Everyone Can Drive. Not Everyone Can Drive Well. Same Goes for AI-Assisted Coding

Table of Contents Overview AI Didn't Remove the Skill, It Relocated the Skill Vibe Coding Is Easy. Doing It Well Isn't Everyone Can Drive. Not Everyone Can Drive Well. The same thing happens with AI It's Not Just About Prompting Overview Lately, I've been writing a lot of technical-heavy articles and I noticed my articles got less popular. Well, what did I expect, technical articles can be a little... boooooring. So this time, let's take a different route. A lighter, more easy-going article, something less about complex concepts and more about sharing ideas, stories and perspectives. Time to win some readers' hearts back... But if you're a bad driver, you might hate me after that. AI Didn't Remove the Skill, It Relocated the Skill I've heard people say that you don't need any skills to use AI to write code. ❌ Errrrr... Wrong. Stick around. I'll explain it with a car-driving analogy. Vibe Coding Is Easy. Doing It Well Isn't Even if you're vibe coding, which is probably the most abstract way of creating software, you still need skills. Just a different kind of skills. You need to know how to use your tokens effectively so you can ship products faster, safer, with better quality, and without setting your wallet on fire. That's what I call "hypermiling your tokens". There are actual techniques for using your tokens intelligently rather than just throwing more of them at the problem. And those techniques are harder to master than you might think. Not because they're particularly complicated to understand, but because mastering them takes a ridiculous amount of experimentation. You try different approaches. Different prompts. Different models. Different levels of context. Different tasks. You screw things up (inevitable). You spend way too many tokens on something that should've taken half as many (also inevitable). You ask the model to fix something it didn't even break (probably with some attitude as well). Then you ask it to fix the fix (recursion anyone?). Then you realize the original code was fine (but your top-notch model has already inhaled your entire token budget). Congratulations. You just spent 200,000 tokens going in a circle. But, don't get discouraged, because you also started finding that sweet spot where you're getting the most output for the least amount of waste. And that's a skill. Anything that's earned through trial and failure, is a skill. Simple as that. Everyone Can Drive. Not Everyone Can Drive Well. Who would you call a good, skilled driver? A skilled driver is someone who knows the rules and follows them, while also trying to be as efficient as possible. They don't overspeed. They park accurately, between the lines. They can drive inside a lane consistently. They use just enough fuel to get where they're going. They're predictable on the road. They care about the car. They care about other drivers. They don't floor the gas pedal from every single traffic light like they're in a drag race. And they don't slam on the brakes like they're playing Forza with a keyboard. A trully good driver knows how to be both safe and efficient. And they might even have no idea how engines or transmissions work and still be great drivers. I mean I'm not a mechanic, I've been driving for about 7 years now and I've never gotten a ticket, or maybe once, but that was unintentional and once I realized I'd crossed the wrong line, it was already late. And I've seen too many car mechanics who drive like they're gonna spawn again like an NPC after they crash. Now imagine the opposite. You go 30 km/h over the speed limit. Cut people off. Floor the accelerator the millisecond the light turns green. Brake so hard your passengers start reconsidering their life choices. Meanwhile, you're wearing out the car, burning fuel, and spending six times more money than you need to, while also being a danger to everyone around you. You're not a skilled driver. You're just driving recklessly. In fact, you probably have no idea what good driving actually is. What being a car operator actually is. What it actually means to be a human element in the equation. And that's an important distinction. The same thing happens with AI Luckily, in vibe coding, being reckless with AI is considerably safer because, thankfully, your Claude session isn't going to rear-end a bus. The worst thing you're probably going to hit is your own token budget. But being reckless with tokens doesn't mean you're skilled. Everyone can be reckless. Whether you're driving or coding. The skill is knowing how to find that sweet spot between speed, quality, cost, and safety, and staying close to it while you're working. And that sweet spot isn't the same for every situation. Sometimes you need a huge amount of context, sometimes you don't. Sometimes a massive reasoning model is exactly what you need. Sometimes you're asking it to rename a variable and you've basically brought a semi-truck to the grocery store, not to deliver goods, but to buy them... for the day. Sometimes you should let the AI explore the problem. Sometimes you should give it very strict boundaries. Sometimes you should ask it to implement something. Sometimes you should stop and think about the architecture yourself first. And sometimes the best prompt you can write is: "Don't change anything. Just tell me what's wrong." That's where the real skill starts showing up. It's Not Just About Prompting I think this is where a lot of people misunderstand AI-assisted development. They reduce the skill to: "Can you write a good prompt?" - That's actually only one small part of it. You need to know how to scope a problem, how much context the model actually needs, when to split a task into smaller pieces. You need to recognize when the model is going down the wrong path and understand the output well enough to know whether it's actually good. And there's another skill: knowing when to stop. This is probably one of the most underrated parts of AI-assisted development (and driving too, some people just can't recognize huge red stop signs sometimes). More tokens don't automatically mean better software. More iterations don't automatically mean better software. More context doesn't automatically mean better answers. Sometimes you're already 95% of the way there, and another twenty prompts are going to make things worse. You start optimizing things that don't matter or introduce abstractions nobody asked for. You refactor perfectly functional code because the model suddenly decided that your entire architecture needs to be "more scalable." You don't necessarily need to be the person manually turning every screw anymore. Not every driver should be a car mechanic or a mechanical engineer. You need to know which screws to turn, when to turn them, how hard to turn them, and when to stop touching things before you break something. And the better you become at working with AI, the less you'll think about "getting AI to write code" and the more you'll think about orchestrating a system that produces good software efficiently. Similarly to driving, you don't need to build the engine to drive the car. But you still need to know the rules of the road, you need to understand what the pedals do. You need to recognize when something sounds wrong. You need to know when to slow down and definitely need to know that flooring the accelerator doesn't make you a better driver (it actually makes you look like a terrible driver). The same applies to AI coding. Throwing more tokens at a problem isn't a skill nor is having the biggest context window or using the most powerful model for everything. Being able to consistently get the result you need with the least unnecessary wastem, while keeping quality and safety high, is. And the more time you spend operating around that sweet spot, the better you become at recognizing it. And that's why: "AI can write the code for me" doesn't mean: "I don't need skills anymore." The skill just moved. So yeah. Be efficient. Drive safe. Hypermile. Enjoyed this deep dive? Let's stay connected! I share more software engineering insights, projects, and experiments across these platforms: 💼 Connect with me on LinkedIn 💻 Explore my projects on GitHub 💬 Follow me on X 🎥 Watch my videos on YouTube