← Scrum Master Toolbox Podcast: Agile storytelling from the trenches
Scrum Master Toolbox Podcast: Agile storytelling from the trenches
AI Assisted Coding: From Designer to Solo Developer - Building Production Apps with AI With Elina Patjas
00:41:09
Loading summary
Transcript55 lines
- [00:04]
A
Hey there, agile adventurer, just a quick question. What if, for the price of a fancy coffee or half a pizza, you could unlock over 700 hours of the best agile content on the planet? That's audio, video, E courses, books, presentations, all that you can think of. But you can also join live calls with world class practitioners and hang out in a flame war free and AI slop clean slack with the sharpest minds in the game. Oh, and yes, you get direct access to me, Vasko, your Scrum Master Toolbox podcast. No, this is not a drill. It's this Scrum Master Toolbox membership. And it's your unfair advantage in the agile world. So if you want to know more, go check out scrummastertoolbox.org membership. That's scrummastertoolbox.org Membership. And check out all the goodies we have for you. Do it now. But if you're not doing it now, let's listen to the podcast. Hello everybody. Welcome to a very special AI Assisted coding bonus episode. And for this bonus episode we have with us Elina Patias. Hey Alina, welcome to the show.
- [01:24]
B
Hi, thanks for invitation.
- [01:26]
A
So Elina is in Helsinki, Finland and she's building Lexi, an AI powered studied course study tool. Pardon me, for kids age 10 to 15. Frustrated by this ineffective read for exam routines and gamified edtech fluff, she designed Lexi to turn handwritten notes into adaptive quizzes that reveal knowledge gaps. Private ad, free and subscription based, Lexi is learning simplified. So Elina, that's a short intro. Tell us a little bit more about yourself and about Lexi. Why did you start developing Lexi and why are you using AI to develop Lexi?
- [02:12]
B
Okay, so about my background. Like, I have been building software for more than 15 years here in Finland and I have been mainly working as a designer and like product management side of stuff. And about a year ago I kind of, I resigned my old job back then and, and I was like doing this, I was updating my portfolio then and I was picking, I was in a mindset that I'm going to look for another job and, and so on. And I decided then that I will try using cursor, which was like a really new tool back then. And, and I was doing that and I was like, well this is actually pretty easy. So that was like the moment then I realized that, okay, if I can do this, what else can I do? And I had this idea of an app in my, in my head back then. I have two kids by the way, they are like 11 and 6 right now and my younger one learning to read. And I wanted to kind of help her to practice those reading out loud skills. So I tried to, I was thinking that what if I tried to kind of build an app to help her to kind of overcome her fear? And I did that and I launched it to App store and I was like, okay, so if I can do this, so what else can I do? And this all happened, by the way, within like two weeks. And then I had this idea of building Lexi for my background. I have been working in educational technology for quite some time and I have been working with this, designing these online learning platform for Finnish schools and students and teachers. And I had this really big desire to be able to help kids to practice better. Because not all kids, even though we have superbly good educational system here in Finland, not all kids can kind of get the help that they need, like when they need, especially when they are at home alone, not all have like parents who can help them effectively practice for exams, for example. And I had gathered like a lot of insight about that subject and then I was like, okay, I'm going to try to do this. I'm going to try to build Lexi and I'm going to try to do it by myself. Because I was super curious of seeing if I was going to able to do this. I was a bit skeptical. This was a year ago. But the more I spent time doing it, the more convinced I was that I might be actually able to kind of.
- [05:30]
A
So was this, these early experiences with Cursor, which is, I guess one could call it an AI assisted coding integrated development environment. Right, like, because it has everything, right, like from the text editor to anything that you can connect it to whatever models in the back end and whatever.
- [05:49]
B
But.
- [05:49]
A
But was it that experience of writing those first portfolio contributions and then later even an app with it? Was that what kind of triggered you to say, hey, I can definitely build something much bigger with this?
- [06:07]
B
Yes. Yeah. And I think that what kind of happened inside my head was that, like I said, I've been building software for quite some time, but I've always struggled with the fact that, how can I say this nicely, that I have this vision in my head because I am a designer, so I know the users, I know the user needs and I know the business requirements, I think in systems. But there is so much this overlapping things happening simultaneously when you are working in a bigger team that most often what happens after dev handouts and all those tens of thousands of meetings that maybe like 10% of that your original vision is like, you're able to execute it, and most often you're not ever going to even able to iterate it again. And I have always struggled a little bit, well, a lot with this fact. And what happened to me with those first experiments was that I kind of realized, even though the tools, the AI tools were not completely there yet, but I was able to see like the shift in the paradigm. So it was changing from the thing that we are writing code to this, that you are just kind of building the product. So. And I was able to do it by myself. Like, I was like, I was really excited about that.
- [07:58]
A
Yeah, I can see that excitement even today as you talk about it, and I really appreciate that. So if I got you right, what you were saying is your own previous experience of software development was that because of the way software is developed with a lot of people and a lot of processes and so on, a lot of the ideas, great ideas that you might have for the product end up not being implemented, that vision that you were talking about. But when you work on your own with just this AI tool, you can actually bring many of those ideas into fruition. I hear you didn't say it, but I hear you mean a lot more of those vision ideas than in a traditional software development environment.
- [08:41]
B
Yes, yes, definitely. And not like, just like, I can bring more ideas, but I can like create them exactly like I was designed to do that because it's always like, there are people who hate making compromises and I'm one of those. And I think that is like a trade of a lot of designers. But now I am able to kind of execute the thing that I want, like 100% the way I want it. And I can do it really fast and I get the feedback immediately because I can test the thing. For example, I can share you an example from last week. In this product, my product, Lexi, I have this feature that users can send me feedback request or if they find a bug or something from depth directly to my personal WhatsApp number, which is pretty, pretty insane. Not scalable, but because I'm just starting this out so I can do things that don't scale. But I got this request from a girl who was studying and she was like, she would really appreciate if she would be able to kind of iterate the study set after the creation of the study set, because she really likes to listen to text out loud and if there are like spelling errors and stuff, it makes the reading experience harder. And I was like, that's a fucking brilliant idea. Part of my language. But yeah, I can execute that. And the next morning it was 9:15. I sent her a screen capture of this new feature and I sent her that. I'm going to roll this out in the next release, which went out two days ago. And that makes me really happy because I'm actually able to do what I was, what I have always wanted to do as a designer and the reason why I'm actually even building anything.
- [10:46]
A
So, yeah, absolutely. Because we build for people, right? Like, that's a very good point. And you're not saying this, but what I also hear you mean with that story is that as an, as a designer, you now feel empowered to bring ideas to life rather than feel dependent on others to accept and maybe bring some of those ideas to life.
- [11:12]
B
Yes, definitely. That is a very huge thing for me. And one of the main reasons why I kind of feel that I want to talk about this and I like, I want to want that other people are able to kind of experience the same freedom. And I really hope that we are getting away from the gatekeeping more and more in the future.
- [11:38]
A
When you think about this AI assisted coding, how do you define. There's this word going around everybody's talking about. It's called vibe coding. How do you define vibe coding and how is it different, maybe from what you're doing and what you would call what you're doing, or maybe even other AI assisted coding approaches?
- [12:01]
B
Yeah, well, I have a little bit mixed feelings about the word wipe coding. I do know what the origin is, like what Karpathi meant, but what has happened is that it has translated into this kind of thinking that it's sloppy, that our people are using it just to build quick prototypes or not like reading the code at all and stuff like that. And then there is this, the other side of the spectrum that you're using AI as a tool. When you are kind of implementing new features, you ask help. And those are the two ends of the spectrum. I don't think that I'm neither because, for example, what I do is that I use natural language. When I start my, like, a coding experience, for example, I can share like a practical example which might be easier to understand. Like how I do this is that last week I was doing design for this new feature which is called Fill in the Plank Question. So this kind of assignment type when kids need to, like, find words to fill in like sentences, and especially helpful when you're learning languages. What I did is that I spent like 30 minutes designing the whole workflow inside my head. So kind of this implementation plan that I wrote out in plain text and it had all the UX interactions to dataflow and the kind of the overall architectural decisions that I made during that phase. So I spent a lot of time writing like really, really good spec and then I gave that to Claude code and I started kind of like when I work with Claude code, it's like I treat it as like my, like in a sense, like a companion. It's like I discuss the different scenarios with it and it produces me the code and I do not write it anymore. I did that when I started with Cursor, but along the way I've noticed that actually AI is able to write really good code if you just know how to read it. So that's like, I do read the code what like Claude code writes. And it's actually super helpful for me because I understand also like my own systems better when I can easily spot like what are the differences it done and stuff like that. But I do not write any code. I haven't written a single line of code in a long time.
- [15:01]
A
So did I understand correctly that now the way you developed with cloud code is that first you spend some time thinking about, I'm guessing, the feature spec, you called it, the data flow, the architectural decisions and the user interaction. So all of those aspects. And once you are happy with the spec, then you give it to Claude code. But how do you then interact with it? Because one of the characteristics of these tools is that they can easily get confused. They can try to do too many things at the same time and then things don't come out right. So how do you interact with it? Right, you've written the spec, you're happy with the specific, but what do you do next with the AI assistant? In detail?
- [15:43]
B
Yeah. Yeah. Okay. So I obviously don't give the entire specs to Claude and say like yolo and see you in three hours. But at that point, when the implementation plan is ready, then I split it into smaller tasks because I can. Claude knows very well my entire code base because I also update the Claude MD all the time. Whenever I like make changes, I make sure that Claudia MD is really up to date. So it basically knows everything about my code. And I have already. Someone asked me last week how many lines of code there is in my app at the moment. So I have 32,000 lines in the client side and 8,000 on the server side. So few lines, a lot of lines to be like writing by yourself. And so after that I give the overall the idea what we are trying to execute together. So I said, he, no, it.
- [16:53]
A
Yes, very good.
- [16:55]
B
So it knows what we are trying to achieve. And then I kind of split it into smaller prints. And that's something that I have been like learning a lot how to prompt really well. So like it knows, it has to know really well what I am trying to do. And I kind of have these checkpoints all the time that I kind of check what is what he's like, it is trying to suggest me to do and how to proceed. So we have like this dialogue all the time. So it's not like just like. So you take a coffee. Yeah.
- [17:35]
A
So you take the whole specific that you written on your own and then you do maybe some small updates to the Claude md, which kind of the project file that Claude always reads before doing anything. And then you say, okay, but out of this, let's only do this thing now.
- [17:52]
B
Yeah.
- [17:53]
A
How it goes?
- [17:54]
B
Yeah, mainly. Yeah. And sometimes I've noticed now that Claude code is getting so much better. So sometimes it can actually suggest like, like what if we do this, like this way? And then I kind of think about it, would it be like a good approach or not? And then usually say no, because normally it is better than my original approach. But yeah, so I treat it really much like this.
- [18:22]
A
And how about the spec? Do you also use an AI to help you write the spec out or do you write it on your own?
- [18:28]
B
Yeah, I write it with Claude. Yeah. So I basically have this, like, I start doing the planning together with Claude. I have like the original idea is always like really clear in my head, like what I want to achieve. And then I kind of have this conversational partner because I am working solo, like I mentioned before, so I don't have anyone to kind of discuss this. So Claude is also like my planning partner and then he. It is my executional partner.
- [19:01]
A
So it's interesting that you started way back with Cursor. Right? Like when you were first getting acquainted with these tools and getting used to the idea and nowadays you use Claude code. Is there a specific reason why you change tools? Like why did you decide to go from something like cursor into something like Claude code? And how do you use cloud codes? Is it command line? Is it integrated into VS code or something else?
- [19:28]
B
Yeah, I think that when I started, it was a year ago, then Cursor was just like new thing and I wanted to try it out because. Well, not that new thing, but like a, like, like pretty new thing and I wanted to try it out and then I kind of. Because I was just starting doing that I wanted to. I was thinking more like, I want to be able to learn how to code. So that was my mind shift then, and I really used it as a tool to help me. Like, actually I was writing the functions myself and kind of trying to get, like, better at coding, but at some point during this year, I realized that I actually don't need to do this because it will do it so much faster than me anyway. And it would require so much time if I would start from the beginning to kind of start, like, learning to write the code myself. So it shifted like, a little bit at a time to the direction that I actually do not want to write the code myself, but I want to be able to understand what like, the system is, like, doing for me. And, like, and when the cloud code came out, I didn't actually first started using it because I was quite happy with the current system back then. But then Cursor did some, like, really stupid changes to their pricing systems and I was like, no, this is not working. And I also noticed kind of a, that due to the fact that my mindset was shifting, I was kind of already wanting to have a different type of a companion. And then I did test with the cloud code and I've been pretty happy.
- [21:23]
A
You use in the command line or I use it?
- [21:25]
B
Both. I use it in the command line, but I also use it in my ide. So I use jet, actually, because it's super fast and. Yeah, so. And I can actually run like, multiple, like, commands simultaneously. So, like, one cloud is doing something, another one is doing another thing.
- [21:48]
A
And I kind of IDE that you. That you use.
- [21:52]
B
It's called Z Zed.
- [21:54]
A
Yes, very good. I'll put the link to that in the show. Notes for people to check it out. And throughout this experience, what have you learned about what makes AI assisted coding work for you?
- [22:11]
B
Well, I think that I've already kind of said it, that it's like, what works for me is that I finally have, like, tools to execute, like, my own vision without no compromises. And even though I have always loved working really closely with devs, there's been some nagging thing in my backside of my head. It's like I have spent so much time of explaining how things should work, like, instead of doing stuff, and now I'm finally able to kind of do what I want and just the way I want it to happen. So that was like, the main reason why I think that using AI as a coding assistance for me is like, a good choice.
- [23:18]
A
And what are some of the, you know, looking at it from the other side, what are some of the things that don't really work? What are some of the anti patterns that you've discovered in working with AI assisted coding tools?
- [23:31]
B
Yeah, I think that is like something that quite often if you hear people about like talking about the problems of wipe coding, so obviously if you just pull it like full let any AI tool to run rampant in your code, shit will happen. So you need to be actually like really specific about what you want to do and how do you want to do it and kind of a treat the AI as this highly skilled companion that you need to be able to communicate yourself super well and be on top of everything. So that works for me. I have no idea how other people are using it and I've seen actually super good like solutions people have like truly 100% prompt, like wipe coded like without no previous coding experiences and doing stuff with lovable and blah blah, blah. But I don't know anything about that because I don't work like that. But how I work, this works for me super well.
- [24:47]
A
So what would you say? Like, because one of the things that you talked about is how frustrating it felt for you before when you needed to spend so much time explaining how things should work and why they should work that way to other human developers. Right. But now with Claude code as an example, you also need to explain, so what is the big difference right now versus what was back then when you work with teams of developers?
- [25:14]
B
Yeah, I think it's like, it's so much easier. I think that I'm super bad at communicating out loud. Like you can maybe hear. I work actually quite well when I write things because it's easier and it's more natural for me to write things. And also I can. The iteration loop is so fast because when you kind of fail to explain something, you get the feedback so fast that you can actually say, no, no, no, this is actually not what I wanted. And you can clarify the thing that you needed like in really fast. So the feedback loop is so fast that it makes it so much easier. But it's true, you do need those communication skills. Are you working either with a human developer or an AI developer?
- [26:08]
A
The other thing that you said that I want to highlight is the speed of the feedback loop. This is something that we very often underestimate because when we are working with other humans, we are not telling them what to do. Right. Like we're saying, this is what I want us to do. In the case of A designer or an architect or whatever. But then the developers, they might have other tasks that they need to complete first and then they'll complete those and then at some point they start working on the things that they've agreed with you and only after that you start getting some output that you can look at and get feedback from. But by that time you might already be in a completely other type of work. Right. Like you might be thinking about features that are coming up a month from now, whatever that is. And because of that, like that lack of synchronicity creates this problem of lengthening the feedback loop. But when you're working with an AI assisted assisted coder, then what you're doing is that you're, you're prompting and getting the feedback at the same time. It's not real time because it takes time for the LLM to produce the output, but it's so much faster. Right. Like it's a few minutes, it's not a few weeks.
- [27:18]
B
Yeah, yeah, definitely that. And that's something also like when I have been working as like in a traditional like development teams, is that like when you are producing designs, you're producing some kind of like figma file? Well that's not designed, that's just like a step inside of the pipeline. And then you have to do this really tenuous handover to someone who is trying to interpret what you were actually trying to do. And most often a lot of things vanish between that phase. And I think that quite many people who work in software development do know this struggle. And there are like a lot of ways we've been trying to like solve this issue. But for, for me, like I've been doing this for more than 15 years, I haven't seen not that much improvement. The only way I've noticed that it actually works when you are like bare programming or working alongside weather dev. So you can actually like share a room or you can actually share a slack conversation while you are doing. But then basically what the devs are doing is you're guiding them at the same time that you're developing the feature. Yeah.
- [28:52]
A
And then with the AI it just happens faster. Because also one of the advantages of AI is that it's faster.
- [28:59]
B
Yes.
- [28:59]
A
Right. So if the feedback loop is there and if we have the communication skills to transfer the knowledge to the AI or transfer the ideas to the AI, then the typing also happens faster. But when you look at your own experience, I mean you started I guess a bit more than a year ago as we record this, when you look at your own experience how it has evolved over this last 12 months and you kind of project that into the future. How do you see this trend going? Like where do you think this AI assisted programming is going to be? Let's not be too far into the future, just even six months from now. Like what do you see are the trends and how do you see that approach to software development evolving?
- [29:44]
B
Yeah, well this is something that I'm going to most likely get like a lot of hate because I already get that whenever I write about this, for example in LinkedIn or somewhere else about like my what I think was happening, like not even just coding but like how we are building software. And this is solely based on my own experience. And also what I can see by following what is happening outside of like our Finnish bubble maybe is that I think that kind of the whole paradigm is changing like from like how we are like interacting with software. I think that we are going to the direction where people are like the software gets built in an instance. Like it's like it's going to this like intent based mode when we actually don't even need apps or software as we know them right now in the future. But we kind of are able to kind of shift to the direction where the software just is happening like in the minutes. Like for example, if I could think about Lexi, my own product. I actually do hope that eventually I will not need an app to do this. But for example, when kids have they need to practice for exams, for example, they kind of just can tell the AI that I need to practice for this and they will create instant practice and it is gone after the kid has used it and they will never use it again. The next time they need to solve the same problem, the same UI gets built all over again. This is like a really raw thinking in my head. I don't know if it's going to happen and what is it, six months or three years or whatever. I don't have that kind of knowledge. But I can see that something is changing in the way we are building software. And one thing that I've noticed that I and I think that will change is how much one people can actually do with the proper skill set. Because I think that I can say that I am a walking example of if you have the proper background and what good looks like you can actually be able to do several things at a time. What used to require like 10 people at least to build before. So yeah, I think that there is some kind of shift coming.
- [32:52]
A
Absolutely. And one of the interesting things that you said there is this concept of emerging software in the moment, right? Like if you need something done then through the interaction with this higher level tool, whatever that might be, like Lexi Square or Lexi number two, or Lexi V10, whatever that is, then things just emerge as necessary and then they don't stay there forever. Because right now we have this idea that software is something that first you need to develop and then you need to maintain eventually, forever, if it's successful, right? Because if it's not successful, nobody cares. But if it's successful, then you need to maintain it forever. But what I hear you say is that the way you see this AI assisted coding, or even AI programming or AI software development approach is that it will eventually just create the software that is needed at the moment and then when it's not needed anymore, it just vanishes, right?
- [33:50]
B
Yeah, yeah, that's like, that's what I think right now. I don't know if it's going to happen. I have no idea. I don't have like a magic ball I can look into. But that's how I feel like as a designer, I think that one of the key skills is to be able to kind of read all the silent signals. And these are not even silent anymore because these are actually quite loud signals. But this is also something that I've noticed that if I speak out loud about this, people get pretty pissed off. And I understand because it's also super scary because you can actually see that if this is happening, it will change a lot of people's lives, including myself. I have been wondering this and thinking about this a lot because, for example, for me it's taken like 15 years to gain this knowledge that I have and a lot of practice now I can see that basically good AI tools can't do the, like the actual design, like the user interface design or UX design pretty well already. So you can like. I'm like, okay. So I also do feel that it's at the same time it's scary, but at the same time I'm really excited because I can see also like the bonus side, what is happening. So I'm just kind of trying to adapt myself and my own thinking that what is happening, what is going on.
- [35:31]
A
At the end of the day, the tools are nothing, right? Like at the end of the day, if we're solving problems that people have, not problems we invented, but problems people really have, then there's going to be value there. And however that gets delivered, you know, we can innovate there and maybe it's AI maybe it's something else.
- [35:47]
B
So.
- [35:48]
A
But definitely a great and inspiring vision. Thank you for sharing that, Elena. Now for those that want to maybe dive a little bit more, is there like a book, a video, a podcast, whatever that might be a resource that you could recommend for those that after this episode are saying maybe I should give it a try?
- [36:09]
B
Honestly, I think that the best resource is just your own frustration with some existing tools. Because I always, if someone is asking like, should I take a YouTube tutorial or whatnot, don't do that. Just open whatever tool that you're using is. It's like Claude or ChatGPT and start interacting and discussing, kind of getting into this mindset that you're exploring what you can do and then just start doing. Actually, for example, I told our mutual friend of ours, you'll see that for example, if you have struggles that your kids basketball team like how you need to make the reservations for the team, like rebuild that one, you know, like take time and think about how you can do that better and just use the tools and see how it goes. Because that's the quickest way of learning and quickest way of actually understanding how this might affect on how you work yourself. And so I actually don't have any books or YouTube tutorials or anything to recommend because also if I recommend something that I've watched like yesterday and people go look at tomorrow, it might be outdated already.
- [37:29]
A
It might be out of date, especially in this field. For sure.
- [37:32]
B
Yeah.
- [37:33]
A
All right, everybody, check out Lexi. It's an AI powered study tool for kids age 10 to 15 that Elina is writing. I'll put the link to that in the show notes. And how about you, Elina? Where can people go to find out more about you and the work that you're doing?
- [37:51]
B
I'm not super active like in. Well, you can go to my LinkedIn or my X or Twitter. I actually call it Twitter if you want to see like something like I should post my thinking there sometimes, but most often I. Because my users aren't there. So my focus is not there either. So my users are in TikTok and WhatsApp on channels and YouTube. So you can find me those channels if you're interested in the product that I'm building. And so if you want to know me and what I do, just download the app and try it yourself because that is maybe the fastest way of kind of knowing like what I do and how I do it. Yeah.
- [38:38]
A
Elina, thank you very much for being here with us and for your generosity with your time and your knowledge.
- [38:44]
B
Thank you. Thanks for inviting me. It was nice to be able to share some of my thinking.
- [38:53]
A
Alright, I hope you liked this episode, but before you hit next episode, here's the deal. This podcast is powered by people like you, the members who wanted more than just inspiration. They wanted real tools and real connection to people who are practicing Agile. Every day we're talking access to over 700 hours of agile Gold, CTO level strategy talks, Summit keynotes, live workshops, E courses, Deep Dive interviews, books, and if you're into no Estimates, we got the pioneers of no Estimates in those Deep Dive interviews as well. Agile Business Intelligence, creating product visions, coaching your product owner courses, you name it. You'll get invites to monthly live Q&As with agile pioneers and practitioners, plus a private Slack community which is free of all of that AI slop you see everywhere. And of course without the flame wars. It's a community of practitioners that want to learn and thrive together. It's the best place to connect with community and learn together. So if this podcast has helped you before, imagine what you will get from this podcast membership. So head on over to scrummastertoolbox.org membership and join the community that's shaping the future of Agile. We have so much for you, so check out all the details@scrummastertoolbox.org membership because listening is great, it's important. But doing it together, that's next level. I'll see you in the community Slack we really hope you liked our show. And if you did, why not rate this podcast on Stitcher or itunes? Share this podcast and let other Scrum masters know about this valuable resource for their work. Remember that sharing is caring.