Why LLM-driven software rewrites hit a wall: kernel breaks compat, compat blocks provability
jd_pressman · x · 2026-10-23
jdpressman discusses with lusichu the practical difficulty of LLM-assisted code hardening: application software is nearly disposable, so a 'toothpicks and tweezers' approach there is insane, and dependencies remain in C, limiting protection. Starting from the kernel breaks downstream compatibility for all other software; insisting on compatibility forces hard-to-prove programming patterns. His takeaway: only by actually doing it with an LLM can you see the real difficulty.
More from coding & agent
- Railway and OpenCode breached as attackers hunt for your AI API keys — thursdai_pod · 2026-10-23
- Open-source browser agent Jev cuts median task time from 9.45s to 7.09s — abhishek__AI · 2026-10-23
- Onyx: open-source enterprise AI chat and search platform with 32.9k GitHub stars — adnan_hashmi · 2026-10-23
- Hermes Agent hits 395 real user stories; one IT staffer runs 80% of a 165-person company's work — NousResearch · 2026-10-23
- AWS makes Continuum vulnerability-fixing AI agent generally available, tops CyberGym-E2E benchmark — bookwormengr · 2026-10-23
- Running parallel Claude sessions means rebasing n-1 sessions after each merge — tristanbob · 2026-10-23