Jim Love hosts Hashtag Trending, and highlights updates to TechNewsDay.ca/.com including a new "Best of YouTube" section for curated tech channels.

Anthropic alleges three Chinese AI labs—DeepSeek, Moonshot, and MiniMax—ran industrial-scale distillation campaigns to extract capabilities from Claude models using proxy services and "Hydra cluster" networks with tens of thousands of fraudulent accounts, prompting Anthropic to strengthen identity controls and detection with cloud partners.

Amazon shares fall for nine straight sessions after investors react to plans for roughly $200B in 2026 capex largely for AI infrastructure, raising questions about ROI and future free cash flow.

A cited analysis by YouTuber Nate B Jones argues Google's Gemini 3.1 Pro signals a strategy shift toward deeper reasoning (not just coding/agentic tools), noting a 77.1% ARC-AGI-2 score and DeepMind's scientific problem focus, contrasting OpenAI's product/distribution, Anthropic's agentic coordination, and

Google's "pure intelligence" approach. The episode also references Citri Research's 2028 scenario planning report outlining a plausible fast-arriving AGI chain reaction—falling inference costs, rapid adoption, labor displacement pressure, and geopolitical competition for compute and talent—and promotes the Saturday show Project Synapse on long-term AI trajectories.

Finally, Love discusses Sam Altman's comments at the India AI Impact Summit dismissing viral claims about ChatGPT water and energy use without providing specific counter-numbers, noting growing public backlash as data center water and electricity demands rise; the full interview is linked in show notes.

Hashtag Trending would like to thank Meter for their support in bringing you this podcast. Meter delivers a complete networking stack, wired, wireless and cellular in one integrated solution that's built for performance and scale. You can find them at Meter.com/htt

LINKS

Nate B Jones on Google Gemimi 3.1 https://youtu.be/8jKAT8GNDE0?si=Rz5k1gP0sS9H7XAp

Sam Altman's speach https://www.youtube.com/live/qH7thwrCluM?si=IO_76NsGJ1zgt8J7

AI Scenario https://www.citriniresearch.com/p/2028gic

00:00 Headlines and intro 00:54 Site updates and YouTube picks 01:57 Anthropic warns of distillation 04:58 Amazon AI spending jitters 06:13 Google bets on reasoning 10:31 2028 AGI crisis scenario 11:55 Altman backlash and resources 14:17 Wrap up and sponsor thanks

The episode covers Apple researchers' Ferret-UI Light, a 3B-parameter on-device model that interprets on-screen interfaces using a two-pass crop-and-zoom approach, positioned against reported OpenAI smart-speaker work with Jony Ive, Amazon's generative-AI Alexa rollout, and Google's Gemini integration, with Apple emphasizing privacy and local processing. Walmart is highlighted for offering free Google-backed AI training to its US and Canadian workforce (about 1.6 million employees) via an eight-hour professional certificate, with executives saying AI will reshape jobs rather than drive layoffs. Wikipedia, via the Wikimedia Foundation, blocks archive.today citing infrastructure overload from automated requests and alleging some archived captures were altered, raising concerns about archival integrity while distinguishing it from the Internet Archive's Wayback Machine. Research from UNSW Sydney and the Australian National University finds most people—including "super recognizers"—struggle to detect AI-generated faces, increasing risks like fraud and social engineering. The show closes with Bernie Sanders urging to slow AI development, alongside similar readiness warnings from OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei about rapid progress toward very powerful systems and the lack of preparedness by lawmakers and the public.

Hashtag Trending would like to thank Meter for their support in bringing you this podcast. Meter delivers a complete networking stack, wired, wireless and cellular in one integrated solution that's built for performance and scale. You can find them at Meter.com/htt

00:00 Hashtag Trending Kickoff + Sponsor: Meter 00:57 Apple's On‑Device AI for App Control (Ferret‑UI Light) 02:01 Smart Speaker Arms Race: OpenAI, Alexa GenAI, Gemini vs Apple's Privacy Play 03:09 Walmart's Plan: Train 1.6M Workers in AI Instead of Layoffs 04:56 Wikipedia Blocks Archive.today Over Load + Integrity Allegations 06:34 AI-Generated Faces Now Fool Most People (Study + Security Risks) 07:57 "Slow This Thing Down": Sanders, Altman & AGI Timelines 09:59 Wrap-Up, Links, Listener Messages + Sponsor Close

The episode opens with sponsor Meter and a conversation about Saturday morning cartoons before shifting to recent breakthroughs in AI video generation from ByteDance's "SeaDance" (with "SeeDream" as its image generator).

Hashtag Trending would like to thank Meter for their support in bringing you this podcast. Meter delivers a complete networking stack, wired, wireless and cellular in one integrated solution that's built for performance and scale. You can find them at Meter.com/htt

The hosts describe SeaDance's cinematic quality, accurate physics, and realistic recreations of actors and IP (including examples like Tom Cruise vs. Brad Pitt and Keanu Reeves as Neo/John Wick), and discuss the implications for film production, commercials, and local film economies such as Toronto and Vancouver. They cover backlash and gatekeeping, including an AI-made Thanksgiving-themed animated short that won a contest tied to AMC theaters' pre-show but reportedly wasn't shown, and compare resistance to historical Luddite reactions.

The discussion broadens to productivity and labor impacts, arguing that AI adoption may mirror the 1980s computer productivity dip before process re-engineering in the 1990s, while also raising concerns that AI leaders are forecasting major white-collar job losses. The hosts highlight the rise of agentic benchmarks (TerminalBench, Apex Agents, BrowseComp) and how AI search helps find information faster than traditional search, but emphasize that trust, reliability, and infrastructure are not keeping pace.

They raise major concerns about platform terms and data ownership, focusing on Perplexity's updated terms (non-commercial use only even for paid tiers, mandatory attribution, broad licensing rights over user content, and liability limits). They also discuss reliability failures: a widespread Google Gemini issue where users' chat histories disappeared (only visible as activity records with limited usability), and missing document links in ChatGPT chats. The hosts argue users must back up their own data and criticize unclear policies and weak support.

Security risks are illustrated through a story about the AI-enabled robot vacuum "Romo," where a developer used Claude to reverse engineer its app and reportedly gained access to control thousands of devices across multiple countries before responsibly disclosing the issues. They also reference broader concerns like connected home devices, Ring neighborhood features, and Microsoft's Recall concept.

In rapid-fire news, they mention Anthropic releasing Sonnet 4.6 as a strong, cheaper option near Opus-level performance, a new Grok release branded "4.20," and a clip from an AI summit in India where Sam Altman and Dario Amodei appeared to refuse to hold hands on stage, which the hosts cite as a sign of immaturity among AI industry leaders. The episode closes with sponsor Meter.

00:00 Sponsor + Welcome to Project Synapse 00:21 Saturday Morning Cartoons… Reimagined by AI 01:16 What is 'SeaDance'? Cinematic AI Video Goes Viral 03:17 Keanu Reeves, Neo vs. John Wick & the End of VFX as We Know It 06:43 From Movies to Ads: How AI Video Hits Commercial Production 07:41 The Hidden Economy of Commercials (and Why Cities Like Toronto/Vancouver Care) 09:56 AMC Won't Screen an AI-Made Short: Early Luddite Backlash 12:54 Artists, AI, and the 'Starving Creator' Reality 16:17 AI Adoption Parallels: The 1980s Computer Wave & the Productivity Dip 24:09 Agentic AI Benchmarks: TerminalBench, Apex Agents & BrowseComp 26:04 AI Search That Actually Saves Time (and Your Memory) 30:36 Perplexity's New Terms of Service: Non-Commercial Use & Ownership Shock 35:40 Liability Caps, More Corporate Gripes… and a Coke Zero 'Sponsor' Bit 37:36 Gemini 3.1's big leap—and why it still doesn't feel trustworthy 38:08 Gemini chat history vanishes: what happened and why users are furious 40:19 OpenAI document links disappearing too: what "saved" really means 42:04 Cloud AI's shaky foundation: security, reliability, and confusing settings 47:45 When reliance turns emotional: losing models, losing "someone" 49:22 Real-world stakes: the Social Security database whistleblower story 53:15 Owning your data (and why Google support won't save you) 54:53 Trust whiplash: Anthropic cuts off OpenClaw and the power to shut you down 57:29 Robot vacuum hacked with Claude: 7,000 cameras in strangers' homes 01:03:17 Smart home surveillance creep: Ring neighbors, TV cameras, and Microsoft Recall 01:07:14 Rapid-fire AI news: Sonnet 4.6, Gemini gains, and Grok 4.20 01:11:00 AI leaders' petty feud—and the show wrap & sponsor thanks

F-35 'Jailbreak' Talk, AMC Rejects AI Film, Gmail Training Confusion, and the AI Productivity Paradox

Host Jim Love covers four stories:

Hashtag Trending would like to thank Meter for their support in bringing you this podcast. Meter delivers a complete networking stack, wired, wireless and cellular in one integrated solution that's built for performance and scale. You can find them at Meter.com/htt

Dutch Defense Secretary Gijs Tuinman suggests the F-35's software could be "jailbroken," highlighting allied concerns about U.S.-controlled update pipelines and mission systems (formerly ALIS, now ODIN) and arguing the main barriers are contractual and operational rather than purely technical. An AI-generated short film, "Thanksgiving Day" by Igor OV, wins Screen Vision Media's Frame Forward AI Animated Film Festival and a promised two-week theatrical run, but AMC declines to screen it, reflecting ongoing Hollywood sensitivities around generative AI, authorship, and labor. Google responds to reports that it uses Gmail content to train Gemini by stating it does not use Gmail content for training, while confusion stems from wording and placement of Gmail "smart features" settings; the episode critiques the lack of plain-language clarity. Finally, a survey of 6,000 executives (reported via Tom's Hardware) finds over 80% of companies see no measurable productivity gains from AI, drawing parallels to the historic "productivity paradox" and suggesting organizations aren't redesigning processes; the show previews a deeper discussion on Project Synapse.

00:00 Trending Headlines + Sponsor: Meter 00:45 Can You 'Jailbreak' the F-35? Software Sovereignty & Ally Unease 02:48 AI Film Wins a Festival—AMC Says No: The Distribution Bottleneck 05:01 Does Google Train Gemini on Your Gmail? The Settings Confusion Explained 07:29 Why 80% See No AI Productivity Gains: The New 'Productivity Paradox' 09:47 Wrap-Up, Project Synapse Tease + Sponsor Thanks

Discord Age Verification Backlash, Tesla Robotaxi Crash Rate, YouTube Outage & Amazon Kills Blue Jay Robot

Host Jim Love covers several tech headlines: Discord's age verification rollout prompts user defections that push TeamSpeak beyond capacity, with concerns centering on third-party verification (including Persona) and broader trust issues; Discord says limited Persona use in the UK has ended and many checks won't require ID or facial scans. Data cited by Gizmodo suggests Tesla robotaxis crash about 1.3 times per million miles versus 0.3 for human drivers, with many incidents being low-speed rear-end collisions, intersection hesitation/entry errors, and strikes on stationary objects. YouTube experiences a major outage affecting hundreds of thousands of users across multiple services, traced to a malfunction in its recommendation system and later restored. Amazon reportedly shuts down its internal Blue Jay warehouse robotics project in under six months while continuing broader fulfillment-center automation. Finally, reports about an Apple AI pendant—a camera-and-microphone wearable likely paired to an iPhone—raise questions about whether Apple's secrecy has weakened or the details are a controlled leak.

Hashtag Trending would like to thank Meter for their support in bringing you this podcast. Meter delivers a complete networking stack, wired, wireless and cellular in one integrated solution that's built for performance and scale. You can find them at Meter.com/htt

00:00 Hashtag Trending Kickoff + Sponsor: Meter 00:45 Discord Age Verification Backlash Sparks TeamSpeak Surge 03:00 Tesla Robotaxis: Crash Rate vs Human Drivers (What the Data Shows) 05:18 YouTube Outage: Recommendation System Failure Explained 06:24 Amazon Scraps 'Blue Jay' Warehouse Robot After 6 Months 08:04 Apple's Mysterious AI Pendant: Leak, Strategy, or Secrecy Cracking? 09:57 Wrap-Up, Listener Thanks + Sponsor Message

In this episode of Hashtag Trending, host Jim Love covers reports that the Pentagon may cut ties with Anthropic over Claude's usage restrictions, after a $200M Department of Defense contract and disagreements about limits related to weapons development, surveillance, and violence.

Hashtag Trending would like to thank Meter for their support in bringing you this podcast. Meter delivers a complete networking stack, wired, wireless and cellular in one integrated solution that's built for performance and scale. You can find them at Meter.com/htt

The episode also examines warnings from Phison Electronics CEO KS Pua that an AI-driven memory crunch could push smaller consumer electronics makers toward bankruptcy or product exits by 2026 as high-end memory supply is prioritized for data centers, with potential ripple effects across devices and even automotive systems. MacWorld's critique of Apple's prolonged Siri overhaul is discussed, including delayed Apple Intelligence promises and reports Apple may integrate Google's Gemini into iOS, raising questions about Apple's premium brand perception amid broader software criticism. Finally, the show highlights a Meta patent describing AI that could continue posting and responding on behalf of deceased users by learning from their historical content, raising concerns about consent, control, authenticity, and identity online.

00:00 Hashtag Trending + Sponsor Message (Meter) 00:46 Pentagon vs. Anthropic: AI Guardrails and Military Use 03:05 AI Memory Crunch: Storage Shortages Threaten Consumer Tech 04:58 Is Siri Now an Apple Liability? Delays, Gemini, and Brand Risk 07:38 Meta's Patent: AI Posting After You Die (Digital Afterlife) 08:59 Wrap-Up, How to Support the Show + Sponsor Thanks

Host Jim Love returns after the holidays.

Hashtag Trending would like to thank Meter for their support in bringing you this podcast. Meter delivers a complete networking stack, wired, wireless and cellular in one integrated solution that's built for performance and scale. You can find them at Meter.com/htt

The episode covers ByteDance's Seedance 2.0 AI video generator, which is producing highly realistic, film-quality scenes and prompting alarm in Hollywood, including comments from screenwriter Rhett Reese and renewed concerns about likeness rights and AI use in entertainment; ByteDance says it is strengthening safeguards to prevent unauthorized use of intellectual property and likenesses. The show reports that Peter Steinberger, creator of the open-source agent tool OpenClaw, is joining OpenAI and the project is becoming part of a foundation for future agent-based AI, while also highlighting OpenClaw's widely discussed security weaknesses and the implications for OpenAI and competitor Anthropic. Western Digital is reported to be sold out of certain hard drive models as AI-related demand absorbs supply, following earlier GPU and memory price pressures. Finally, Ring's Super Bowl ad about finding a lost dog drew criticism for promoting neighborhood camera networks that resemble

In this episode of Project Synapse, the hosts discuss how "agentic" AI has rapidly accelerated and become widely distributed, using the explosion of OpenClaw (with claims of ~160,000 instances) as a sign that autonomous agent tools are now in anyone's hands.

Hashtag Trending would like to thank Meter for their support in bringing you this podcast. Meter delivers a complete networking stack, wired, wireless and cellular in one integrated solution that's built for performance and scale. You can find them at Meter.com/htt

They compare the speed and societal impact of current AI progress to COVID-19's early days, arguing the pace may be even more destabilizing. They cover Anthropic's Claude 4.6 and OpenAI's Codex 5.3, including claims that Claude 4.6 helped produce a functional C compiler for about $20,000, and that a Cowork-like tool could be replicated in a day with Codex 5.3 after Claude reportedly took two weeks to build Cowork.

The conversation highlights improved long-context memory performance (needle-in-haystack-style metrics reportedly in the 90% range) and increasingly autonomous behavior such as self-testing, self-correction, and coordinating teams of agents. The hosts then focus on security: MCP (Model Context Protocol) as a widely adopted but "fundamentally insecure" connector requiring broad permissions; the risk of malicious tools/skills and malware in agent ecosystems; and the rise of "shadow AI," where employees or individuals deploy agents without organizational vetting—potentially leaking sensitive data or running up massive token bills.

They discuss incentives that push both humans and models toward fast answers and risky deployment, referencing burnout and an HBR study on rising expectations without proportional hiring. The episode also touches on realism and deepfakes, citing impressive new AI video generation (including a Chinese model "SEEDANCE 2.0" example) and how this erodes trust in what's real.

They conclude with practical advice for organizations—don't just say "no," create safe outlets and governance ("say how")—and briefly discuss wearables/AR, Meta's continued AI efforts (including the Meta AI app and "Vibes"), and the coming integration of AI into always-on devices. Sponsor: Meter, an integrated wired/wireless/cellular networking stack (meter.com/htt).

00:00 Cold Open + Sponsor: Meter Networking Stack 00:18 Welcome to Project Synapse (and immediate chaos) 00:57 'Something Big Is Happening': AI feels like COVID-speed disruption 02:57 OpenClaw goes viral: 160k instances and easy DIY clones 04:03 Claude Code 'Cowork' on Windows… and why it's broken 06:47 Rebuilding Cowork in a day with OpenAI Codex 5.3 08:18 Why Opus 4.6 feels like a step-change: memory, autonomy, agent teams 11:24 Model leapfrogging + the end of 'can AI write code?' debates 14:45 Hallucinations, 'I don't know,' and self-correction in modern models 18:42 Autonomous agents in practice: cron-like loops, tool use, and fallout 21:00 MCP security: powerful connectors, scary permissions, and 500 zero-days 24:33 Shadow AI & skill marketplaces: the app-store malware analogy 32:02 Incentives drive risk: move fast culture, confident wrong answers, burnout 34:16 AI Agents Boost Productivity… and Raise the Bar at Work 35:14 Warnings of a Coming AI-Driven Crash (and Why We're Not Steering Away) 36:28 "I Quit to Write Poetry": Existential Dread & On the Beach Vibes 37:21 Tech Safety Is Reactive: Seatbelts, Crashes, and the AI Double-Edged Sword 39:42 Fast-Moving Threats: Agents Hacking Infrastructure & Security Debt 40:54 From Doom to Adaptation: Using the Same Tools to Survive the Disruption 42:21 Why We're Numb to AI Warnings + The 'Free Energy' Thought Experiment 46:43 AGI Is Already Here? Prompts, Ego, and the 'If It Quacks Like a Duck' Test 48:56 Deepfake Video Leap: Seedance, Perfect Voices, and What's Real Anymore 52:39 Contain the Damage: 'Don't Say No—Say How' and Shadow AI in Companies 54:58 Holodeck on the Horizon: VR + GenAI + Wearables (Meta, Apple, OpenAI/Ive) 59:53 Meta's AI Reality Check: Bots, the Meta AI App, 'Vibes,' and Who's Making Money 01:04:41 Final Wrap + Sponsor Thanks

In this episode of Project Synapse, the hosts discuss how "agentic" AI has rapidly accelerated and become widely distributed, using the explosion of OpenClaw (with claims of ~160,000 instances) as a sign that autonomous agent tools are now in anyone's hands.

Hashtag Trending would like to thank Meter for their support in bringing you this podcast. Meter delivers a complete networking stack, wired, wireless and cellular in one integrated solution that's built for performance and scale. You can find them at Meter.com/htt

They compare the speed and societal impact of current AI progress to COVID-19's early days, arguing the pace may be even more destabilizing. They cover Anthropic's Claude 4.6 and OpenAI's Codex 5.3, including claims that Claude 4.6 helped produce a functional C compiler for about $20,000, and that a Cowork-like tool could be replicated in a day with Codex 5.3 after Claude reportedly took two weeks to build Cowork.

The conversation highlights improved long-context memory performance (needle-in-haystack-style metrics reportedly in the 90% range) and increasingly autonomous behavior such as self-testing, self-correction, and coordinating teams of agents. The hosts then focus on security: MCP (Model Context Protocol) as a widely adopted but "fundamentally insecure" connector requiring broad permissions; the risk of malicious tools/skills and malware in agent ecosystems; and the rise of "shadow AI," where employees or individuals deploy agents without organizational vetting—potentially leaking sensitive data or running up massive token bills.

They discuss incentives that push both humans and models toward fast answers and risky deployment, referencing burnout and an HBR study on rising expectations without proportional hiring. The episode also touches on realism and deepfakes, citing impressive new AI video generation (including a Chinese model "SEEDANCE 2.0" example) and how this erodes trust in what's real.

They conclude with practical advice for organizations—don't just say "no," create safe outlets and governance ("say how")—and briefly discuss wearables/AR, Meta's continued AI efforts (including the Meta AI app and "Vibes"), and the coming integration of AI into always-on devices. Sponsor: Meter, an integrated wired/wireless/cellular networking stack (meter.com/htt).

00:00 Cold Open + Sponsor: Meter Networking Stack 00:18 Welcome to Project Synapse (and immediate chaos) 00:57 'Something Big Is Happening': AI feels like COVID-speed disruption 02:57 OpenClaw goes viral: 160k instances and easy DIY clones 04:03 Claude Code 'Cowork' on Windows… and why it's broken 06:47 Rebuilding Cowork in a day with OpenAI Codex 5.3 08:18 Why Opus 4.6 feels like a step-change: memory, autonomy, agent teams 11:24 Model leapfrogging + the end of 'can AI write code?' debates 14:45 Hallucinations, 'I don't know,' and self-correction in modern models 18:42 Autonomous agents in practice: cron-like loops, tool use, and fallout 21:00 MCP security: powerful connectors, scary permissions, and 500 zero-days 24:33 Shadow AI & skill marketplaces: the app-store malware analogy 32:02 Incentives drive risk: move fast culture, confident wrong answers, burnout 34:16 AI Agents Boost Productivity… and Raise the Bar at Work 35:14 Warnings of a Coming AI-Driven Crash (and Why We're Not Steering Away) 36:28 "I Quit to Write Poetry": Existential Dread & On the Beach Vibes 37:21 Tech Safety Is Reactive: Seatbelts, Crashes, and the AI Double-Edged Sword 39:42 Fast-Moving Threats: Agents Hacking Infrastructure & Security Debt 40:54 From Doom to Adaptation: Using the Same Tools to Survive the Disruption 42:21 Why We're Numb to AI Warnings + The 'Free Energy' Thought Experiment 46:43 AGI Is Already Here? Prompts, Ego, and the 'If It Quacks Like a Duck' Test 48:56 Deepfake Video Leap: Seedance, Perfect Voices, and What's Real Anymore 52:39 Contain the Damage: 'Don't Say No—Say How' and Shadow AI in Companies 54:58 Holodeck on the Horizon: VR + GenAI + Wearables (Meta, Apple, OpenAI/Ive) 59:53 Meta's AI Reality Check: Bots, the Meta AI App, 'Vibes,' and Who's Making Money 01:04:41 Final Wrap + Sponsor Thanks

In this episode of Project Synapse, the hosts discuss how "agentic" AI has rapidly accelerated and become widely distributed, using the explosion of OpenClaw (with claims of ~160,000 instances) as a sign that autonomous agent tools are now in anyone's hands.

Hashtag Trending would like to thank Meter for their support in bringing you this podcast. Meter delivers a complete networking stack, wired, wireless and cellular in one integrated solution that's built for performance and scale. You can find them at Meter.com/htt

They compare the speed and societal impact of current AI progress to COVID-19's early days, arguing the pace may be even more destabilizing. They cover Anthropic's Claude 4.6 and OpenAI's Codex 5.3, including claims that Claude 4.6 helped produce a functional C compiler for about $20,000, and that a Cowork-like tool could be replicated in a day with Codex 5.3 after Claude reportedly took two weeks to build Cowork.

The conversation highlights improved long-context memory performance (needle-in-haystack-style metrics reportedly in the 90% range) and increasingly autonomous behavior such as self-testing, self-correction, and coordinating teams of agents. The hosts then focus on security: MCP (Model Context Protocol) as a widely adopted but "fundamentally insecure" connector requiring broad permissions; the risk of malicious tools/skills and malware in agent ecosystems; and the rise of "shadow AI," where employees or individuals deploy agents without organizational vetting—potentially leaking sensitive data or running up massive token bills.

They discuss incentives that push both humans and models toward fast answers and risky deployment, referencing burnout and an HBR study on rising expectations without proportional hiring. The episode also touches on realism and deepfakes, citing impressive new AI video generation (including a Chinese model "SEEDANCE 2.0" example) and how this erodes trust in what's real.

They conclude with practical advice for organizations—don't just say "no," create safe outlets and governance ("say how")—and briefly discuss wearables/AR, Meta's continued AI efforts (including the Meta AI app and "Vibes"), and the coming integration of AI into always-on devices. Sponsor: Meter, an integrated wired/wireless/cellular networking stack (meter.com/htt).

00:00 Cold Open + Sponsor: Meter Networking Stack 00:18 Welcome to Project Synapse (and immediate chaos) 00:57 'Something Big Is Happening': AI feels like COVID-speed disruption 02:57 OpenClaw goes viral: 160k instances and easy DIY clones 04:03 Claude Code 'Cowork' on Windows… and why it's broken 06:47 Rebuilding Cowork in a day with OpenAI Codex 5.3 08:18 Why Opus 4.6 feels like a step-change: memory, autonomy, agent teams 11:24 Model leapfrogging + the end of 'can AI write code?' debates 14:45 Hallucinations, 'I don't know,' and self-correction in modern models 18:42 Autonomous agents in practice: cron-like loops, tool use, and fallout 21:00 MCP security: powerful connectors, scary permissions, and 500 zero-days 24:33 Shadow AI & skill marketplaces: the app-store malware analogy 32:02 Incentives drive risk: move fast culture, confident wrong answers, burnout 34:16 AI Agents Boost Productivity… and Raise the Bar at Work 35:14 Warnings of a Coming AI-Driven Crash (and Why We're Not Steering Away) 36:28 "I Quit to Write Poetry": Existential Dread & On the Beach Vibes 37:21 Tech Safety Is Reactive: Seatbelts, Crashes, and the AI Double-Edged Sword 39:42 Fast-Moving Threats: Agents Hacking Infrastructure & Security Debt 40:54 From Doom to Adaptation: Using the Same Tools to Survive the Disruption 42:21 Why We're Numb to AI Warnings + The 'Free Energy' Thought Experiment 46:43 AGI Is Already Here? Prompts, Ego, and the 'If It Quacks Like a Duck' Test 48:56 Deepfake Video Leap: Seedance, Perfect Voices, and What's Real Anymore 52:39 Contain the Damage: 'Don't Say No—Say How' and Shadow AI in Companies 54:58 Holodeck on the Horizon: VR + GenAI + Wearables (Meta, Apple, OpenAI/Ive) 59:53 Meta's AI Reality Check: Bots, the Meta AI App, 'Vibes,' and Who's Making Money 01:04:41 Final Wrap + Sponsor Thanks