Dev Reflects on AI Coding: Models Make Basic Reasoning Errors, Hand-Coding Wins

jsuarez · x · 2026-08-13

The author shared recent experiences using Grok 4.5 and other models for coding. While Grok 4.5 is faster, models still make basic reasoning errors on extended research tasks (like deleting control conditions) that invalidate hours of work.

Frustrated by these failures, the author is reverting to manual coding for core tasks. He now limits AI to extremely narrow refactoring jobs or treats it as a tool for occasionally surfacing manually verifiable results. He advises new developers to learn foundational coding (e.g., writing C by hand) rather than relying heavily on AI.

Related event: Developer Tests Reveal AI Models Still Prone to Basic Errors(3 posts)→

Original post →

More from coding & agent

coding & agent channel →