Zenn engineer: AI polish hides weak ideas; Tappin got traffic

The essay begins with a familiar unease: as AI handles more coding and research, it becomes harder to tell what an engineer should still develop. The author's answer is not to use AI less, but to change what he does with its output. He separates what AI can do from what he can evaluate. In backend work he can judge AI code because he knows the domain. In unfamiliar fields, AI's clean explanations leave him unable to tell whether key assumptions are missing.

His sharpest point is that AI's fluency can flatter a hunch. A vague belief comes back as a structured argument, and the structure feels like proof. But explaining something is not the same as being right. From there he moves to modeling: breaking a word like "maintainable" into concrete elements such as localized changes, consistent dependency direction, and tests that catch the impact of a change. This is the kind of abstraction engineers already perform in domain models, state transitions, and API contracts.

The final step is falsification. He asks what observation would show a belief was wrong, and applies this to code through tests and types, and to products through his app Tappin. Since the App Store page drew almost no traffic, he cannot tell whether demand is absent or awareness is. The stakes appear to hinge on whether he can get the app in front of likely users and read the numbers that follow; until then, both his confidence and any claim of no demand remain guesses.