OpenAI merged ChatGPT and Codex into a single product and released GPT-5.6, putting agentic AI capability in front of roughly one billion users. The new model tier introduces three variants named Sol, Terra, and Luna, and the pricing structure is described as competitive enough to change the build-or-buy calculation for most companies.
The conversation between Riley Brown and Ras Mic gets specific fast: they benchmark GPT-5.6 against Claude Code and Fable 5 using the Replit and Lovable test, walk through real Codex workflows including agent loops for marketing automation, multi-thread execution, social media scraping via plugins, and a record-and-replay feature for knowledge work. The building-block economy section around the 30-minute mark is worth watching closely: it frames tools as composable primitives, not features.
The argument the hosts are making is structural, not promotional. Every company that does not redesign its software around agent-native architecture is building technical debt against a new default. Computer use, updated browsers, and a new Sites feature point toward an agent-first operating system layer that most users have not encountered yet. Read the full breakdown to understand where the workflow boundaries actually sit today.
[WATCH ON YOUTUBE →]