Product Agility

Kinga Magyar: Vibe Coding: Building Software with Natural Language - Productized 2025 TalkInTen

• Ben Maynard, Gábor Suhajda, Kinga Magyar

Share episode

Send us a text

Productized Lisbon - an exceptional conference for product people. 

We're honoured to partner with Productized for the third year running - the energy, practical thinking and community in Lisbon make this event a must-attend for product leaders, and we're proud to bring you these live "Talk in Ten" conversations from the balcony.

In this quick-fire live recording from Productized, we talk to Kinga about "vibe coding" - using natural language to instruct AI to write code - and dig into real use cases, limits and practical prompting techniques to get useful results fast.

Key topics discussed

  • What vibe coding is and why it matters for rapid prototyping and non-technical builders
  • When vibe coding works best (MVPs, internal tools, solo builders)
  • When to stop and bring in engineering (integration with existing codebases)
  • Prompt engineering: practical tips - structure, clarifying questions, version control
  • Community, tooling and how to avoid long bug loops while experimenting


Guest bios

Kinga - Product thinker and speaker focused on applied AI and no/low-code tooling. Kinger teaches practical approaches to using natural language-driven development to ship prototypes and internal tools quickly.
Ryan Lane - CTO at Bobcats Coding, an AI engineering and product studio based in Budapest and Lisbon. Ryan helps organisations apply AI-focused engineering practices and supports community education efforts, including Bobcats' AI economics guidebook.

Thanks to our Sponsor, Bobcats Coding, for making the Lisbon series possible. Check out Bobcats' free AI economics guidebook at bobcatscoding.com and support the teams who make events like Productized and this podcast possible.

Host Bio

Ben is a seasoned expert in product agility coaching, unleashing the potential of people and products. With over a decade of experience, his focus now is product-led growth & agility in organisations of all sizes.

Stay up-to-date with us on our social media📱!

Ben Maynard

🔗 https://www.linkedin.com/in/benmaynard-sheev/

🐦 https://x.com/BenWMaynard

💻 https://sheev.co.uk/

Product Agility Podcast

🔗 https://www.linkedin.com/company/productagilitypod/

💻 https://productagilitypod.co.uk/

🖇️ https://linktr.ee/productagility


Listen & Share On Spotify & iTunes


Want to come on the podcast?

Want to be a guest or have a guest request? Let us know here https://bit.ly/49osN80

Welcome to the Product Agility Podcast where we explore the ever changing world of product leadership and org design, helping you navigate complexity and build better outcomes for your people and your customers. This week we're coming to you live from Lisbon for the third year in a row at the Productize conference where I'm grabbing 10 minute conversations with product thinkers, leaders and innovators from around the world. These quick fire chats are all about what's shaping our industry right now, from AI and product strategy to the human side of building great products. Now a huge thank you goes out to Bobcatz Coding for making this Lisbon series possible. Bobcats is a Budapest and Lisbon based digital product studio specializing in AI engineering and end to end digital product development. They're also on a mission to educate the market, exploring a new topic every six months and this fall is no exception. Their latest AI economics guidebook is out now and you can download it for free@bobcatscoding.com now here's your talking 10. Good afternoon or good morning listeners, whatever. Time it is for you. It's afternoon here, it's post lunch, things are quieting down and we're joined by Kinger Kinga, whose surname I'm just gonna make a hash of if I try and say it. What's your surname? Kingaar. Exactly. Hi. Hi. Thanks for having me. It's really nice to have you here. Really excellent. Have you here. Kinger and Ryan, CTO of Bob Katz. Hello sir. Hello. Good to be here. Great to have you here. King. First time at Productized? First time speaking at Productized. I attended last year. Oh, really? Did you enjoy it last year? Yeah, absolutely. So much so inspired me to apply. Look at that. And here you are. Here you are. It's going up and up in the world. Can you give our listeners a little bit of an overview of what it is you're going to be doing here at Productised? Yeah, absolutely. So I'm giving a talk on Friday about Vibe coding and prompting and my hope is that by the end of the talk, the audience will be inspired to try wipe coding. Those of them who haven't tried yet, so they wouldn't feel like that they have to prepare more or watch tutorials about it, but just try their hands at building. And those part of the audience who have already tried Vibe coding, I hope that they will live with some actionable practice or techniques rather and they will be motivated to just continue and push forward because sometimes it can be very frustrating. So for those who might not be familiar with it or who are Struggling to keep up with the rapid changes we're living through now. It's October in 2025 when we're recording this. How would you define Vibe coding? So I would define it that you're giving instructions to AI about what you want to build. You give these instructions in natural language and AI will write code and ultimately create the software. So you basically can build a software product without knowing how to code. So there's a lot of discourse in the industry currently about the limits of Vibe coding or what it's good for and what it's not good for. Can you describe how you see its usage and how you see that might change in the near future? Yeah, absolutely. So I think Vibe coding is amazing for several different use cases and not so great for others. So let's start with the amazing ones. I think it's great to create minimum viable product quite quickly so you can test something with your potential users and validate your idea or the problem you are trying to solve. It's also great to create some internal tools if you are having some problems in the organization and want to solve that quickly. It's also amazing if you are not a technical person and have a problem. For example, I've seen this in one of the competitions that Lovable organized and I took part in where person who is teaching English for young learners and she needed a platform to do that and she hired a development agency who couldn't deliver the product and then she had to build it herself or pay another 20,000 Euro. And she actually built this product during this competition and now she's using it. So that's. I think that's a really inspiring story. Nice. It reminds me, I was sitting down with my 9 year old son, the burgeoning opera singer and he. He's not an opera singer, he just wishes he was. And we sat down, he had a maths test and he's called the Ultimate Challenge. It's a thing of a UK where you have to do 100 questions in five minutes and various degrees of difficulty and he's reached the top of the thing and he has this test coming up. So we sat down in front of ChatGPT and I went through and we spoke to. He got him. I got him talking to ChatGPT to make an application for him, to test him in the Ultimate Challenge and then we went back and forth and refined it to actually come up with a neat little tool that he could use as a practice. And I thought, this is amazing. When I think back to my childhood, sitting down with my spectrum 48k with rubber keys trying to figure out what Basic meant and doing little programs there to now be able to Fast forward like 30 odd years and you can have a convers with something and it produces it. You know, since all those years I spent mucking around in PHP and MYSQL making small web apps like now, it's just so much simpler. But there are limits to what we can do. In your experience, what are the. When does this become. Get to a point where somebody. People should be saying, well I've taken this as far as I can go. I just need to throw in the towel and get some extra support. Right. So I think it gets a bit more complicated when you already have an established product and you want to introduce a new feature. So then you already have your code base and maybe the Vibe coding tool has a different tech stack. So for example, lovable that I've been using the most heavily, it has a fixed text tag that it works with. And so it might be more challenging to integrate a new feature into your existing product by Vibe coding, but then you could still use it to validate an idea. So you build it quickly and put it in front of users. And even though it doesn't exactly look like how your product looks like, maybe the functionality is already there. So you can see how users interact with it, but in a test environment. Right. So you are not rolling it out to production yet. Yes, yes, yes. And what about the individual people, individual builders who are here getting excited because they feel the barriers to kind of being able to turn their product into something real? What advice would you give them? Because like for one day I Vibe coding at the weekend building some ish building native automations. It could get to a point where I'm actually, I cannot do this part. I need to call upon a friend to help me understand how to kind of write the JavaScript to interrogate the API because it just. I couldn't get it to work. So I knew my limit because I know roughly what I'm capable of. But lots of people don't know their limits with this. So how would you advise people? Yeah, so the nice part is that you don't. Many of these tools are marketed in a way that you don't have to understand how software works. And of course it helps a lot if you do, if you have at least a basic understanding or a basic technical literacy because oftentimes you get into bug loops that you are trying to maybe add a new feature and then AI is telling you that oh it's there, you try it, it's not working. And then you go in loops for could be a couple of hours. So in those cases, I found that it's often best to. Most of these tools have version control, so you can go back to the version that was before you started implementing this new feature and getting into that bug loop. So I find that it's a good practice to then revert to that version of the software instead of just continue fighting with the AI, because then you can put into your prompt all the learnings that you learned during trying to fix this bug. That's one. And the other one is that most of these bytecoding tools have their own communities. So you could go into those communities and maybe ask for help from other community members. Because based on my experience, people are quite helpful and friendly in these communities. So maybe they can help you solve the issue where you got stuck. Maybe we could talk a little bit about the art and science of prompting. So what would you recommend people keep in mind as they start to figure out how they should talk to these different AI tools? Right. So I found that I can be pretty lazy when it comes to prompting, and that's a very bad strategy because the more context and background you give to AI, the better the output will be. And so sometimes what helps me is that I explain in a prompt what I want to build or do and then add a question to the end. Do you have any clarifying questions? And then the AI actually comes back with maybe five questions. That helps me to figure out, okay, I forgot about this and this and these details. And then I can reply and then the output will be better. Another technique would be to structure your prompts so you could use markdown formatting, bullet points, numbered lists, or even XML style tags, because then the large language model will be able to parse your message easier if you have structured your prompt that way. Great tips and really useful. I think the, the clarifying questions one is a great one, but I think these are I talking to Tamash earlier about it and a few other people. There's lots of skills which are really necessary to be able to communicate clearly. And there's little things we begin to learn. You know, I remember years ago writing requirements documents like way way back when and having to being told the thing to put on each requirement was including but not limited to because it covered your ass. But I think that this is where we've AI. There are some catch rules that we need, but the clarifying questions, adding structure to it are just fantastic. Fantastic tips. If you're listening to this, hop onto a AI chatbot and chatgpt, Claude, whatever you like and try and get it to coach you. I think some of the biggest hurdles we have to making the most from these isn't at all. It's us. Yeah, absolutely. And it's even hard to communicate in real life with people and with the machine. It's even more complicated. Yeah. When you're vibe coding and you're chatting in what language do you use? Your native language or do you feel I use English? We've been having this internally recently where we had people using a tool called Onyx, which we've really thoroughly blown us away. What it's capable of in a very quick time from building context around a large language model. But people's prompts weren't that great. And so I said to them, would you find it easier to prompt in Germany or Lithuanian? They're like, oh, yeah. Well, prompting that language, then that's the hidden power of some of these models is that particularly when it has your own organizational context, is it can translate it for you. And I think that helps people understand it a lot easier sometimes. I mean, I'm English. I know one language. I can't imagine what it is today. Multiple. But I think there's so much. There's so many things we can. Yeah. And the good part with AI is that it doesn't mind if you're making typos or making grammatical mistakes. So that's an advantage of it. Although I tried to correct my typos because I don't want it to use the extra compute. I feel like I'm already destroying the environment enough. Anyway, thank you so much, King. Yeah. Again, talk about this forever. I would absolutely love to. We'll get you on a. We're going to be launching a new podcast of a new year. Maybe we can get you along to that title. TV confirmed. But Ryan will be talking to you about this as well. But thank you so much for coming along, Ryan. Thank you for your awesome questions, everyone. Thank you for listening. This has been another talking 10, and there's plenty more to come, so do remember to subscribe to this podcast on the podcast platform of your choice. Thank you.



Head Shot

Ben Maynard

Host