AI News Today | Julian Goldie Podcast

How I Built My Own Agent Operating System (Agent OS) With Claude — No Coding in 2026The script explains how the creator built a free, custom “Agent OS” from scratch using Claude through conversation rather than coding, describing it as a browser-based dashboard where multiple specialized agents coordinate in one place with voice input, journaling, goals, and an Obsidian-connected memory system. It contrasts “vending machine” AI usage with a system that knows the user’s goals, tasks, notes, and history via a front end, a set of role-based agents, and a memory/data layer. The creator shares lessons learned, including starting with one simple agent like daily planning and expanding over time, and emphasizes that the key skill in 2026 is clearly describing outcomes and writing strong system prompts. The script also promotes the AI Profit Boardroom’s prebuilt Agent OS, roadmap, coaching, tutorials, and community, and outlines steps to start building: pick one workflow problem, have Claude scaffold a dashboard, set up Omi + Obsidian, and iterate daily.00:00 Building Agent OS00:25 What Agent OS Is01:01 Inside the Dashboard02:08 Core Components02:40 Start Small Strategy03:40 Building With Claude04:34 Outcome Thinking Skill04:53 Boardroom Shortcut Offer06:03 Memory System Unlock07:45 Not Technical Barrier08:42 Agents In Action10:06 Data Layer Basics12:18 Getting Started Steps13:17 Join and Final Push

Hermes Beats OpenAI Codex, DeepMind Drops Antigravity CLI, and Musk Teases AI SatellitesThe episode covers four fast-moving AI stories relevant to building AI-driven businesses in 2026: OpenAI Codex is reportedly slowing down and losing reliability, with over 1,700 posts in 24 hours prompting public switching—mainly toward Claude Code—while early benchmarks show the self-improving Hermes Agent outperforming Codex on key speed tasks. Google DeepMind also released Antigravity CLI, a new terminal-based coding agent tool positioned as a replacement for Gemini CLI, enabling more complex agentic coding workflows from the command line with free testing under token limits. Elon Musk teased a forthcoming detailed design reveal for TerraFab, a Tesla/SpaceX/xAI project aiming to put AI compute in orbit via satellite-based data centers powered by constant solar energy and different cooling dynamics, potentially lowering AI costs industry-wide. The script advises building AI workflows in modular layers so tools can be swapped without rebuilding everything, and promotes the AI Profit Boardroom’s coaching and agent operating system.00:00 AI News Rundown00:33 Codex Slowing Down01:30 Hermes Beats Codex01:57 What It Means Business03:54 DeepMind Antigravity CLI05:21 Google Agentic Future06:02 Musk AI Satellites07:22 Cheaper Compute Impacts08:27 Switch Tools Without Rebuilds09:49 Zoom Out 2026 Rhythm10:46 Final Sign Off

Claude Mythos-1: 23,019 Vulnerabilities Found + Sonnet 4.8 Leaks (June Release?)The script covers two major Claude updates: Mythos-1 and leaked details on Sonnet 4.8. It claims Claude Mythos Preview scanned 1,000+ open-source projects in 30 days, found 23,019 vulnerabilities (6,202 high/critical) with a 90% true-positive verification rate, including large sets of issues for Cloudflare and Firefox, and it reportedly blocked a $1.5M wire-fraud attempt at a partner bank by detecting behavioral patterns without human intervention. It describes the challenge that AI can find vulnerabilities faster than humans can patch them, Anthropic’s response via Claude Security auto-patching, and an open-sourced bug-finding pipeline, plus Cisco building a similar system. It also discusses brief sightings of Mythos-1 Preview in Claude Code/Security despite public caution, Jack Clark’s Oxford remarks on AI risk, and accidental leaks indicating Sonnet/Opus 4.8 testing in Google Vertex with an expected mid-to-late June timeframe and new reasoning improvements.00:00 Mythos-1 and Sonnet Leaks00:26 Mythos Scanning Results01:05 Real World Bug Finds01:32 Bank Fraud Stopped02:13 Patch Backlog Problem02:54 Claude Security Auto Patching04:16 Mythos-1 Preview Spotted05:05 Safety Messaging Tension06:27 Sonnet 4.8 Leak Details06:55 Vertex Testing Signals07:27 Expected Sonnet 4.8 Upgrades08:11 Big Picture Takeaways09:09 Business Implications09:28 Three Common Objections10:25 How to Get Started11:15 Final Wrap Up

uild a Free AI Agent Operating System (One Dashboard: Voice, Goals, Video, Search & Memory)The presenter explains how to build a free, unified “agent operating system” that makes AI agents far more powerful by combining voice chat, autonomous goal execution, video creation, research/search, and a searchable memory system in one dashboard. After spending about 100 hours refining their own setup, they outline a five-part “build anything” stack: voice agents for hands-free conversations, Hermes goal mode for long-horizon autonomous work that runs while you’re away, a video agent using Hyperframes CLI, a Twitter-linked search tool that returns markdown guides, and an Obsidian-based knowledge vault that stores every output so nothing is lost and can be reused as context. They contrast this with juggling multiple tools and subscriptions, share community examples, discuss Hermes login/model options, and mention an optional ready-made zip, playbook, coaching, and community in the AI Profit Boardroom. They note SEO/blog content can be deployed directly to Netlify without GitHub.00:00 Build an Agent OS02:22 Why an OS Beats Tools03:31 One Dashboard Demo05:32 Five Power Moves06:13 Voice Agents Setup06:36 Goal Mode Autonomy07:10 Video Agent Workflow07:44 Search and Research08:11 Memory in Obsidian11:00 How to Build Yours12:52 Q&A and Access Options17:26 Recap and Next Steps19:36 Final Q&A and Outro

NotebookLM’s Free Upgrade: Build a One-Click Agent OS with Hermes, Hyperframes & ObsidianThe script explains a new free NotebookLM upgrade and a recommended workflow that plugs NotebookLM into an agent operating system to automatically organize notebooks and outputs, sync everything to an Obsidian memory vault, and turn research into ready-to-post content (videos, podcasts/audio views, infographics, slide decks, mind maps, flashcards, quizzes, tables, and reports) in one click. It introduces a framework called Goldie Knowledge Studio with a five-layer pipeline: capture notebooks, generate content in a studio UI, auto-save/sync to Obsidian, organize assets in one place, and publish. The key improvement is placing NotebookLM beside the Hermes Video Agent and using Hyperframes to edit and improve generated artifacts, plus optional AI avatar customization to reduce generic output. It also mentions using an MCP to connect tools, adding Google’s Antigravity via CLI (replacing Gemini CLI by June 18), and offers a paid “AI Profit Boardroom” for the full setup, tutorials, coaching, and community.00:00 NotebookLM Upgrade Hook00:40 Why Most Use It Wrong02:12 Agent OS Overview03:34 Goldie Knowledge Studio03:59 Five Layer Workflow05:08 Obsidian Memory Vault06:59 Hyperframes Editing Bridge08:30 Old Way vs New Way11:41 Shared Dashboard Benefits14:17 Live Q&A Setup Tips16:26 Recap and Next Steps18:22 Final Thanks and Outro

Build a Hermes Agent Operating System (Memory Layer, VPS Setup, Workflows & Context Tips)The video explains how Hermes Agent has become much more powerful and shows how to build an “agent operating system” that organizes multiple AI agents and workflows in one mission-control setup that can run tasks automatically 24/7 without needing a powerful or technical computer. Using community questions, it covers creating and improving workflows (including a video workflow), adding a memory layer via an Obsidian knowledge graph so agents can store, update, and organize personal notes for more relevant outputs, and why the system improves continuously over time. It compares Paperclip to a custom Hermes Kanban-based team of agents, demonstrates running an agent OS on a VPS using Hostinger with Cloudflare Tunnel/Access for anywhere access, discusses organizing files and skills in Hermes, handling context via /new and /compact, and suggests focusing on Hermes (and Claude) while keeping things simple to avoid overwhelm.00:00 Hermes Gets Supercharged00:42 What Is An Agent OS01:32 Build Workflows Fast01:52 Add A Memory Layer03:10 Paperclip Vs Hermes04:07 Run It On A VPS05:24 Organize Hermes Workspace06:51 Delegate Big Projects07:38 Sync Telegram And Obsidian08:19 Context And Best Tools09:31 Groq Talk Mode Demo10:50 VPS Vs Local Options12:07 Obsidian Migration Advice13:16 Don’t Get Overwhelmed13:59 Community Wrap Up And CTA

OpenClaw 5.24 Beta: Discord Meeting Notes, iMessage Approvals, Live Voice Steering + Token-Saving ImagesOpenClaw 5.24 is a GitHub pre-release beta with new features including a meeting notes plugin that lets an agent join Discord voice calls, capture who said what, auto-start recording, and provide transcripts for summaries, action items, and search. The update adds iMessage action approvals via thumbs up/down, more interactive voice conversations with real-time status checks, cancel/redirect/queue controls, and Discord wake-name gating. Images now auto-adjust quality/compression by model with override modes to save tokens, and gateway startup is faster by reusing persistent data. Dashboard chat search is improved, the install package is smaller, and Grok web search moves to OAuth login rather than separate API keys. Fixes include Telegram topic concurrency, follow-up cancel-signal issues, Claude image support detection, DeepSeek tool cleanup, faster non-blocking memory search, and quicker skill refresh/updates. The script warns not to deploy beta on live setups and outlines backup, beta update, testing, and rollback steps.00:00 Biggest 5.24 Beta Features00:50 Beta Warning and Safe Testing01:10 Discord Meeting Notes Plugin02:20 iMessage Emoji Approvals02:58 Real Time Voice Controls03:41 Smarter Images and Faster Startup04:38 Dashboard Search and Grok OAuth05:32 Smaller Install and Ecosystem Pitch06:24 Stability Fixes and Performance08:24 How to Update and Roll Back08:58 Agent OS Offer and Wrap Up

OpenClaw Studio (Agent OS): One Dashboard for Images, Video, Voice, Apps + Infinite MemoryThe script introduces OpenClaw Studio inside an Agent Operating System, positioning it as a free, customizable dashboard that turns OpenClaw from a basic chat into a “studio” where outputs never get lost. It demonstrates built-in Twitter search (with a markdown previewer), image generation, video generation with format and resolution options, text-to-speech with selectable voices, and real-time voice conversations with an AI agent, plus a workspace that previews and organizes everything created (images, videos, voice notes, apps, websites) with local folders and download access. A control room shows agents, logs, scheduled tasks, and memory. The key missing piece for most users is a memory system, implemented via OpenClaw + Obsidian and an “infinite context engine” that logs work and improves results. The creator offers prompts, tutorials, zip files, and updates via the AI Profit Boardroom community.00:00 OpenClaw Studio Reveal00:54 Tour of Studio Features02:29 Twitter Search and Markdown Preview03:23 Voice and Video Creation Hub03:42 Old UI vs New Dashboard06:27 Workspace Previews Everything07:58 Control Room and Agents08:37 Build It Yourself or Get Template10:29 What Agent OS Really Is13:28 Old Workflow vs New Workflow14:40 Memory Layer with Obsidian16:42 Workspace Buckets and Context Engine17:57 Free Setup and Common Objections20:20 Recap and Community Outro

Build Your Own Agent Operating System (Free 7-Layer Blueprint + Infinite Context Memory)The speaker presents a step-by-step blueprint for building a personal “agent operating system” that centralizes multiple AI agents into one local dashboard with searchable, previewable history and persistent memory. After spending 100+ hours and nine failed attempts duct-taping tools together, they outline five key mistakes to avoid: treating AI as a tool instead of a system, paying for subscriptions before using free options, relying on weak in-app memory, building brittle n8n automations, and letting outputs disappear into random folders. The solution uses a seven-layer architecture—foundation, memory (Omi + Obsidian), brain (routed models), agents (e.g., Hermes, OpenClaw, Claude Code), command center (Next.js Mission Control), production surfaces (studio/SEO/goals/workspaces), and a loop that writes outputs back into memory—so agents can work autonomously and improve over time.00:00 Why Agents Need OS00:56 Dashboard Tour02:13 Seven Layer Stack03:08 Why I Built This05:36 Nine Failed Attempts06:18 Mistake One Tools07:12 Mistake Two Free Stack09:21 Mistake Three Memory11:14 Mistake Four Skip n8n13:31 Mistake Five Organize Outputs15:35 Old Way vs New17:26 Blueprint Seven Layers18:03 Seven Layer Blueprint19:15 Hardware And Memory20:14 Models And Agents22:08 Mission Control Dashboard22:53 Production Workflows24:37 Feedback Loop Layer25:23 Build Journey And Proof26:52 Limiting Beliefs Debunked28:09 Live Q And A29:56 Final Recap And Offer31:05 Community Coaching Support32:34 Closing Thoughts

Hermes Agent Goals Mode: Run Autonomous 24/7 Tasks for Free (No More “Keep Going”)The video introduces Hermes Agent’s new free “Goals” update, which lets users set a long-horizon objective once and have the agent run autonomously until completion without repeated prompts. It demonstrates launching a goal to create and host a blog, then explains how Goals mode works via a persistent standing objective, a conservative judge model that decides after each turn whether to continue, and an adjustable default 20-turn budget with auto-pause and resume. The script contrasts this with the old workflow that required constant “keep going” prompts, context loss, and human babysitting. It covers commands like /goal status, pause, resume, and clear, notes local goal-state persistence and per-goal directories, briefly discusses preferred tools/LLMs (including Grok 4.3 and Antigravity CLI), and ends by promoting the AI Profit Boardroom for an agent OS setup, roadmap, coaching, and community.00:00 Goals Update Overview00:43 Launching Your First Goal02:11 How Goal Mode Works03:37 Old Way vs New Way05:50 Live Q&A Tooling Talk06:36 Judge Budget Loop Explained07:28 Commands Examples Persistence08:15 Free Setup Use Cases08:59 Common Objections Recap09:56 Boardroom Offer Closing11:03 Thanks For Watching