Mani Doraisamy traced a Google Forms add-on’s slow welcome screen to AI-written installation logging. Successive attempts to fetch an email address and user ID made multiple network calls before the screen appeared. He reports a 15-second wait.

His proposed fixes start with deciding what the product actually needs:

  • Log the installation timestamp and tell marketing that an install alone cannot supply a follow-up email address.
  • If identity collection matters, show the welcome screen first and assess the security tradeoffs of logging afterward.
  • Specify the data model and architectural boundaries before asking AI to implement a feature. Require approval before changing them.

Doraisamy calls this an unannounced promotion into architecture: humans must negotiate requirements, choose tradeoffs, and review the implementation. The incident makes a concrete case for that responsibility. His claims that coding is going away and AI will never push back go well beyond the evidence of one debugging story.

The 138-comment thread on Hacker News adds practical warnings and direct counterexamples.

What the thread adds

  • bikeshaving — on reviewing a Claude-assisted keyboard project: “Claude would leave the debugging tooling around unflagged, which essentially made the program a key-logger.”
  • classabove — on simply asking whether code is performant: “With prompts this vague, I often see massively complicated preemptive optimisations.”
  • cluckindan — against the claim that AI never pushes back: “Multiple times I’ve asked it to do things against guidelines and plans, and every time it has pushed back.”
  • whycombinetor — on whether architecture is a new responsibility: “Am I an outlier in that every SWE job I’ve had has expected you to push back on requirements and design the data model for tickets you are handed?”

Handles are pseudonymous; HN exposes no per-comment scores and orders comments by its own ranking. This is a slice of the thread, not a consensus.