AI nie nadaje się do dużych systemów IT? | Kuba Kubryński

👉If this episode resonated with you and you want to go beyond just "using AI," we have good news for you! The DNA: Dear Modern Architect team is back with a new training course. More information is available at: https://architektjutra.pl/?utm_source... Is your system AI-ready, or does it just "work" thanks to humans? Many people claim that AI excels in large, complex systems. However, the truth is much more brutal: AI isn't too weak for your code—it's your code that often has a "rotten" architecture, masked over the years by the team's "tribal knowledge." In this episode, we discover why AI has become the world's best tester of architectural debt. This is a key lesson for anyone who wants to transition from the role of a "code whiz" to an architect building systems ready for the new era of automation. 🎙️ Kuba Kubryński - a systems architect and entrepreneur operating at the intersection of technology, business, and the IT market. He is the co-founder of DevSkiller and today focuses on the impact of AI, reskilling, and technological change on the role of programmers and the long-term stability of companies. Table of Contents 00:00:00 - Is AI really struggling in large systems? 00:00:45 - Rotten architecture and "tribal knowledge" as a blocker to development. 00:01:34 - Why AI is like a new employee (onboarding in 5 minutes). 00:02:15 - What happens when business logic is "smeared" across code? 00:02:54 - Three ways of calculating the same thing - how AI exposes the lack of a source of truth. 00:03:52 - AI as a litmus test for architectural debt. 00:04:40 - How to save tokens and money with clear facades. 00:06:05 - The evolution of the architect's role: designing for AI agents. 00:06:46 - Homework: how to test your system with LLM. 00:07:47 - Where to start? (Spoiler: not with prompts). What questions does this episode answer? 💡 Why does AI "break" things in legacy code that you think should work? 💡 What architectural features prevent a language model from guessing and improvising? 💡 What is the difference between a system that "works thanks to humans" and a system with good architecture? 💡 How can you use AI to quickly verify the quality of onboarding in your project? 💡 Why are modularization and clear contracts the only way to combat limited context? 💡 Will your role as a programmer change into a "designer for automation"? In this episode, you'll learn: ✅ How to set clear module boundaries to make them readable for algorithms. ✅ Why "hidden knowledge" in developers' minds is the biggest enemy of AI implementation. ✅ How to avoid confusing responsibilities that leads to refactoring errors. ✅ What is "truth serum" in the context of system architecture? ✅ How to use facades and contracts to optimize token consumption and the cost of working with AI. ✅ Why errors we attribute to AI are often flaws in the structure of our code. Check out our other materials – each one is equally substantive and addresses specific problems you face every day. If this video opened your eyes to architecture, the next ones will likely turn your approach to development upside down. We put a lot of energy into creating this content, so take advantage of it – especially since it's completely free. ▶️ Be sure to check out other episodes on the channel ▶️ Are programmers threatened by AI? The truth about the IT market | Jakub Kubryński    • Czy programiści są zagrożeni przez AI? Pra...   ▶️ Why aren't seniors good at AI? | Kuba Kubryński    • Dlaczego Seniorzy nie umieją w AI? | Kuba ...   And if you want to learn how to make such architectural decisions in a structured and informed manner, check out the course: https://architektjutra.pl/?utm_source...