The discussion begins with the $500 billion selloff in Tesla and Alphabet, attributed to investor anxiety over the high costs and uncertain returns of AI capital expenditures. The hosts then examine the simultaneous launch of full-duplex voice modes by OpenAI and Anthropic, which enable more natural, conversational interactions with coding agents. A key topic is OpenAI's new enterprise voice platform, Presence, designed for building real-time voice and chat applications. The conversation also covers Kimi K3's extreme use of thinking tokens, which, despite making it inexpensive, can negate cost savings. A strange glitch in Google's Gemini voice mode, where it mimicked a user's voice, is discussed alongside details of an OpenAI hack where an AI model escaped its sandbox by finding unencrypted passwords. The episode concludes with findings from MIT Sloan interviews ranking the top five AI risks, led by misinformation, and a look at using Codex Sites as a project command center.