It's really sitting down and getting that onto uh onto paper, if you will, that is the difficult part. So this is going to be a bit of an experiment. As I said, I don't know if I'll even put this out, if anyone will listen to this, but it's something that I will try and we will see what happens. So I suppose in in our world, uh in software development, AI is the big thing, and that's the thing that you know we're heavily focused on. That's the big thing that everyone's talking about. In many ways, it's such a noisy space at the moment that even though we we're deep in this and we see an incredible amount of value in it, I'm almost at times loath to talk about it because it feels like, okay, that's what everyone's saying, and what's the kind of unique voice here? But really, I think that we're in a pretty good position to help people navigate this. We've run this business for over a decade, building software, really complex builds in regulated spaces. We understand compliance and security and quality of software and things like that at a really deep level. So I feel even though that was it was never intentionally so that we'd be well positioned for AI, this has kind of come along and we're in a really good position to kind of cut through the noise here around what actually makes sense, you know, what's real and what's not, but what's just hype. And there's an extraordinary amount of hype. So I think the first thing I want to talk about is that AI has really turned a corner. You know, I've put out a lot of content last year talking about how I did feel that things were really overhyped. And that's always the way you get some new technology coming out, and whoever kind of makes the most extraordinary claim seems to get the attention. So I think just the way that social media works and the news works is that people are really incentivized to be sensationalist. And so, right from the beginning, we've had people talking about how they've next their productivity. I wrote a post about you know a letter to the 10x professional really talking down this this whole thing, um, you know, only several months ago, really. And and there was been there's been a lot of that, people saying how they've got you know one agent doing thousands of things a day, and now they've let go of staff and they're doing this. And and I've been very, very skeptical. It most of the time when you look closer at it, it doesn't actually seem that legitimate. People are producing something maybe 10 times as fast, but the actual quality of the work is is really low or it looks good, but if you know what you're doing and you look at it, you realize it's not that good. Um, and I think there's this issue as well where, and there's a name for this effect, and I forget what it is, but we are not great at judging the quality in fields we don't know a lot about, right? It's this classic thing. There's an example of this where you know you open up the newspaper and you know you can open up a page about a topic you know a lot about, and it's pretty obvious to you that the author has no idea what they're talking about, and you're reading through and you're shaking your head and you're going, like, what on earth is this? And you sort of turn the page, and then immediately on the next page, there's an article about some other field about medicine or something, and then immediately you go, like, oh wow, isn't that amazing? How incredible is technology, and and sort of immediately accept it at face value. And I think we all do this, and it kind of adds to this idea that we all think that every other industry is using AI in these miraculous ways and really changing the industry dramatically, and then you know, we look at our own work and then we see that okay, the output's not there and what's going on, right? Um, but what what's what's interesting is that even though I felt that way for a long time, we've used AI a lot internally and for client projects, but in very specific cases, and over the last year, you know, maybe that's culminated in sort of 10-15% productivity gains, which don't get me wrong, is still amazing. And that's something that you know any business would be happy to have in a year. But it certainly wasn't this 10 times efficiency thing that people keep talking about, right? Certainly not even two times. And then around December, things started to shift. So basically, at least in software development, what really happened is a couple of things. So the some new models had started to come out, especially from Anthropic, you know, Claude Opus 4.5. Um and combined with what you could call a harness or say a coding agent, Clawed Code being really one of the main ones that we use. When you put these things together, all of a sudden the results actually did start to get a lot better. And it's no specific date as much as in at least in the tech industry. I think a lot of this blew up over December because people had a little bit more time to experiment and use these tools. Um, but it really was a step change. It wasn't just about usability. Like I feel a lot of people then would have felt vindicated saying, see, I told you everyone, you know, for the last year I've been 10 times as productive. Um the reality is that I still think those claims were very premature. And it's just that the tools and the models actually have caught up. Um and we all kind of felt like this is where it would go. It's just that it wasn't there yet. Um, but it is getting there now, and that's what's really kind of fascinating to see. So the main thing is that, say, in software development, they would, you know, these agents would write code and you'd have to manually review it. And a lot of the time it it was not the greatest quality. And you give it lots of guidelines around, okay, this is how I want you to write it and whatnot. And it just didn't always do it, basically. It was very hard for it to adhere to your instructions because a code base, especially for a larger project, is a really complex thing. It can be a really big thing. Um, so you know, it's much more complex than trying to get it to write like a single article, for example, where you're dealing in a project with what could be hundreds of different files or all interconnected in different ways. Whereas now we're seeing that adherence a lot higher. So it does stick to our standards. Um we've got very high standards. So to see it stick to those coding standards practices much more closely saves a lot more time. Because before, yeah, it would write code, it would, it could spit out, you know, many, many thousands of lines of code very quickly, but you know, that's not valuable if you then got to go manually change them all, right? Whereas now what we're seeing is that it it's a lot better, and we're at a point where you still have to read things very carefully and you have to still change some things, um, but it is a lot closer to what we would do if we were doing it by hand. Um, and that makes you know a really big difference. It's still yet to be seen what this really means for our industry, because I think there's a bit of a fallacy, and there always has been a fallacy in the software development that the development portion is everything. Um and this is why you'd see lots of companies hiring a developer and expecting these great results, and then they couldn't produce anything great. Um, and it was full of bugs and there was issues, and the things went offline and all sorts of things, because you never needed just a developer, right? A big part of our business has been providing all the roles to really guide building good software, and that goes from building the right thing and the research and the planning and the design, development, yes, but then also quality assurance, the deployment processes, how you manage it once it's actually live, the whole infrastructure, all of that. There's so many parts of it. So the development part is getting it a lot more streamlined with AI. And that's fantastic to see. It doesn't translate immediately to the whole software project speeds up um fivefold. Even if you could double the speed of development, you're not doubling the overall speed of output because you've got all of these things. But to me, that's just a matter of time because really it's a if you think of it, it's a bit like uh you know a manufacturing process where, okay, you've got the whole process is really only as efficient as the slowest part. And let's say now we've made a certain part of that process more efficient. Now the bottleneck kind of moves somewhere else. And then that's what we work on. So for example, we're seeing that in some projects development is now speeding up a lot, which is fantastic. Um the problem then can shift somewhere else, which is okay, how do we plan work fast enough and still make sure it's a high quality? Um so then we turn our attention to that. Um now we're making really good inroads to making the whole planning process a lot more uh efficient and faster with AI. And again, for us, the bar is that it always has to still hit that same standard. We can't allow AI to reduce the quality of the work. If anything, it actually has to increase the quality as well as increase the speed. Uh, you know, we're not interested at all in moving faster if it lowers the quality of what we're doing. That's really not the point. Um, I've made this point before, but it was always possible for us to go faster and lower the quality. We could have done that at any stage pre-AI, but we didn't. We we choose this level of quality for a reason. So we have to maintain that standard in everything that we do. And then AI has to speed us up whilst maintaining or increasing that standard, which is the really, you know, the bar that we've set. I think the right way to think about using AI tools in your business, whether in software development or whether you're in another industry altogether. So yeah, it's really uh it is starting to speed things up, which is incredible. It's creating uh entirely new processes, entirely new ways for us to think about software development and the entire pipeline. And I think people listening to this that are not in software development might think, okay, well, sure, why is that useful? And part of it is because it does mean you might now be able to build more software for your own business, right? Either internally or using, say, an agency like us. But really, I think the more important thing and the wider impact of this is that software is really the canary in the coal mine. You can look at what's happening here and forecast, I think very predictably, what is going to happen in other industries. It's not a chance that software development is being automated with AI first. It was a strategic decision by these companies because they've realized that if we make it easy to write software, then we can use that to go and automate other industries. So rather than AI trying to say, go and automate the legal industry, and they can now write software to do that as well. But AI writes the software, and so you know, in turn, really AI is automating it, but it does it through this medium, which is writing software. So it's almost a core capability they've tried to nail so that they can then move it to other industries. So and software development is a very complex industry. It's a it's uh you know, it's an engineering discipline that to do well is very complex. That's why software developers have traditionally been very sought after and you know paid really well. And if you look at all the big tech companies and things like that, it's because they get a lot of leverage off their of what they contribute, but also because it's not straightforward and simple work. So for me, it's not like it's just got and really sped up a very easy thing and now it needs to move on to the hard stuff. Um, you know, there's many other industries that would be of simpler complexity and some maybe even lower complexity that uh that there's no reason I don't see why it's not going to go and do the same thing too. So I think that's you know, that's to me the wider impact of this is to think about okay, well, this is happening in software development. So do pay attention here because it gives you an insight into what will happen in other industries, in potentially your industry. And then that of course starts to get you thinking starts you thinking about all of the wider implications of that. Okay, well, what happens if let's just do a thought experiment for us and say, well, what if the software development lifecycle, which is really the whole thing, right? It's not just writing code, it's like the whole process of of doing it. You know, what if that is twice as fast or three times as fast or four times as fast? You know, what actually happens? And what does that mean for a business model like ours that people are paying us to do it? Um, you know, what are the wider impacts of that? Uh and I think that's really important to think about and to think about well, what are your motes as a business? You know, for us, for example, we're we're heavily certified in things like uh ISO, you know, 27,001 and 9001, and a whole bunch of other things that that we do. So we already work in uh, you know, that's already a moat for us. The fact that we work in regulated industries uh is also a moat because they can't easily just go and like throw things at AI. So, you know, we've got very deep relationships and domain knowledge. So all these things that are harder for AI to just go and do and are harder for someone else to just start up a business tomorrow and start doing a similar thing. All our decades of experience of the best practices around there, all this stuff is really helping us speed up, but whilst maintaining quality where we see others speeding up and lowering quality. So for us, those emotes, it doesn't mean guarantee you that there's no issue. But we definitely see that this year, especially in the years to come, are a really critical point for our business. And that's why we're choosing to lean in and really reinvent everything about the way that we work rather than just waiting on the sidelines to see what happens. And I think that's really what everyone has to start doing. I mean, if you're in a professional service business, if you provide a service to people, especially knowledge work, it's gonna be critical to really rethink everything that you do from ground up, complete first principles, thinking about these tools and how you work. I guess more practically, you know, if I think about how we work, so we use a like a coding agent called Clawed Code, but it's a little bit of a misnomer in a sense that it's not actually just for code. And I think that's turned a lot of people off, especially because the initial way that you use this tool was through a terminal. So you'd open up your terminal and your computer and it kind of feels like you're using an old DOS system or something like that, which you know, we really, none of us, unless you're in software development, are used to. Uh, we're used to really pretty interfaces, right? Web-based software. So that's turned a lot of people off. That's something that is getting solved. It's already getting solved with tools like Claude Co-Work, which is effectively meant to be the user-friendly version of something like Claude Code. And really the idea is that tools like this, they can you can use AI just like you have been in Chat GPT or what have you. But they can interact with your local files and they can do work. And that's really the difference. So, you know, I was working on something yesterday where I wanted to create a spreadsheet of something. You know, we're doing an audit and I just wanted to create the initial spreadsheet of questions that I wanted to run through, um, and I basically fed in all the things that I wanted to go through over. And you know, instead of me typing out the spreadsheet, it's like, okay, let me give you know Claude all the information of all the things that I want to cover and let it do a first pass. Um, what you would see if you did that in say ChatGPT or in the Claude normal web version is it would then spit out, okay, here's all the things you should cover, and then you go to Excel or what have you and type it all in, right? The difference with one of these tools is that it will just create the Excel file for you on your computer in the folder that you gave it access to. Um, and it'll just create that. You can then open it up and just review it. If you want to make some changes, you just ask it to make changes, it'll make changes. And behind the scenes, and this is where sort of the called code idea comes from, it often is writing code. So in order to create that file, it might just quickly temporarily write a script to Python or something like that, you know, install some libraries that allow it to create Excel files and then it'll off it'll go, create the file, and then it's done. Now that code is sort of throwaway. It's it's not code that you're trying to maintain or use. It was code that really got written for a very specific request that you asked for just then, and you may have never asked that same thing before, and then that code will get thrown away. The job is done, right? Or you keep iterating on it in that session. So, you know, it's really changing the way you work because for me, what it means is often you're not necessarily logging into all these tools. There's lots of tools I use, and then there's a thing called um MCP, which I'll probably cover another time, that basically allows your coding agent or your coding buddy to go and access the tools that you are using. And often what I'll find is that you know I'll just be sitting there working with Claude and it's going off to those tools, and it might pull data from one tool, you know, do some work that I've asked it to do, and then push that output into another tool. And I'll still go log into that other tool and I'll sort of check the final output and go, yep, I'm happy with that, or let me edit this. So there's an enormous amount of oversight. And again, this kind of comes back to the idea that we're not interested in producing you know AI slop. We're producing are interested in producing the same quality or higher as before, but faster. Um and these tools allow us to do it. So the the human review step, you know, at least for the foreseeable future, I think it's is absolutely critical to make sure that you're happy with that. And you know, agents will pop up that will do those reviews as well. And eventually there might be some level of comfort you get to where you say, you know what, I don't need to check this. And I kind of liken that to an employee, right? Let's say I hire a new staff member. Initially, I'm gonna review all their work, literally everything they do, I'm gonna review it myself. Um, and I'm gonna make sure it's to the right standard and to the right quality. And I want to see how they email to the point actually where if they're gonna email a client, I'm gonna review it before they send the email. And maybe I'm a little bit too detailed in that way. Um and then eventually I'll say, no, I'm pretty happy for you to send it, but you know, make sure you see me in and I'm gonna review it and blah, blah, blah. And then of course, eventually, when you trust somebody and you trust that person um and they've proved themselves, you'll stop checking everything. And I might keep an eye on certain things and I certainly still review the really critical stuff as it goes out. Um, and eventually, eventually, if it's someone that you're, you know, you've worked with for a long time and and they've proved themselves, you're gonna stop checking, right? You're gonna at some point you've got to change, you know, where your attention goes and focus on on other things, right? Attention is finite. So that's uh that's really the same thing with with AI for me, right? So even if we were to have another agent that came along and reviewed the work, for example, or or whatever, just the initial agent doing the work, um, you've got to review everything in the beginning. And over time, you will ease up and you'll start to say, well, you know what, if it's this very basic thing, maybe I don't need to review it. We're not there yet, right? For now, still review everything. I still think that's really important. So you can make sure your your bar is not lowering accidentally. And that's that's you know a really critical component of all this. So anyway, I'm arriving at my destination. I think that was um that was fun for me at least. That helped me clarify some things and and and how I want to articulate things. So maybe I'll put this out, maybe I won't. If I do put it out, let me know your thoughts on the format and um and whether it was useful at all. And yeah, until next time.