021 - Anthropic's Mythos: The AI Model That Changes Everything

025 - Google I/O 2026

Ryan, Kevin, and Travis recap Google I/O 2026 announcements, emphasizing Google’s rapid AI push across its products. They discuss Gemini 3.5 Flash as a fast, lower-cost frontier model optimized for long agentic tasks, rolled into Search and YouTube at scale, and Gemini Omni/Omni Flash for multimodal creative generation using models like Veo and Imagen. Google Search is overhauled into a multimodal, generative, interactive results experience, raising concerns about provenance and web economics as answers bypass creators. They cover Gemini Spark, a 24/7 cloud-based consumer agent integrated into Chrome and Google apps, plus daily briefings and background “information agents.” Other topics include a universal cart via a commerce protocol, SynthID watermarking with major partners, and announced Android XR audio/display glasses with partners like Warby Parker and Samsung. 00:00 Welcome and agenda 01:36 What Google I/O is 03:06 Gemini 3.5 Flash debut 04:21 Speed cost and rollout 05:59 Google comeback narrative 09:22 Omni Flash creative suite 12:55 AI video realism concerns 14:54 AI Search overhaul 17:11 Trust provenance and web economics 22:18 YouTube summaries and Ask YouTube 24:29 Gemini Spark personal agent 26:52 Ultra plan lock in and privacy tradeoffs 35:05 Convenience vs Societal Costs 35:46 Pandoras Box and Downstream Risks 36:37 Humans Leaving the Value Chain 39:09 Agency vs Algorithmic Feeds 41:16 Gemini Daily Briefing 42:08 Antigravity and Developer Orchestration 43:09 Information Agents and Signal vs Noise 50:44 Universal Cart and Commerce Protocol 52:52 Will Merchants Opt In 55:14 Google Product Longevity and Compute Limits 57:09 SynthID Watermarking and Truth Concerns 01:00:39 Android XR Glasses and Attention Economy 01:05:45 Who Actually Needs XR 01:07:44 Wrap Up and Listener Feedback This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit aiunprompted.substack.com [https://aiunprompted.substack.com?utm_medium=podcast&utm_campaign=CTA_1]

024 - Exploring AI's Cutting Edge: From Cognitive Debt to Conversational Agents

Ryan, Kevin, and Travis discuss how impossible it is to keep up with AI’s pace and use recent OpenClaw updates to illustrate what’s happening at the “tip of the spear.” They recap an OpenClaw community-driven plugin architecture overhaul that caused short-term instability but created a smaller stable core with extensible plugins. They cover new voice interaction options, including Discord real-time voice features (buffers, barge-in detection, echo control), bringing agents into Google Meet via Twilio dial-in, and broader implications of voice and multimodal “thinking machines” interaction models. The hosts explore cognitive debt/coherence challenges as AI builds faster than humans can comprehend, and highlight OpenClaw’s security hardening (1,300 advisories processed) plus major memory upgrades that create structured person cards and a wiki-like knowledge base, raising governance and compartmentalization concerns for enterprises. They also note improved commitment tracking, self-modifying/self-building capabilities, and auto-generated skills. 00:00 Welcome Back Setup 00:55 Why Youre Behind 01:13 OpenClaw Overview 02:22 Community Moves Fast 03:04 Plugin Overhaul Fallout 04:25 Stable Core Plugins 05:39 Pick Your AI Strategy 08:48 Cognitive Debt Explained 10:26 Daily Reps Mindset 12:07 Voice Comes to OpenClaw 13:04 Discord Voice and Meet 18:21 Metacognition Modalities 21:30 Do You Need Code 26:09 Voice vs Text Context 29:35 Thinking Machines Tease 29:39 Interaction Models Demo 30:53 Voice Latency Tradeoffs 33:27 Conversation Cues Vision 36:39 OpenClaw Security Hardening 38:31 Memory And Knowledge Base 40:04 Enterprise Governance Dilemma 45:21 Corporate Brain Example 48:59 Auto Commitments Heartbeat 51:45 Stability Updates Skills 52:44 Wrap Up And Thanks This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit aiunprompted.substack.com [https://aiunprompted.substack.com?utm_medium=podcast&utm_campaign=CTA_1]

023 - AI in Hollywood, Mining Innovations, Memory Models, and Podcast Automation

Ryan, Kevin, and Travis return for a news roundup and announce Kevin’s resignation from Microsoft after getting funded for a startup applying “human-led, agent-operated” AI to small, underserved mining operations (Kai Radian). Listener feedback highlights Tauric Research, a multi-agent financial trading framework with analyst, research, trader, and portfolio-manager roles, raising concerns about overconfidence and automation. They discuss Anthropic previewing Claude’s “dreaming” memory consolidation and frame “harnesses” as orchestration/scaffolding that improves signal, manages context, and mixes deterministic workflows with LLM judgment, including models checking each other. They cover Anthropic using SpaceX’s Colossus I compute in Memphis to boost Claude Code limits, Mozilla using Claude Mythos to find 271 bugs in a month, reports that 39% of new podcasts are AI-generated “slop,” and a four-year Actors Guild deal adding AI protections for voice/likeness and writers’ rights. 00:00 Welcome Back Updates 00:52 Kev Leaves Microsoft 01:37 Mining Startup Vision 04:18 Mine Talk Banter 04:48 Listener Feedback Trading Agents 09:11 Claude Dreaming Memory 12:58 What Is A Harness 17:57 Determinism Vs Judgment 20:58 Anthropic SpaceX Compute 26:54 Mythos Finds Firefox Bugs 30:22 AI Security Attack Vectors 31:21 AI Podcast Flooding 34:33 AI Slop in Communities 38:25 Brain Atrophy and AI 39:51 Actors Guild AI Protections 44:31 Oscars and Human Eligibility 49:07 Democratizing Creativity 52:46 Human Storytelling Matters 54:08 Wrap Up and Feedback This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit aiunprompted.substack.com [https://aiunprompted.substack.com?utm_medium=podcast&utm_campaign=CTA_1]

022 - Omar Shahine, Microsoft CVP of OpenClaw + Microsoft 365

ai.u crew talk to Omar Shahine, Microsoft Corporate Vice President of OpenClaw and Microsoft 365, about his tech origins and career. Omar recalls getting an Apple IIe in third grade, automating tasks with tools like FileMaker Pro, and arriving at Microsoft via a 1995 blog and a 1999 tester internship after being rejected from medical school. He highlights formative work in the Mac business unit during Apple’s revival and scaling OneDrive to hundreds of millions of users. Omar describes leadership lessons centered on customer focus and empowering teams, then explains how using Claude Code and building an OpenClaw assistant named “Lobster” (e.g., proactive meeting texts, family coordination, automation tools) led to a viral blog post, a presentation in a Satya-hosted forum, and a role transition to build this capability for Microsoft 365, emphasizing trust, feedback, and personalized, agent-driven productivity.00:00 Welcome and Guest Intro01:12 Early Tech Spark Apple II03:08 From Pre Med to Microsoft05:15 Thrown in the Deep End06:47 Pinch Me Career Moments09:08 Leadership Lessons at Scale12:27 Why OpenClaw Matters14:11 Building Lobster Assistant19:54 Going Viral Inside Microsoft22:40 Joining the OpenClaw Team24:45 The Story Behind the Hype25:43 Why Software Feels Hard27:21 Agents Over Buttons29:16 Personalized Agent Loops31:17 Trust and Accountability34:56 Customer Pull and DIY Agents37:30 Agents Talking Together40:08 Tooling Everyday Life41:55 Agent Friendly Internet46:24 Advice for Newcomers48:11 CoWorker Demo and Wrap This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit aiunprompted.substack.com [https://aiunprompted.substack.com?utm_medium=podcast&utm_campaign=CTA_1]

021 - Anthropic's Mythos: The AI Model That Changes Everything

ai.u crew discuss the announcement of Claude Mythos preview, a new “frontier model” not released publicly but deployed through a cybersecurity coalition called Project Glasswing. They describe Glasswing’s 12 founding partners (AWS, Apple, Broadcom, Cisco, CrowdStrike, Google, JP Morgan Chase, Linux Foundation, Microsoft, Nvidia, Palo Alto Networks, and Anthropic) and report that Mythos found thousands of zero-day vulnerabilities across major operating systems and browsers, including a 27-year-old OpenBSD bug, a 16-year-old FFmpeg issue, and autonomously chained Linux kernel vulnerabilities to escalate privileges. They note benchmark gains (e.g., 66.6% to 83.1% on a security exploit test and 53% to 64% on “Humanity’s Last Exam”), partner feedback that exploit windows are now minutes, concerns about abstraction and cognitive debt, and Anthropic’s $100M credits plus $4M open-source donations, with ongoing U.S. government discussions and future safeguards before broader capability release. 00:00 Welcome and Setup 00:57 Mythos and Glasswing 02:31 Coalition Partners 03:41 Zero Day Discoveries 05:09 Chaining Exploits Explained 06:14 Benchmarks and Scores 08:16 Not Just Cybersecurity 11:27 Oppenheimer Moment 15:26 Partner Results 19:18 Governance and National Security 21:16 Digital World Risks 22:20 Digital Fragility Fears 22:54 AI Distance From Work 24:15 Cognitive Debt Explained 25:48 Agents Everywhere Future 27:29 Self Healing Systems Drift 29:36 Alignment Goals And Means 33:04 Autonomous AI Companies 35:10 AI For AI Economics 38:39 Governance Tool Access Risks 40:54 Mythos Security Outlook 42:58 Blackwell Training Breakthrough 44:02 Costs Credits And Zero Days 45:53 Model Therapy And Dreaming 46:58 Safeguards Wrap Up This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit aiunprompted.substack.com [https://aiunprompted.substack.com?utm_medium=podcast&utm_campaign=CTA_1]