You know that rush when you build something huge in a weekend? That feeling of invincibility? It’s a trap.
Every developer I know has felt it. You fire up Cursor, Claude, or Copilot, and suddenly you’re shipping a custom database, a game engine, or even – God help us – your own cryptography. The AI writes the boilerplate, you stitch it together, and it works. For now.
But here’s the truth nobody wants to say: The easier coding gets, the harder it becomes to trust what you’ve built.
I’ve been watching this unfold for months. The agentic coding tools are lowering the cost of ambition. Good. We should build more. But they’re also bypassing decades of hard-won engineering discipline. Bad. Very bad.
Let me be blunt: we’re creating a generation of cowboy coders who can write code faster than they can understand it. And when something breaks – and it will break – there’s no audit trail, no mental model, no testing framework that can save you.
I saw a comment on a recent post about custom cryptography. The writer said, “As someone not in the know about cryptography, I know just enough that the words ‘custom cryptography’ scare the hell out of me.” That’s the voice of someone who still has a functioning survival instinct. But the AI doesn’t have that instinct. It will happily generate a broken cipher if you ask nicely.
Here’s the twist: the very tools that make you feel like a 10x developer are the ones that will turn you into a maintenance nightmare. The shortcut today is the quicksand tomorrow.
I’m not saying stop using AI. I’m saying stop pretending you’re above the rules. You can’t skip the boring parts – testing, code review, domain expertise – and expect the exciting parts to survive.
So what do we do? We bring back the gatekeepers. Not the old ones who hoarded knowledge, but the ones who say “show me your tests” before they say “nice idea.” We build tools that audit, not just generate. And we teach developers that the goal isn’t to write code fast – it’s to write code that lasts.
Because the alternative is a world where every ambitious startup is running on a house of cards, and the first real wind will bring it all down.
FAQ
Q: Aren't we just being dramatic? AI tools are just productivity boosters, not a threat to code quality.
A: Productivity boosters are great when they amplify good practices. The problem is that these tools amplify bad practices too. They let you skip the boring but essential steps – tests, reviews, design docs – and that's where the danger lies. The speed is real, but so is the fragility.
Q: What should I do differently if I'm using these tools?
A: Treat every AI-generated block of code as if you wrote it drunk. Review it properly. Write tests for it. Get someone else to review it. And never, ever let the AI write your cryptography. The tools are enablers, not replacements for your judgment.
Q: Isn't this just the same old 'new technology is bad' argument we've seen a hundred times?
A: No – this is about the specific nature of agentic coding. Previous tools couldn't generate entire systems on a whim. This one can. The risk isn't new technology; it's the scale of ambition that low effort enables. The more you can build in a day, the more you can break in a year.