I pulled ten claims from my January post on agent-first coding and scored each against public evidence: what held up, what was just extrapolation, and the review bottleneck I missed.
In December 2025, LLM coding agents crossed a threshold. My workflow flipped from 80% manual to 80% agent-driven in weeks. Notes on what works, what breaks, and where this is heading.
Create blog posts with natural language commands and voice dictation using Claude Code. From idea to published MDX in minutes - no manual file creation, no YAML syntax errors, no context switching.