The episode opens with Cerebras's successful IPO, highlighting its wafer-scale engine's speed advantage for inference due to integrated memory, though it notes limitations for large context workloads compared to NVIDIA. The discussion then shifts to managing AI agent swarms, suggesting a span of control of three to eight agents per operator, mirroring firefighting doctrine. A Figure humanoid robot livestream is reviewed, followed by a personal account of an AI voice-cloning scam, leading to a recommendation for family code words. The cybersecurity segment covers Microsoft's use of a 100-agent swarm to counter the Mythos threat actor and a separate exploit of Apple M5 memory integrity by Mythos. The episode concludes with news of Recursive Superintelligence raising $650 million for self-improving AI and a warning from Geoffrey Hinton, emphasizing the need for practical security measures like code words and verification protocols.