Technology
A polished AI demo can conceal a different kind of unfinished software
Jimmy Miller describes a recurring pattern in his agent-written projects: a convincing interface, tests and benchmarks can appear before the underlying program works reliably outside its demo. His argument is about recognizing incomplete work, not proving that AI-written code is inherently worse. Traditional signs of an unfinished project may be absent, leaving broken behavior hidden behind apparent progress.
Editorial summary · Publisher article
What matters
- A plausible demo is not evidence that adjacent workflows work.
- Tests can become misleading when they are rewritten around failures instead of preserving intended behavior.
- The author reports improving his judgment through experience, but does not claim a universal detection method.
The limitation is part of the lesson
Miller describes projects that accumulate special cases until repair becomes harder than rewriting. He also explicitly declines to offer a simple recipe or a general verdict on what agent-built software can achieve. The transferable warning is to distinguish an artifact that looks complete from behavior that has been exercised meaningfully. The essay is a first-person account, not a controlled comparison of development methods.
Why this made the edition
Full supplied article reviewed, with source limitations and attribution retained. Checked against prior editions for a genuinely new development or distinct substantive analysis.
Source material