I've been building with AI for about 18 months now. And I'll be honest, it's changed everything about how I work.
But here's the thing. I don't treat every project the same way. When I'm prototyping a quick idea or building a personal tool I'll probably never maintain? I vibe code it. Just me and the AI riffing, seeing what comes out. It's fast, it's fun, and honestly, it's kind of addicting.
For production apps? The stuff I actually care about? That's a completely different process. That's AI-assisted engineering. Real planning sessions, careful reviews of everything the AI generates, thoughtful architecture decisions. The AI is doing the typing, but I'm doing the thinking.
And that distinction matters more than most people realize.
Coding Used to Be a Scarce Resource
For most of the software industry's history, coding was genuinely scarce. It was slow. It was tedious. And it definitely wasn't for everyone.
Some people found it too hard. Others just couldn't stomach sitting there for eight hours a day writing code. And honestly? I get it. This job was reserved for those of us with enough skill and, more importantly, enough patience to survive it.
Because let's be real. We're talking about a career filled with hundreds of hours spent hunting bugs that turned out to be typos. A single = instead of ==. Floating point nonsense. You know the drill. Not everyone could handle those late nights staring at a codebase that "just doesn't make any sense," refusing to give up until you found that one line buried three files deep.
That patience combined with skill is exactly what made coding scarce. And scarce things command high prices. That's why developer salaries got so inflated. Not because the work was always brilliant, but because so few people could actually do it consistently.
That Era Is Over
Coding isn't scarce anymore. Full stop.
AI agents can write code as well as most developers. Often better. And they do it at a pace that even the most talented engineer can't touch. That effectively makes code free.
So by extension, all software should be free too, right?
Well, no.
Code by itself doesn't solve problems. You need the right code. And this is where things get interesting.
The Difference Between Coding and Engineering
This is what historically separated great engineers from good coders. A great engineer doesn't just write code that works. They take time to understand the problem, the business, the product, the users. They make sure the technical solution actually fits.
And early on, a great engineer fights to keep things simple. When there are still a ton of unknowns and the product direction hasn't fully emerged, the goal is velocity. The ability to iterate fast, add features, remove features, change direction entirely. That requires a codebase with as little complexity as possible, so changes can happen quickly.
This is exactly where vibe coding falls apart.
A vibe coder will iterate on the product, sure. But there's no architectural direction behind it. The AI just solves whatever problem is right in front of it, one prompt at a time. Layer after layer. What you end up with is code stacked on top of itself with no cohesion, no plan. Some people call it code slop, and that's exactly what it feels like. It leads to bugs, weird side effects, and eventually the product grinds to a halt because every small change breaks something else.
Velocity dies. And that's the whole game.
Rebuilds Are Finally Possible (But They're Not Magic)
Here's something that's actually changed in a meaningful way. In the old days, rebuilding a working product was almost never worth it. The time, the cost, the risk. Those projects rarely finished. Because the thing about real software with real users is that 95% of it is edge cases. Years of fixes, patches, and workarounds that accumulated while you were busy solving actual customer problems. Most of it undocumented. Just buried in the code.
In the AI era, rebuilds become feasible. The coding part isn't the bottleneck anymore. AI handles that.
But here's the catch. If the engineering isn't done right on the rebuild, you'll end up in the exact same mess. Different code, same problems. That's where engineers still matter. They're the custodians of the software. Non-engineers simply don't have the experience to maintain complex products through the iteration cycles that real software demands.
So What's Actually Valuable Now?
Coding isn't the activity that commands those high salaries anymore. Engineering is.
Engineering quality code to solve real problems in a way that maintains velocity. That's the single most valuable asset a software company has. And it's a skill that's still in high demand and still in short supply.
Most companies don't even know how to hire for it.
So Is Coding Dead?
No. Coding is very much alive. The patterns matter. The architecture matters. Your choice of languages, frameworks, and libraries matters. What doesn't matter anymore is the act of typing it all out yourself.
My Advice to Developers Right Now
If you're a developer struggling with where things are headed, here's the most important thing I can tell you: embrace AI agents. Learn how to multiply your productivity with them. You can still reach a solution that's 100% what you would've coded yourself. You just have to get better at communicating what you want.
Get good at planning with the AI. Get good at reviewing quickly. Create skills that capture your patterns, even the nitpicky ones you're slightly embarrassed about, and invoke them on every commit. But always review the work. Review the architecture. Review the code. Especially if you have customers, real revenue, or sensitive data flowing through the product.
The degree of review depends on risk. A side project? Maybe you skim it. A payment flow handling customer credit cards? You read every line. But leverage AI at every step. Use your actual skills to build a cohesive architecture that models the business well. And above all, maintain velocity. Maintain simplicity to the extent possible.
Vibe coding alone can't do this. Humans still need to be in the loop, especially when you're not just copying an existing product spec. Code still matters. But the developer's focus needs to shift from writing code to engineering it.
The Era I'm Actually Excited About
As AI agents get more embedded into the fabric of every company, software will get better. How could it not? Code is effectively free now, but quality products are not, and creativity can move at a pace it never could before. More experiments can be tried. More ideas can be tested.
But this means engineering matters more, not less. We're not going to be building copy-pasted solutions. We'll be inventing new ones. New product patterns, new features that haven't been done before. Your ability to plan and communicate that properly, your ability to maintain a product that feels simple on the surface but handles incredible complexity underneath, that's going to require engineers who really know their craft.
I'm excited for this next era of software. I've been living it, and it's the most fun I've had building in years. And you should be too. Just don't fall behind.
Don't be a naysayer. Remember, plenty of people were skeptical about the internet. Then mobile. And plenty of people missed the boat on both. AI agents are the most economically transformative technology yet. It will work, if you learn to use it well.