'Agents Need Control Flow, Not More Prompts' Hits 470 Points — Practitioners Converge on Deterministic Scaffolding
Hacker News·high signal
A blog post arguing that reliable AI agents require coded control flow rather than prompt engineering hit 470 points and 228 comments on HN. Practitioners shared concrete evidence: one QA agent that missed files and took 10+ minutes became reliable in 3 minutes after switching to a deterministic harness. The top-voted consensus is that LLMs should be treated as components within coded workflows, not as the orchestration layer itself — 'use the LLM to write code, not to execute workflows.'