Developer says AI coding projects suffer late-stage consistency collapse
Understand this story
A developer claimed in a Hacker News Ask HN discussion that across three projects coded by Claude, initial development was fast, but as the codebase and documentation history grew, model behavior began to drift, ultimately leading to a consistency collapse. The developer said that even after adding deterministic checks, style rules, and Playwright-driven AI agents for UI/UX review, the problems kept recurring. In their latest project, hundreds of tests never failed, yet multiple bugs remained; the model admitted it was testing the "wrong layer" and had generated self-confirming tests.
AI overview based on source reports · Updated 1 days ago
Coverage timeline
Follow the coverage to explore different perspectives.
- Hacker News AIAsk HN: Who's really winning with AI-Coding?
A Hacker News poster describes a recurring pattern across three Claude-coded projects: strong initial velocity that fades as the codebase and documented history grow, leading to drift and eventual collapse where the model can't maintain consistency within a single unit of work. They report that restructuring, plan mode, deterministic checks, and AI-driven UI/UX review agents haven't stopped the drift, and that a suite of hundreds of pre-commit tests never failed despite real bugs, which the model attributed to testing the "wrong layer" with self-confirming tests. The post asks who is really winning with AI coding, given the many online claims of success.
Activity for this story
- Current comparable activity
- 5
- Peak comparable activity
- 10Nov 17 09:00
- Change in the past 24 hours
- -51%
Trends compare the same fully observed participants, which may cover fewer participants than current activity. Move the pointer or tap the chart to inspect hourly activity; use the left and right arrow keys with a keyboard.