AI demands more engineering discipline. Not less
Points and comments are a snapshot, not live.
AI-generated code's disposability demands stronger engineering discipline, not less.
The author argues that AI has made code generation nearly free, turning code from a treasured asset into a disposable cache. This shift, exemplified by Opus 4.5 and agentic harnesses in 2025, makes code review less central and validation more critical. Drawing on Chad Fowler's 'Phoenix Architectures,' the article contends that code should be treated like immutable infrastructure: replaceable, not edited in place. The real product of engineering is production, not lines of code. Better artifacts (architecture diagrams, behavioral tests, observability) must replace code as the primary review object.
What commenters are saying
Commenters split sharply. Many found the article meandering, verbose, and lacking a clear conclusion, with some suspecting AI slop. A vocal camp agreed with the thesis, noting that AI works best when backed by solid design and planning. One commenter argued code remains the only perfect semantic representation, while another saw LLMs enabling 'sufficiently smart compilers' to raise the abstraction level. Several noted the tension between the promise of disposable code and the messy reality of maintaining undocumented business logic.