About this talk
In this session, Pinny Reznik from Resync discusses the implications of artificial intelligence for software development. He presents various tools and methodologies that allow non-developers, referred to as citizen developers, to build applications without extensive coding knowledge. Reznik emphasizes the shift towards an AI-native approach, where AI handles much of the coding and automates processes, allowing for faster development cycles. He explores concepts like dark factories, which utilize automation to create software efficiently, and introduces several tools like Laravel and NA10 that facilitate this new paradigm. The talk also cautions that while the speed of development may increase, careful planning and architecture remain essential to prevent potential challenges that arise from scaling AI-driven solutions. Ultimately, Reznik proposes a vision for the future where software development becomes more of a systematic, industrialized process driven by human intent.
Full transcript
Uh, one moment. Yeah, now, now, do you hear me? Good. Okay. Okay, that was exciting. The laptop broke down and the microphone in the same time. So, that's uh good start. So, uh my name is Pinny Reznik. I'm from company called Resync. What I'm going to talk about is to certain extent is written in this book, which if you like, you can get for free digital version.
Um I'm going to talk about AI like everyone else. And uh the first thing is a bit of wipe coding. This is application at our CFO built for internal systems. So, he had bunch of uh uh SaaS products to manage internal timing and all kind of other financial stuff. And he hated them. And so, he just went to a bit of lovable, a bit of cloud code,
and built it. Uh and it now we are using this application, which is entirely custom built, only used by us. Uh and he has no, but zero software development background. This is a dashboard I actually built for marketing and sales. Those are not real numbers. Uh this is another project I built last weekend. So, I had this idea to I don't know what AI native actually is.
There's a lot of talking about it, but I will what we know is different from whatever was in the past, but there is no clear definition. So, I thought, "Why wouldn't we build a website that actually ask the question from the people? We can collect observations over over time create some sort of patterns and and uh and consistent definition." So, this is it's it's just a website.
It's not super complex. There is some functionality in it. But the point is it was built in two in two days. I do have some software development background, but I didn't write a single line of code on this. sort of considering all that and considering that cost of software development goes dramatically down, sort of close to zero, do we actually still need software developers? Who thinks we
need software developers? Yeah. Who thinks we don't need software Okay. So, okay, there are few. Uh I got in big fight on LinkedIn around this topic. So, I I was uh Most people said said that it's fun to write code and they want to continue doing that. this is sort of a career path that some predict. then people say the whole SaaS companies are go bust because
no one cares about it about SaaS that everyone can create those financial products very easily. Maybe. So, that's if you haven't heard about uh SaaS apocalypse, that's the thing when uh 300 billion dollars evaporated from uh from their shares prices. Um but if we look into it, everyone sort of agrees that when something becomes cheaper, we do more of it. And we do more of it to
compensate to actually lost revenue. So, we actually we do so much more, we actually spend more money and time on it. And if we consider that, so then this naive way of thinking that if we use AI and we can go 10 times faster, so we can solve all the problems that we have right now in 10% of the time, we are going to fire 90% of
developers. This is a very naive way of thinking about it. And I'm try to explain why. So, obviously we do see that English is the most valuable programming language now. That that's pretty obvious. And Uh, we see across the board, CEOs and uh and tech leaders, they say that most of the code is going to be written by AI. And some companies have a policy that you're
not allowed to write code directly uh without AI. And um there are a bunch of quotes, some of them are older, uh but there's generally consensus that uh AI is going to write code and we're going to instruct it. Same with Elon Musk. Some people say it's better if we're going to be welders or plumbers or something like that because that's not going to be replaced And
there's always one person who says it's never going to happen. >> [snorts] >> And another bit of uh yeah, sorry. But how? Right? What's what's the Is it actually going to work? Now, there's a general direction that development becomes less of a writing code and more of architecture and planning and uh and system thinking. what is again pretty clear, and this this is research by Anthropic, is
that many different disciplines are going to be affected by AI. Um surprisingly, two two things that people didn't see coming is software development and management. Those are probably the most affected by AI right now. But it's coming. but then looking a bit deeper into this, what we can see is that right now, this is how right now software development is done. So, there are micro solutions in
the beginning, can be done by junior developer. What what is micro solution? It's like Excel macros or or really simple scripts. So, it's somebody so somebody who has a bit of development background can do that. But you still need a bit of development. You still need to understand code. And then we have a bit more complex tools like uh uh like portals or websites and more complex
tools like platforms. So, booking.com is pretty big website, but conceptually it's not very complicated, right? It's a platform to sort of share properties or buy hotel rooms. And then we have sort of mega systems like Google or Facebook or AWS. Those are really really big. There are not that many of them and they're really hard to build and we have thousands of people working on building them.
Now, and then there are certain sort of systems that we cannot build because they're too complex. They're unachievable with current means. And uh what's going to happen likely is that this whole thing going to shift. If we can build 10 times faster and we don't need development background, it's very likely that those citizen developers, those are non-developers, people who don't understand coding at all like our CFO,
they will be able to build certain portion of software, but unlikely the entire stack. And the same 100 developers will be able to build systems like Google Cloud or like Facebook. But what 1,000 developers are going to build? Maybe those mega systems that are not achievable to us today. Something like climate simulators or digital city twins. So, imagine a system like SimCity that manages a real city
in real time connected to all life systems in entire scale of the system of a city. Um and and think of it of it in levels of complexity. So, there is complexity from putting a shelf on the wall, right? And building a shed, building a house, all the way to like Burj Khalifa of software development. And each one of these levels of complexity de- require different tooling.
so, if you have uh who has a drill in the house, electric drill? Right, most people have something. So, you can likely put a shelf on the on the on the wall, maybe build a a shed. So, let's talk first about these citizen developers. What what are they going to do? So, first there is a bunch of tools that are um very much appropriate for those non-developers.
So, Lavavel is perfect example. You you get in, there is a prompt, there is no code involved. It goes all, it's packaged, delivered directly to to the cloud. So, there is no understanding of how things work. And of course, you can, you know, you can make a joke about how all the credit cards data is going to be leaked. And and that's true, right? That's true. If
if a developer goes too far and does some stupid things, of course bad things going to happen. But also Lavavel increasingly invest more and more time in security and making sure that those things happen less. But this tool is is really powerful. It's pure um pure prompt, right? So, there is no ac- actual software development skills. The other set of tools are around process of automation. So,
this is particular blog post from Michael, our co-founder, um about NA10. It's quite famous tool. It It automates the the processes inside company. So, if you have like marketing or financial or some other processes or scheduling of something. So, this tool is really really good. It can So, this blog post is about automatically answering to emails. So, he built this flow in NA10. It scans Gmail and
understands from the previous emails what's the context and writes the draft of email. He just needs to approve and click and it goes out. So, this kind of tools today probably you don't need any 10. You can use also code co-work and some other tools. So, there's bunch of tools like that. Another one really good one. I really like this demo from uh Camunda. Uh it's like
15 minutes demo. Really cool about getting a loan and uh and human in a loop and some more advanced workflows. Those things don't require actual software development skills. They require understanding of the business and configuration and not even configuration, prompting and talking to the system. what I see we start seeing right now is this all explosion of development, all those citizen developers, all these people who who
are not developers, they're very likely going to work on the top layer, on on the last mile of software development, on consumption of the software. Right? So, custom front ends, custom different configurations, custom applications, everything's hyper personalized and connected across multiple systems. It's very unlikely that a citizen developer who doesn't understand software can build a large-scale platform to actually manage uh complex systems and run in distributed
way. So, [snorts] let's talk about the second part. How many of you has a crane in the house? Like a crane. No? very unlikely, right? So, uh this this is sort of a point. There is set of tools that although they are available and they they exist in the market, but they're not approachable for non-specialists. Uh so, this is this is a slide we're using in our
training about uh AI coding a sort of maturity of code coding with assistance of AI. Sort of the last point is a last few points is about software factories. So, there are variations of a regular factories, dark Do you know what what does it mean dark factory? Why what's the point of it? That's a rhetorical question. I'll answer it anyway. The dark factory is is a factory
where only robots work without humans. So, you don't need lights. Robots don't need lights. So, you entirely fully automated without human involvement. So, the the potentially the the goal is to create this kind of dark factories that are sort of churning software on large-scale, reliable, consistent, and repeatable ways. there's few implementations of this kind of factories. One of them Gusteau. These are real visuals from the blog
post from Steve Yegge who actually built it. It's very cool read, very interesting system. This this is my interpretation of architecture using AI. The idea of of Gusteau is to orchestrate swarms of agents. So, it's a replacement in Steve's language, it's a replacement to ID. So, instead of coding in ID, you have bunch of specialized agents that each one does job like architectural planning, architecture development, testing,
security. And then you the job of Gusteau is to orchestrate those agents. To make sure that at the end you actually have appropriately built functional software. And well tested and everything. Now, this tool is is beyond the the capabilities of of a non-developer. That requires really complex setup. It's actually quite difficult to to manage. But the results of of using those tools this tool is very early,
so I'm not sure it's production ready and but there are a bunch of other tools like that. So, Lucky mode is another tool which is quite similar. It also manages bunch of specialized agents each one with different responsibilities. And eventually it can create sort of flow of orchestrated agents that all together build software. And we as humans need to make sure that we express our our desires
intent and and everything and then just make sure that they're building the right thing. at the end the goal is to to really create AI agentic systems that build code that that is reliable and we can use it. And there's a bunch of other tools. Open AI SDK is it's totally different concept doing something else, but those are also complex tools and there is another one recently
saw in in a meetup elastic platform so which is also complex setup to to build larger scale software systems. Actually, before me this is the presentation GitLab that was before me. They're talking about exactly the same thing. So, larger scale complex for professional software developers. And then once we get there, then they're all kind of weird questions start popping up. Is the are the microservices still relevant
architecture? Should we move more into even smaller agentic swarms or the opposite to this sort of massive monolithic architectures massive uh hyper monoliths because but I don't know this this this are difficult questions that we are still looking for answers because we don't know how best to leverage the AI systems to build Another thing is Scrum. It's quite obvious that Scrum is breaking down. It's not a
good management or process for agentic software development. Because multi-week sprints are sort of irrelevant because the the the time it takes to build piece of software in the past it took few hours to plan and then weeks to build and test and then and that's why you you needed this one or two weeks of sprints. Now it takes few hours to think then you give it to
AI and it builds it in few minutes. It's not perfect and we're only starting to do it this way but but still the cycle of development is much more heavy on thinking and less on on typing the code. So the the current systems of managing software development also breaking down. generally what we really want to get to is some sort of intent or spec-driven development. So what
we really want to get to is when we as humans express intent because AI doesn't doesn't want or need anything. It only does what we tell it to do. So we need to be very specific about expressing our intent and invalidating the results. And that's where the AI needs to also allow us to to actually do those things. Everything else eventually is going to be done by
by the machine. really the question is longer term. Can we actually get to this level of industrialization with software where we can divide a piece of software or a full system into pieces of consistently repeatable well-speced, well-tested, validated pieces of software that we can sort of combine together as a Lego. When we are in that situation, then then we stop being sort of craft and become an
industry. And in this way we can create we can create this kind of specific specs for pieces of software and then create assemblies around that using those uh sort of from the boat to the engine, right? And then eventually create supply chains of the software. Now I I don't know if that's achievable or when it's achievable, but uh but that could be one of the directions AI
is just a trigger for a much bigger change. Right? And then the last part is uh is about the transformation itself. How we actually move from the current way of building software towards the AI native way. And every change comes in in uh in waves. So, every technology comes in waves and uh the the sort of cognitive revolution that brought by AI is most comparable to industrial
revolution because of the scale of replacement, right? So, uh massive amount of people are replaced were replaced by by engines and move into cities and uh to factories and then later to service jobs by a tractor, So, now we sort of have a machine that replaces cognitive jobs. Um and the waves concept we are using this this is a a slide from my colleague's presentation from over
10 years ago. So, that that is a presentation we used for promoting microservices and cloud native. And uh there's another uh another wave. This wave is interesting. It shows that AI grows faster than others, but this is also not new because the previous innovations like internet or cloud, they also were faster than the innovations before. The problem is that every wave of of innovation comes with innovators
dilemma. So, the question is it's obvious that the next wave will be better, but for every company, for every person, it's a question where to jump to the next wave. If you're jumping too soon, then the technology is not ready and you're you're not leveraging the previous technology to get maximum out of it. If you try If you jump too late, then uh uh you're too late
and you're losing the competition. So, that that's sort of the story of the book that I mentioned in the beginning that all these waves, they come in uh consistent, repeatable, predictable ways and uh there is a specific way of actually going around that, which is start small, build the MVPs, few few teams and then expand from there. And uh what we see in real life when we
transition to AI native or start adopting this that it always exposes new bottlenecks, right? So, when a developer start writing code much faster, what happens is that uh then the the product owners are sort of lost, right? Well, what do we do? If the developers are going through the backlog much faster, then they need to create more tasks, but if they create more tasks, is it the
same tasks or they need something to build something more complex in the next level? we find out that actually if you don't have a good cloud native setup with continuous integration, delivery and uh and with proper cloud platform and proper testing, it's really hard to go to AI because AI actually speeds up everything. And if you don't have good good infrastructure to speed up, then you you're
speeding up bad things, which is actually counterproductive. that's it. Again, if you like the free copy, you're welcome to get one and I think I'm on original time despite the broken laptop. So, I think the next talk will start soon. So, I don't think we have time for questions. I'm not sure. Yeah, you're welcome to ask me. I'll be here. Yeah. Thank you.
