Coding with AI is like building a plane. You make the specs and you give them to Claude Code. Then you watch it build the whole plane in half a day where it would have taken you weeks, boy it is beautiful.
You fly your plane and it’s amazing.
One day though, there is a problem, maybe when you cross the equator, the plane starts acting strange, or maybe when the wind exceed a certain strength… So you need to investigate what happened to fix it, but you have no idea how the plane works exactly, you just how “overall” how it works. So as a good vibe place builder, you ask Claude.
To investigate, Claude does not take every piece of the plane out to make sure it works, it builds the biggest plane-diagnostic machine mankind has ever seen, in 30min. Which you understand even less than the plane, because you are not a plane-diagnostic machine expert.
But the machine produces more reports and logs that you can read. So you trust it as well.
Thing is the plane is not really getting better. Since you have no idea how the plane works, no idea how the machine works, you cannot even really interpret the data the machine outputs.
So you face the dilemma of either you trust 100% what the machine is saying, which you don’t since it is now obvious the place is not getting better, or you go back to craftmanship in caves like your ancestors did.
After days of suffering, you decide to take your screw driver and you remove every piece of the plane to understand what goes wrong. You feel the joy of building and fighting to understand how this complex system works again. And soon you find the piece that causes the problem. It is a completely different problem that what the machine told you it was (reading all those reports was a waste of time), and on your way to fixing the piece you uncover a few other problems that you were lucky did not happen before.
You look proudly at your screw driver, so simple and good at its job, a tool you can rely on 100%.
Welcome to the new software engineering world.