If it's verifiable, it can be one-shotted: an engineering workflow for AI-built systems
imdigitalashish · x · 2026-11-21
Using the Riemann hypothesis as an example of something verifiable yet hard to construct, the author argues anything verifiable can be decently one-shotted by AI. His team workflow for hard systems: start with the full picture, break it into components with clean boundaries, give each its own pass/fail check and I/O contract, and slice further until every piece is independently checkable — after that, one-shotting the implementation is the easy part. Projects drag, he says, because nobody defined how to check the parts; this process also curbs context drift.
More from coding & agent
- Windows Terminal's agentic "Intelligent Terminal" becomes a daily driver — dfinke · 2026-11-21
- Have Claude spin up a custom wandb-style training dashboard in minutes — rishabh16_ · 2026-11-21
- Jev is just a classifier — but it replaces months of ML team work in minutes — viksit · 2026-11-21
- omarsar: Stop automating everything — help your coding agents write better code instead — omarsar0 · 2026-11-21
- Dev launches remote MCP server for live Pinnacle sports odds, queryable from Claude Desktop — Significant-Sun-9735 · 2026-11-21
- The speed limit for AI science is the cost of verification, field by field — professor_ajay · 2026-11-21