The dopamine is flowing with OpenAI Codex today. I've got Claude and Codex reviewing each other's work systematically and while progress seems slower than the heady days of any early project, it feels like every move is way more considered and the output is improving code quality. A lot of my current work is developing using frontier models to build applications and harnesses using self hosted LLMs and RAG. My next step is going to be to bring as much of the coding "in house" as possible.