AI coding assistants are making us worse at system architecture
by•
AI can write code ridiculously fast now. That part is obvious.
What I’m more interested in is what happens when we get too used to that speed.
If I can ask an AI to build something, see an error, paste the error back in, and have it fix the problem in seconds… there’s a lot less incentive to stop and think about why the system is designed the way it is.
Sometimes I wonder if we’re getting better at building software or just getting better at getting software to work.
Have you noticed this too, or am I overthinking it?
8 views
Replies
I write few rules that must never break as tests, and keep a short file of those rules that the agent reads at the start of every session. In a flashcard tool I build, one of them is "a credit is spent by a single conditional UPDATE, never read-then-write". The agent can rewrite anything around it.
When a quick fix tries to cross that line, a test goes red. That's the moment I actually stop and think about the design.