Why Software Factories Fail (or: harness engineering is not enough)
Points and comments are a snapshot, not live.
Article body wasn't reachable. The HN discussion summary is below.
Points and comments are a snapshot, not live.
Article body wasn't reachable. The HN discussion summary is below.
What commenters are saying
Commenters agree that LLM-assisted coding requires high discipline and clean code, not just skill. One commenter argues that reinforcing the data quality principle is critical: "AI works on data. Code is data." A former StrongDM employee notes the company was sold after the dark factory announcement; the CTO started a consulting firm, which another commenter interprets as a sign of product weakness. A commenter from the StrongDM AI Lab defends the weather report's update frequency.
Two camps emerge on constraints: some advocate for formal verification and programmatic checks, while others propose a taxonomy of constraint mechanisms (generative, interpretive, elicitative). Another commenter criticizes GitHub's PR review UX and praises Linear's AI-driven grouping and commentary. A dissenting voice argues that code review is futile since agents can automate it, and suggests reviewing working software instead.