Respectfully, I disagree. > With AI, knowing when to say “NO” is just as important as knowing what to ask it to build. AI is great at giving you working solutions. But a working solution and a well-designed solution are two different things. Let’s take a simple example. You integrate Razorpay as your payment gateway today. Tomorrow, for whatever reason, you decide to move to another provider. If your payment logic is tightly coupled with Razorpay across your codebase, you’re now looking at a massive PR touching multiple parts of your system. However, if you had designed a payment abstraction from Day 1, switching providers would be much easier. Could AI have done this? Absolutely. But someone needed to think about it, ask for it, or recognise the problem during code review. And that thinking often comes from experience, from getting your hands dirty, making mistakes, and learning why certain design decisions matter. > With all due respect, building side projects and building production systems that are used and improved every single day on every layer are two very different things. The real challenge isn’t always getting something to work. It’s making sure it continues to work as the product grows, requirements change, and more people start contributing to the same codebase. Continuous improvement is the main point that differentiates. > AI has made shipping the first version incredibly easy. Maintaining and extending it is where engineering decisions start to matter. In the early days, you move fast. Extremely fast. And AI has made that even faster. But as you keep iterating, adding features, handling edge cases, and scaling things, you start realising how important a solid foundation is. How do you have a solid foundation without looking at massive PR’s in future? Experience. > Writing the initial code is often the easiest part. Making sure the next 100 features can be built on top of it without breaking everything is the difficult part. And that’s where experience, system design, and knowing when to reject an AI-generated approach become extremely valuable. I’m not saying AI cannot write good code. It absolutely can. I’m saying that getting code that works and knowing whether that code is the right foundation for your product are two different skills. If you are okay with massive PR’s, high token burn, constantly changing architecture with every feature, yes LLD is dead!

Kismat ki chaabi — mehnat!