Andrej Karpathy joined Anthropic this week. That is the headline. Beyond that, Claude Code now surfaces a multi-agent view directly in the terminal, letting developers watch parallel agents work in real time. Desktop improvements also shipped, tightening the gap between Claude and native IDE workflows.
OpenAI pushed four Codex updates worth tracking: a /Goal command for intent-first task framing, shared plugins across workspaces, Design Mode with annotation upgrades, and AppShots, which captures live app state for context-aware edits. Google announced Gemini Spark at I/O, putting on-device agents on Android phones, but the overall I/O messaging was scattered enough that the author flags it as actively confusing to parse.
Cursor shipped Composer 3.5, and the episode includes a live test with real results, not a benchmark slide. That test is the reason to watch the full video. If you are choosing between coding environments or trying to map which frontier tools are pulling ahead in the agent layer, this episode is a useful 25-minute benchmark.
[WATCH ON YOUTUBE →]