OpenAI Codex Sites and Cursor Canvases are now in direct competition for vibe-coding workflows, both targeting users who want to generate and deploy web content without leaving their editor. DeepSeek V4 is the number that matters this week: its cost undercuts both Claude Opus 4.8 and GPT 5.5 while benchmark performance closes the gap enough to make the price difference hard to justify for most agent workloads.
Apple approved Poke.com ahead of WWDC, which is the clearest signal yet that iMessage is becoming an agent distribution channel. The Hermes Super App segment is worth your attention specifically because it frames the consumer-facing layer of agents, not the model layer, as the next platform war. Anthropic's pause on new model releases gets coverage here too, with context on what that signals about their internal roadmap.
The reason to watch the full video is the section-by-section breakdown of where each product sits on the spectrum from power user tool to mass-market app. Riley Brown works through Codex Sites versus Cursor Canvases as a concrete feature comparison, not a vibe, and the DeepSeek cost analysis is specific enough to inform actual vendor decisions. The Microsoft segment closing the episode adds one more data point on enterprise agent infrastructure that the headline does not flag.
[WATCH ON YOUTUBE →]