AI News Today | Julian Goldie Podcast

Fable 5 Got Pulled in 72 Hours—Here’s the Better, Cheaper Replacement (Fusion 5)After Anthropic launched Fable 5 on June 9, 2026 and it was pulled worldwide three days later (along with MiFaS 5), the script argues you can still reach “Fable-level” performance using a new OpenRouter model called Fusion 5 for about half the price. Fusion uses a panel of 3–5 top models answering in parallel (examples include Opus, Gemini, Claude, Groq Build, and Gemini 3.5 Flash) with live web search, then a sixth “judge” model merges the results into one final answer. Side-by-side demos (glacial valley, voxel Minecraft world, isometric RPG, city block simulator, Hogwarts) claim Fusion is generally smoother, less buggy, and more detailed than Opus 4.8 alone. OpenRouter testing on 100 deep research tasks reportedly scored Fusion above Fable 5 and beat GPT 5.5 and Opus 4.8, though Fusion is slower and uses more tokens.00:00 Fable 5 Vanishes00:48 Fusion 5 Alternative01:24 What Happened Timeline01:43 Fusion Panel Explained02:25 Glacial Valley Demo03:41 Minecraft Voxel Test05:01 Isometric RPG Showdown06:10 City Block And Hogwarts08:32 Research Benchmarks Proof09:22 System Integration Pitch10:22 Best Use Cases11:31 Tradeoffs Speed Cost12:09 New Way To Build14:16 Final Steps And Wrap

Build Your Own AI Agent Operating System: Scheduling, QA Judges, Obsidian Memory & Team Workflows (Hermes, Claude, Paperclip)The video answers common questions about building an “agent operating system” where teams of AI agents collaborate using tools like Hermes, Claude, Obsidian, and Kanban boards. It explains how to give an agent long-term ownership of tasks (e.g., ongoing SEO) via scheduled cron jobs managed in one place, and recommends Claude over Hermes for heavier coding work while noting model/API choice affects performance. For quality control and reliable handoffs, it demonstrates adding a dedicated “judge” agent that critiques, scores output, and forces iteration until standards are met. It compares workflows across OpenClaw, Hermes (native Kanban for teams), and Paperclip (multi-tool, company-style orchestration), and argues that adding memory/context through Obsidian improves coordination and creates an auto-updated second brain. It also covers Windows options for an “infinite context engine,” and showcases a custom voice agent example, then directs viewers to the AI Profit Boardroom for the full setup and training.00:00 Agent OS Overview01:04 Long Running Tasks02:56 Coding Model Choices04:18 Quality Control Judge07:00 Benchmarking Free vs Paid07:48 Managing Agent Teams10:11 Obsidian Memory Vault11:19 Infinite Context Windows13:00 Voice Agent Showcase14:09 Community Setup Wrap Up

GLM 5.2 vs Claude Opus 4.8: I Ran 5 Build Tests (China Won 4) + The CatchThe video compares China’s new GLM 5.2 (by ZAI, released June 13, 2026) against Anthropic’s Claude Opus 4.8 (released May 28) using the same five build prompts. GLM 5.2 produces better results in four tests—a running game, an Apple-style landing page, a liquid-sloshing animation, and a neon arcade game—while Claude wins the fifth test with a cleaner moving solar system map. The script highlights pricing differences (Claude’s per-token API costs vs GLM’s flat monthly coding plan) and argues the best approach is to run both: use GLM for cheap “building” tasks and Claude for deeper thinking inside a unified “agent operating system.” It notes key catches: no independent benchmarks yet, and GLM’s API/chatbot/open weights are coming next week.00:00 GLM vs Claude Setup01:28 Build Test 1 Runner Game02:01 Build Test 2 Landing Page02:26 Build Tests 3 and 4 Animations03:48 Test 5 and Price Shock04:54 Stop Picking Sides05:26 Agent OS Workflow07:01 The Catch and Missing Proof08:14 Agent OS Offer Break09:13 Daily Business Use Example11:13 Why Agents Beat Model Chasing12:28 Final Verdict Run Both

Build Your Own AI Operating System with Hermes Agent (Free Models, Memory, Video Pipelines & More)The video explains how to build an “AI operating system” using Hermes Agent to manage multiple AI workers (writing, editing, judging) in one place, swap in new models as they appear, and keep workflows like video creation, SEO, games, music, and lead gen organized and reusable. It contrasts this approach with Hermes Desktop (limited to Hermes and less visual) and n8n (more technical, messy, and break-prone), and shows examples like building games and fully edited videos using the Hyperframes skill plus optional HeyGen. It covers cost and token control via coding plans (e.g., Kimi, GLM 5.2), free OpenRouter APIs, OAuth logins, Headroom token reduction, and free models via News Portal. The system uses an Obsidian “second brain” for shared memory and context, includes voice features like Jarvis, supports VPS setups, and is distributed and updated via the AI Profit Boardroom community.00:00 Build Your AI OS00:48 Why Agent OS Wins02:18 Hermes Desktop Limits04:09 Custom UI Examples04:36 Cut Costs With Free07:24 Obsidian Memory Brain09:29 AI Video Factory11:45 n8n Versus Agent OS12:50 Safety And Guardrails13:34 Free Models Setup14:48 Jarvis Voice Control16:12 VPS And Updates17:10 Build One Workflow18:28 Model Swap Flexibility19:28 KimiCode Review20:32 Community Wins Showcase22:47 Markdown Token Saver23:05 Wrap Up And Next Steps

Fusion (OpenRouter) Lets You Combine Multiple Models to Reach Fable-Level Intelligence for LessThe script covers a new OpenRouter Fusion API update that runs a prompt across a parallel panel of up to eight models (with web search and bash tools), then uses a judge model to extract consensus, contradictions, unique insights, and missing coverage before returning one fused answer. Fusion is presented as a way to boost benchmark performance and reduce token costs versus relying on a single frontier model, with tests on 100 hard deep-research tasks showing much of the lift coming from synthesis rather than diversity. Examples compare solo models versus panels, including a “budget panel” of cheaper models landing within 1% of Claude Fable 5 on intelligence tests, and demonstrations of using Fusion in chat and via API to generate outputs like SEO research and a clean landing page.00:00 Fusion Update Overview00:51 Panels Beat Solo Models01:43 Budget Panel Near Fable02:23 How Fusion Works03:05 Live Panel Demo03:53 Benchmark Results Breakdown05:02 API Integration Ideas05:50 Boardroom SEO Example06:56 Judge Fusion Output08:05 Draco Benchmark Explained09:10 Landing Page Results10:28 Wrap Up And Offers

Claude Banned: US Gov Pulls the Plug + New AI Alternatives

The US Government recently ordered Anthropic to pull their most powerful Claude models, sparking a major shift in the AI landscape. This video breaks down why Fable 5 was removed, the rise of Chinese competitors like GLM 5.2, and how to build a resilient 'Agent Operating System' for your business.

00:00 - Intro: The Claude Ban

00:46 - Why the US Government Banned Claude

02:59 - New Chinese AI: GLM 5.2 vs Claude Opus

06:00 - Kimmy K2.17: Long Horizon Agents

07:23 - Fusion Panels: Combining AI Brains

09:51 - Hermes Automation & NotebookLM Upgrades

12:29 - Google Managed Cloud Agents

13:50 - How to Build a Resilient AI Strategy

GLM 5.2 vs Kimi K 2.7 vs Opus 4.8: Which AI Model Builds the Best Apps?The script compares GLM 5.2, Kimi K 2.7, and Opus 4.8 by giving them identical build prompts inside an agent operating system and judging results across five tests: a Temple Run–style voxel runner (GLM 5.2 best), an inner solar system/orbit HUD simulation (Kimi K 2.7 best for zoom, speed, and customization), a liquid-in-a-bowl particle/metaball interaction (GLM 5.2 best), an Apple-style AI model landing page (GLM 5.2 best), and a neon arcade game (GLM 5.2 most fun). The narrator notes GLM 5.2 is very new and not yet on OpenRouter, contrasts origins and context windows, highlights that Kimi and GLM can be used inside AI agents unlike Claude/Opus, and concludes GLM 5.2 wins four of five tests while promoting the AI Profit Boarding community and agent OS download.00:00 Model Showdown Setup00:57 Test 1 Temple Run Runner02:01 Test 2 Solar Orbit Map03:27 Test 3 Liquid Metaballs04:39 Test 4 Apple Style Landing Page05:40 Test 5 Neon Arcade Game06:24 Benchmarks And Model Specs07:51 How To Use Them Together08:15 Get The Agent OS09:30 Community Wrap Up

GLM 5.2 + Hermes Agents: Building Videos, Games, and SEO Websites with AI TeamsThe script explains how the GLM 5.2 update performs when plugged into Hermes Agent, showing a video created by a team of agents that generated an avatar, voiceover, script, editing, effects, and portfolio visuals. It contrasts GLM 5.2 with Claude, noting Claude is harder and more expensive to integrate with Hermes via API, while GLM 5.2 can be added as a model/profile and used to run coordinated agent teams. The speaker highlights projects built with GLM 5.2, including 3D games, an operating system with apps, and a multi-agent website/SEO workflow with roles like researcher, writer, builder, and judge. It also describes ways to organize agents (profiles, chats, Paperclip org charts, Kanban boards, and voice-activated Hermes Jarvis) and promotes Goldie Agency and the AI Profit Boardroom for setup, tutorials, and coaching.00:00 GLM 5.2 Meets Hermes00:11 AI Avatar Video Demo01:23 Games and OS Builds01:44 Head to Head Model Tests02:13 Connect GLM in Hermes02:50 Agent Profiles and Roles03:45 Team Coordination Workflows05:07 More Visual Project Examples05:37 SEO Website Case Study07:56 Why This Beats Claude09:33 Setup Tips and Mindset10:28 Recap and Next Steps10:41 Join the Profit Boardroom11:10 Final Thanks

GLM 5.2 Just Dropped: Building 3D Games, an Agent OS, and Content Fast (Zed AI)The script showcases China’s new GLM 5.2 flagship model from Zed AI and what the creator has built with it within a few hours, including multiple 3D/open-world style games, an agent “operating system” with apps and a CLI integration, and visual experiments like particle galaxy/fireworks and plasma wallpapers. They explain they improved earlier projects made with Kimi K 2.7 by plugging them into GLM 5.2, and also used a Kanban-orchestrated team of AI agents to script and assemble an AI-avatar video with music and to create an SEO-optimized blog with three articles. They highlight GLM 5.2’s coding plan value, planned local/open-source availability, 1M-token context window, long-horizon performance, compatibility with existing tools, and how to access it via API and CLI.00:00 GLM 5.2 Demo Builds00:50 How The Games Were Made01:19 Agent OS And CLI Apps01:55 Video And Blog Automation02:46 Local And Open Source Plans03:34 More Fun Project Demos04:06 Specs And Ecosystem Integrations04:57 Pricing Setup And Workflow05:54 Agent OS Memory And Switching Models07:37 How To Get Started Fast08:09 Limitations And Buggy Bits08:21 Wrap Up And Recommendation08:32 AI Profit Boardroom Pitch09:30 Community Training And Support10:12 Final Call To Action

Rank #1 with AI SEO: Hermes Agent Auto-Creates, Edits & Judges Avatar Videos in One PromptThe presenter demonstrates how Hermes Agent can generate fully edited AI avatar videos from a single prompt for AI SEO, showing three examples created with Hermes plus KimiK 2.7 (including scripts, edits, camera moves, and avatar voice/face) without manual work. They explain that targeting a keyword with these videos can help rank in Google, including AI Overview/AI Mode, and show proof of rankings. The workflow uses an agent “team” (writer, editor, judge, and KimiK 2.7) managed in a Kanban board, where a judge agent scores outputs and forces iterations until quality is high; music is created via Suno API, and video creation/editing is handled with HeyGen CLI integrated into the agent system. The episode ends by offering the full setup, tutorials, and files in the AI Profit Boardroom.00:00 AI SEO Video Breakthrough00:34 Prompt Made Video Demo01:43 Second Video Preview02:12 Music and Editing Stack02:31 Ranking Proof on Google03:08 Third Video and Judge Loop03:49 Kanban Agent Workflow04:34 Tools Behind the Scenes05:32 Results and Use Cases07:51 Setup Profiles in Hermes08:24 Recap and Next Steps08:58 Get the Agent OS09:50 Final Thanks