AI News Today | Julian Goldie Podcast

Infinite Video Engine: One Topic In, Fully Edited AI Video Out (Live Web Research + Avatar + B-Roll)The video introduces the “Infinite Video Engine,” a new system built into an Agent OS that can research a topic live from the web, write a script, generate an AI avatar and voiceover, create matching B-roll, and assemble a finished, edited video from a single topic prompt. An example video is shown about GLM 5.2, described as an open-source coding model with 753B parameters, a 1M token context window, a near tie with Claude Opus on long coding tasks, a 62 score on SWE-Bench Pro beating GPT 5.5, and MIT-licensed weights. The creator explains the five-stage workflow, emphasizes speed versus manual video production, highlights scene-by-scene planning with B-roll descriptions and research notes, and says users can review and edit scripts before generating the final video.00:00 Infinite Video Engine Intro00:25 Demo Video Example01:25 Live Topic Script Build02:39 Scene Planning And B Roll03:50 Five Stage Workflow05:07 Old Way Vs New Way05:41 Agent OS Dashboard Tour07:25 Quality And Use Cases08:24 Editing And Workspace Preview09:25 Finished Fusion Video Playback10:20 Get The System And Community11:19 Final Thanks And Goodbye

Build Your Own AI Agent OS: The Ultimate Step-by-Step Guide

Stop juggling 14 different AI tabs and learn how to build a unified Agent Operating System with a shared brain. This guide shows you how to orchestrate multiple AI models, automate complex workflows, and maintain a persistent memory for all your projects.

00:00 - Intro: What is an Agent OS?

02:35 - The Problem with Multiple AI Tabs

04:14 - The 6 Layers of an Agent System

05:19 - Building a Shared Brain with Obsidian

06:38 - Creating a Persistent Workspace

09:05 - The Old Way vs. The New AI Workflow

11:11 - Step-by-Step Implementation Guide

15:31 - How to Scale Your AI Team

Claude Agentic OS: One Dashboard for All Your AI Agents (Shared Memory, Automation, Fusion & Hermes Jarvis)The script showcases a “Claude Agentic operating system” that centralizes multiple AI agents and models in one dashboard with a shared memory, mission control, and a pipeline that takes ideas to built apps with minimal approval steps. It highlights built-in tools like Hermes with API profiles, live chat, and voice control via a Jarvis-style layer, plus one-click generation for images, video (including AI avatars), voice, and music, with workspaces that save and preview everything created. The system integrates new model releases (e.g., Groq Build, GLM 5.2, Kimi K 2.7) and uses a Fusion panel where multiple models debate with a judge model for high-stakes decisions, providing resilience when models are removed. It automates daily workflows like SEO and content production using multi-agent teams and a quality-control judge, and promotes access via the AI Profit Boardroom with tutorials, updates, and coaching.00:00 Wild Agent OS Intro00:28 Hermes Voice Control01:12 Plugging New Models02:13 Agent Mastermind Hub02:51 Memory Galaxy System03:36 Model Bans And Fallbacks04:47 Content Factory Workflow06:46 Goldie Dream Team Vision07:20 Mission Control Dashboard08:04 Games Built In One Click09:18 Why Automation Wins11:00 New Updates And Fusion13:18 Pipeline From Idea To App13:58 Hermes Jarvis Hands Free14:45 Costs Objections And Urgency15:56 Join The Profit Boardroom16:57 Final Thanks And Outro

GLM 5.2 + Hermes Agent: Build a Multi‑Agent Content Machine (Kanban Workflow + Auto‑Publish)The script demonstrates GLM 5.2 integrated into Hermes Agent, including a voice-activated Hermes Jarvis setup and multiple model profiles, and showcases projects built with teams of agents such as open-world games, an agent operating system, and a fully automated “content machine.” Using a Kanban-board workflow, separate agents handle keyword research, writing, editing, and video production while a judge agent scores drafts, rejects weak or fabricated work, and forces iteration until quality passes, after which content can be published and deployed to a website. The speaker shows how to switch models via terminal commands or the Hermes model manager, highlights GLM 5.2’s near-Opus 4.8 benchmark performance on agentic tasks, emphasizes lower costs via GLM’s coding plan, and pitches the pre-built system available in the AI Profit Boardroom with coaching, tutorials, and a large community.00:00 GLM 5.2 Meets Hermes00:55 Content Machine Overview02:13 Connecting Models in Hermes03:08 Kanban Agent Workflow04:10 Reviewing Outputs and Iteration05:52 More Builds Agent OS and Games06:37 Benchmarks and Cost Advantage07:32 Why the Pipeline Wins09:05 Four Parts of the Machine10:24 Real SEO Use Cases11:14 Get the Prebuilt System12:01 Classroom Coaching Community12:58 Wrap Up and Goodbye

GLM 5.2 vs Claude Opus 4.8 (and Kimi K2.7): Side-by-Side Coding & Game Tests + BenchmarksThe video compares GLM 5.2 against Claude Opus 4.8 and Kimi K2.7 through side-by-side creative coding tests (raycaster engines, neon city flight, racing, solar system, voxel runner, liquid-in-a-bowl interaction, and landing pages) and discusses benchmark results and practical tradeoffs. Opus 4.8 generally produces smoother, more playable outputs and leads several tests, while GLM 5.2 wins some rounds (notably a labeled, interactive solar system, voxel runner, and the liquid demo) and scores #1 on Design Arena. The script emphasizes GLM 5.2 being MIT-licensed open source on Hugging Face and available via Ollama, cheaper on a coding plan, and easier to plug into agents via OAuth, while Claude is described as stronger overall but less flexible without paid API access. It also promotes an “agent operating system” and the AI Profit Boardroom community for setup files, tutorials, and coaching.00:00 GLM 5.2 vs Claude00:43 Open Source Advantages02:12 Raycaster Showdown03:21 Upgraded Raycaster04:25 Neon City Flight05:13 Racing And Solar06:19 Voxel Runner Winner07:13 Liquid Bowl Test07:41 Benchmarks Breakdown09:44 Landing Page Faceoff10:15 Final Verdict10:46 Agent OS Integration11:17 Boardroom Pitch13:19 Wrap Up

Hermes Agent OS Q&A: Build a Dashboard, Sync Agents, Stop Spirals, and Automate Your BusinessThe video answers top AI Profit Boardroom questions about using Hermes Agent to build an agent operating system, including how to run a business schedule with a simple, staged approach (start with a to-do list and add one automation per week) and three options for implementing it: a dashboard in the OS, a custom built list, or a scheduled daily recap inside Hermes chat. It explains how to get multiple agents working together using a single group chat to avoid copy/paste between tools, discusses naming agents by roles or by model/API for organization, and shares a five-line pre-build workflow (goal, current state, next action, exact question) to prevent agent spiraling alongside a “pinned plan” planner/implementer approach for high quality with cheaper models. It also covers sandboxing integrations like Google Drive/email, community examples (lead gen dashboard, LinkedIn automation on free local models), and ongoing additions like Kimi K 2.7, GLM 5.2, Fusion, pipelines, Kanban, SEO and video agent automation, plus Boardroom resources, tutorials, and coaching calls.00:00 Agent OS Overview01:18 Schedule and Dashboard Setup01:46 Simple To Do Automations04:04 Agents Working Together05:49 Naming Agents and Teams07:46 Stop Agent Spirals10:20 Drive and Email Sandboxing11:17 Community Builds and Feedback13:26 New Models and Demos14:52 Join the Profit Boardroom15:55 Final Thanks and Outro

Build & Automate Anything with GLM 5.2 (3 Simple Methods + Hermes Agent Workflow)The video demonstrates how to build and automate apps, websites, games, tools, and content using GLM 5.2, highlighting projects created quickly (including 3D games and an OS inside an agent OS) and emphasizing its visual quality and large token availability on the coding plan. It explains that GLM 5.2 is an MIT-licensed, open-source Chinese coding model from ZAI (Jerbo AI), available on Hugging Face, with benchmark comparisons claiming frontier-level performance versus models like Opus 4.8, GPT 5.5, and Gemini 3.1 Pro. The host outlines three ways to use it: via a CLI/dashboard for plain-English coding into a workspace, by plugging it into the open-source Hermes Agent (including voice activation and Kanban-based multi-agent workflows with a judge for iterative quality control), or by running it locally with Ollama and integrating it into tools like Claude Code and Codex.00:00 Build Anything Intro01:42 What Is GLM 5.202:07 Benchmarks vs Frontier03:33 Method One CLI Build04:29 Method Two Hermes Agent06:07 Kanban Content Machine08:44 Method Three Run Locally09:48 Why This Beats Old Way11:58 Key Takeaways Recap12:38 Agent OS Offer13:53 Boardroom Community Tour14:37 Final Thanks

3 New Chinese AI Agents That Build & Automate Anything (Free N2, GLM 5.2, Kimi K 2.7)The video showcases three autonomous Chinese AI agents—Next AGI’s N2 (temporarily free via OpenRouter with a ~262K context window), GLM 5.2, and Kimi K 2.7—plugged into an “agent operating system” to build and automate projects like open-world/3D games, websites, apps, and fully AI-generated videos. It demonstrates using N2 inside Free Claude Code (including voice control), organizing multiple model profiles and teams of agents in Hermes with a Kanban workflow, and adding quality control via a “judge” agent that critiques and drives iteration. The script also describes a Fusion API panel approach (combining model outputs and judging/fusing results) that benchmarks competitively, recommends GLM 5.2 as the favorite overall and N2 as the best free option, and promotes the AI Profit Board community and downloadable Agent OS bundle with prompts, tutorials, roadmap, weekly coaching calls, and ongoing updates.00:00 Meet Three AI Agents00:39 What They Can Build01:59 Free N2 With Claude Code02:50 Games And GLM OS Demos03:45 Hermes Agent Teams04:50 Quality Control Judge Agent06:09 Idea Factory To Gallery07:31 Why You Need Agent OS09:25 Benchmarks And Fusion Panel10:48 Which Model To Pick11:15 Get The Full System12:33 Final Thanks

Agent OS Q&A: Running Multiple AI Agents, Video Automation, SEO Workflows & Setup (Local vs VPS)The episode answers common questions about an Agent Operating System built for the Air Profitable community, explaining how it runs multiple agents in parallel with shared workflows using a Kanban/Hermes setup that assigns tasks to different agent profiles, and noting Hermes’ new asynchronous sub-agents feature for background delegation. It covers AI video generation options via a Studio using Groq/Minimax and free open-source Hyperframes, plus HeyGen for avatar videos. The script discusses using Hermes as a Perplexity Computer alternative and integrating Google Managed Agents API via Antigravity for cloud-based agents. It recommends running Agent OS locally for speed and privacy, shows UI customization and chat/vault management tips, advises using free cloud APIs over slow local models on Mac hardware, and promotes the AI Profit Boardroom for the zip file, tutorials, updates, and community support.00:00 Agent OS Q&A Intro00:27 Running Multiple Agents02:28 AI Video Generation Setup04:05 Hermes as Perplexity Alternative05:50 Local vs VPS Installation07:17 Customizing the Agent OS UI08:28 Adding Tools and Saving Chats10:05 Best Models for Mac Mini11:31 Hermes Desktop Troubleshooting13:11 VPS Hosting and SEO Automations14:08 Custom Business Workflows16:00 Wrap Up and Join Boardroom

Kimi K 2.7 Code High-Speed Update: Quality vs Fast vs No-Think (Real Speed Tests)Kimi K 2.7 Code has a new high-speed rollout with three “gears” for the same coding model: Quality (thinking on), Fast (thinking on but served faster), and No Think (thinking steps off for maximum speed but sometimes lower quality). The video compares side-by-side outputs and response times on three prompts—building a 3D solar system, generating a playable game, and creating a spiral galaxy—showing mixed results: some tasks are faster with No Think (e.g., 98s vs 53s vs 39s), while others can be slower, depending on what gets built. The script explains how thinking-on uses a hidden scratchpad while thinking-off writes immediately, and notes Moonshot’s quoted ~180–260 tokens/sec serving. It also demonstrates toggling modes in KimiCode within an Agent OS setup and promotes the AI Profit Boardroom for tutorials, tools, and coaching.00:00 Kimi K 2.7 Speed Update00:25 Three Gears Explained01:22 Solar System Speed Test02:26 Playable Game Comparison03:16 Spiral Galaxy Surprise03:58 Switching Modes Demo05:12 Thinking On vs Off06:27 Which Mode To Use07:00 Quality Mode Showcase07:46 Agent OS And Boardroom08:57 Wrap Up And Thanks