The real AI coding crisis: no one understands the system anymore
Original: The problem is not AI code, but not knowing about system architecture or intent
Why This Matters
Loss of system-level knowledge compounds technical debt faster than AI can generate code.
As AI tools like Claude generate entire codebases, a growing concern among engineers is not code quality but the loss of architectural knowledge and intent. Teams ship fast but no one understands why design choices were made, leaving systems increasingly unmaintainable.
Engineer and blogger Simon Späti argues that the real danger of AI-generated code isn't average quality — it's that entire teams have stopped understanding their own systems. He cites a viral account from an engineer at a large company where every artifact — specs, tickets, tests, PRDs, code — is produced by Claude Code. Engineers from L1 to L7 are reportedly working 12-13 hour days just to hit Enter, with no one reading, reviewing, or genuinely thinking through what's being shipped. 'There is no sense of victory. Nobody is resolving bugs. In reality, nobody is thinking anymore,' the post reads.
Späti notes AI likely raises below-average codebases to average — useful, but not transformative. The sharper problem is institutional: when everyone defers to Claude, no one builds the mental model of why a system is structured the way it is. Data engineering, he suggests, historically required deep product and domain knowledge. AI now makes that knowledge 'seemingly obsolete' for newcomers entering the field today.
The piece also touches on product managers: AI lowers the barrier to building, but without architectural fundamentals, early choices — wrong language, wrong mental model — create fragile foundations. Maintenance, Späti concludes, remains the 'final boss' that punishes systems built without intent.