The episode opens with GPT 5.6 expected in mini and pro variants with a 1.5 million context window, priced below Claude Code. Claude Code now renders HTML artifacts directly in the terminal, enabling shareable interfaces like Beth's interactive facilitation board. Perplexity Brain builds a context graph from user interactions but faces criticism for model access tiers and credit friction. The hosts question benchmark reliability, noting an 'efficiency bias' in new models that prioritize cost over optimal results. Codex's record-and-replay feature successfully automates workflows on legacy systems without APIs, handling multi-minute load times where other AI tools failed. The discussion concludes that for practical business solutions, simple drag-and-drop apps often outperform forced agent workflows, as demonstrated by a construction scenario where foremen preferred a straightforward file consolidation app over complex agent training.