AI News Today | Julian Goldie Podcast

Google I/O 2026: Gemini 3.5 Flash, Antigravity 2.0 & Spark—Why the Agentic Era Just StartedThe script argues that Google I/O 2026 marks a major shift as Google rebuilds its ecosystem around AI agents, citing rapid growth in token processing (3.2 quadrillion per month) and user adoption for Gemini and AI Mode in Search. It highlights Gemini 3.5 Flash as a faster, cheaper, more agent-capable model and introduces Antigravity 2.0 as a standalone desktop command center for running many agents in parallel, showcased by a demo where 93 sub-agents autonomously built an operating system. It also covers Gemini Spark, a 24/7 cloud-based personal agent with tool access via connectors, plus new Search features like Information Agents, generative UI, and a universal shopping cart. The script emphasizes that results compound when agents are paired with a memory-based “agent operating system,” and promotes the creator’s AI Profit Boardroom setup to implement it.00:00 AI Takeover Begins00:27 Adoption Numbers Explode01:10 Gemini 3.5 Flash02:35 Antigravity 2.0 Agents03:59 Why Results Feel Generic05:02 Gemini Spark Always On07:21 Search Information Agents08:32 Google Ecosystem Shift09:13 Objections And Reality10:40 Build An Agent OS12:43 Action Steps And Setup14:28 Final Roadmap Recap

Google Gemini 3.5 Flash: Build Websites & Run AI Agents 24/7 in an Agent OS DashboardThe video introduces Google’s newly released Gemini 3.5 Flash and demonstrates how to use it to automate work, build and deploy full websites, and run multiple AI agents in parallel inside an Agent OS “mission control” dashboard. It shows Gemini 3.5 Flash’s speed and highlights benchmark gains over prior Flash and even Gemini 3.1 Pro on coding and agentic tasks (e.g., 76 vs 58 on TerminalBench and 83.6% vs 78% on agentic tasks). The presenter explains integrating Gemini via CLI and OpenRouter into tools like Hermes and pairing it with Antigravity 2.0 sub-agents to execute multi-step workflows, create animations, generate images, and manage outputs that compound over time. It also notes Gemini Spark uses 3.5 Flash but is beta-only for US Google AI Ultra users, while 3.5 Flash is available to everyone.00:00 Gemini Flash Reveal01:13 Speed Demo Setup01:41 Agent OS Dashboard Tour03:20 Benchmarks Key Benefits04:50 Spark Availability Options05:42 OpenRouter Hermes Setup07:25 What Is An AI Agent08:47 Agent OS Explained10:22 Goldie Command Layers13:56 Get The Agent OS14:38 Belief Shifts Examples17:34 Recap Next Steps

Hermes + AntiGravity Masterclass: Build an Agent Operating System (Context, Models, Migrations & Automations)Julian shares his setup for combining Hermes and AntiGravity so AI agents understand personal context and improve over time, then answers community questions about AntiGravity 2’s changes (removal of the built-in IDE/terminal), agentic workflows, and whether n8n is still worth using versus faster Hermes automations. He covers handling long-document context limits with model choice (including Claude), compacting, and chunking; recommends building a personal context engine in Obsidian over NotebookLM for long-term memory; discusses open-source/local model options via Ollama and LM Studio (including Hermes-focused and quantized models, plus Gemma as a lightweight choice); and explains migrating from OpenClaw to Hermes. He also outlines agency use cases like deploying SEO content, building/ranking sites, Twitter growth, and AI avatar marketing, and invites viewers to the AI Profit Boardroom for training, guides, coaching calls, and testimonials.00:00 Hermes Meets Antigravity00:39 Community Q&A Kickoff00:44 Antigravity 2 Changes01:24 n8n Workflows and Training01:39 Boosting Agent Context02:39 Automation Content Ideas03:25 Budget Friendly Setup04:17 Is n8n Still Worth It05:10 Beginner Automation Advice05:55 Beating Context Limits06:48 Infinite Context Engine08:11 Best Local Models09:50 Obsidian vs NotebookLM10:36 Migrating to Hermes11:21 Claude Code in Antigravity12:05 Agency Stack and Processes13:18 ChatGPT vs Gemini13:39 Join the AI Profit Boardroom

Google Antigravity 2.0: Build a Full AI Agent Command Center (Agent OS + Memory + Gemini 3.5 Flash)Google Antigravity 2.0 is presented as Google’s most advanced agent platform, redesigned with a new UI and powered by Gemini 3.5 Flash, enabling multiple autonomous agents to work in parallel on tasks like building websites, creating content, generating images, and automating work. New features highlighted include dynamic sub-agents, scheduled tasks, artifacts, project grouping, live voice transcription, and added customization via MCPs. The script shows how connecting Antigravity via the Antigravity CLI into an Agent OS “command center” improves usability, organization, and cross-agent collaboration (e.g., Hermes, Gemini, Claude) while keeping history and files in one workspace. A key differentiator is adding a memory loop using Omi (passive daily capture) exported into an Obsidian vault, giving agents persistent context and creating a feedback flywheel the script calls the “Goldie Gravity Grid.”00:00 Antigravity 2.0 Hype00:56 New Features Tour02:27 Agent OS Integration03:08 CLI Demo Setup05:07 Website Builds Showcase07:52 Models And APIs08:42 Memory With Omi12:35 Goldie Gravity Grid15:47 Why Standalone Fails17:13 Mindset And Roadmap19:08 Final Setup Recap

Google Gemini AI Leaks Before I/O 2026: Desktop Agent, Spark Mode, Stream-to-Cursor, Veo Video + Gemini 3.5/3.2Ahead of Google I/O 2026, leaked reports suggest major upgrades to Gemini, including a new desktop app with standard chat plus “Spark Mode,” a local agentic workspace that connects to folders, reads files, runs scripts, organizes documents, and syncs with Google Drive. The script highlights “Stream to Cursor,” a floating overlay that lets Gemini read the context of whatever window your cursor is on, alongside a persistent voice interface (“Gemini Live”) and reported local “skills” support for attaching custom scripts to automate repeatable tasks. Leaks also mention Veo video generation integrated into the Gemini desktop experience, plus new models reportedly Gemini 3.5 Pro and Gemini 3.2 Flash, with testers emphasizing speed and workflow impact. The video ties these leaks to building a connected system via an “Agent OS” dashboard integrating Gemini, Claude, Hermes, OpenClaude, and NotebookLM for business automation.00:00 Massive Gemini Leaks01:24 Spark Mode Desktop Agent02:36 Stream to Cursor Overlay04:00 Build a Connected System05:46 Local Skills Support06:28 Veo Video Inside Gemini07:27 Gemini Live Voice Agent08:02 Gemini 3.5 vs 3.2 Flash10:03 Why This Matters for Business12:13 Agent Layer Arms Race13:04 Boardroom Walkthrough and Wrap

Claude Mythos Leaks: Regulators Briefed, Cloudflare 2-Hour Patches, Apple M5 Bypassed (What It Means for Business)The script claims new Claude Mythos leaks indicate a major capability jump, alongside Sam Altman’s post that ChatGPT has improved and rumors of a GPT 5.6 release. It cites the UK AI Security Institute’s updated assessment of Mythos as a notable leap, including Mythos solving the previously unsolved “Cooling Tower” cybersecurity evaluation in 3/10 attempts and doubling autonomous task length in months. Cloudflare reports Mythos performs at a senior researcher level, prompting some teams to adopt two-hour patch SLAs. A Tom’s Hardware story says Mythos Preview helped find a local privilege escalation bypassing Apple’s memory integrity enforcement on the M5 chip. The Financial Stability Board and other finance regulators are being briefed, and the narrator argues businesses should build Claude “agent operating system” workflows now rather than wait, promoting the AI Profit Board/Ballroom offering.00:00 Mythos Leaks Shockwave00:57 Sam Altman Teases Upgrade01:49 Mythos AISI Breakthrough02:52 Cloudflare Two Hour Patches03:37 Apple M5 Exploit Found04:44 Finance Regulators Briefed05:48 What It Means For Business06:44 Agent OS Pitch07:46 Three Beliefs Holding You Back10:11 Claude Agent OS In Practice10:56 Final Signals And Next Steps

OpenClaw 5.18 Is Live: Real-Time Android Voice Talk Mode, Stable Groq OAuth, Smarter Browser Pop-Up HandlingOpenClaw 5.18 launches with real-time Android Talk Mode that lets users speak to an AI agent aloud and hear responses immediately, including interruption handling, on-screen transcripts, and continued tool access while talking. Groq OAuth is now stable, allowing Super Groq subscribers to log in and use Groq without an API key or extra cost, gaining access to Groq models and related tools. Browser automation is improved because agents can now see and respond to blocking pop-ups and dialogs that previously caused workflows to stall. Telegram reliability fixes keep replies and generated media in the correct forum topics, repair scheduled message link formatting, and stop media downloads when Require Mention is enabled. Additional stability upgrades include faster gateway startup, Discord voice and image handling fixes, more reliable sub-agent handoffs, isolated plugin installs, redesigned Mac settings with faster connection preferences, shorter tool descriptions to save context, and clearer error messages, along with backup and testing advice before updating.00:00 Whats New in 51800:48 Android Talk Mode02:31 Groq Login Stability03:17 Browser Popup Handling04:16 Telegram Reliability Fixes05:20 Agentic OS Pitch06:07 Stability and Performance08:26 Update Advice and Backup09:08 Final Call to Action

Build a NotebookLM + Claude Agent OS Dashboard (Goldie Infinite Knowledge Engine) to Auto-Generate Videos, Podcasts & MoreJulian demonstrates a local “Agent OS” dashboard that integrates Claude with Google’s free NotebookLM to manage notebooks in one place, chat with sources, trigger Studio outputs, and automatically pull generated assets—videos, podcasts/audio overviews, slide decks, mind maps, infographics, flashcards, quizzes, briefs, and reports—into an organized gallery. He explains the “Goldie Infinite Knowledge Engine” as a three-layer loop: a knowledge vault (NotebookLM sources like PDFs, websites, and research), an agent operating core (Claude-connected control room with an MCP bridge), and an infinite loop where new sources continuously regenerate new content. He contrasts manual content creation with automated workflows, highlights a separate Obsidian-based memory vault (“Infinite Context Engine”) to reduce prompting and token usage, and references the AI Profit Boardroom for templates, prompts, tutorials, and support.00:00 Dashboard Overview01:10 NotebookLM Basics02:21 Why Agent OS Matters03:13 Getting The Template04:01 Systems Win In AI04:50 Three Layer Engine07:34 Old Way Vs New08:40 Memory Context Engine09:43 Not Technical Objections11:23 What Is Agent OS12:27 How Integration Works13:42 Recap And Offer15:54 Multi Agent Automation Tips21:17 Token And Cost Savings25:17 Final Wrap Up

Hermes Agent v0.14: Faster, Smarter—But It Only Shines Inside a Full Agent OSThe script explains why Hermes Agent v0.14 is a major upgrade—adding one-line installation via pip, Groq OAuth support, and faster ComputerUse agents—while arguing these gains reach full value only inside a complete Agent OS. It contrasts running Hermes in a raw terminal (no dashboard, visibility, history, or management) with an Agent OS that provides a command center for task tracking, Kanban planning, content history, and a media studio to preview generated images, video, audio, and text-to-speech. The update also enables Groq 4.3 OAuth with a million-token context window and built-in Twitter search without extra APIs, turning Hermes into a full media creation workflow when organized in the OS. The script emphasizes adding a memory system (e.g., Obsidian) so Hermes can personalize outputs, and promotes a downloadable template, tutorials, and coaching via the AI Profit Board community.00:00 Hermes 0.14 Big Shift00:45 One Line Install01:05 Groq OAuth Upgrade01:26 Why Agent OS Matters02:20 Dashboard Tasks and History03:08 ComputerUse Gets Faster04:18 Background Automation Demo05:24 Groq 4.3 Million Tokens06:32 Media Studio Workflow07:59 Memory Makes It Smart10:00 One Command Center11:04 Template and Coaching Offer12:25 Wrap Up and Next Steps

Hermes Agent OS: Automate Your Business (Video Editing, AI SEO, Memory Setup & n8n Replacement)Julian explains how to use Hermes Agent OS to automate business tasks like video creation/editing, content and AI SEO workflows, and replacing traditional tools like n8n. He answers community questions, demonstrating video editing with the Hyperframes skill (including HeyGen avatar video workflows) and video generation with Remotion by installing from GitHub and running simple prompts. He advises keeping setups simple, following the one-line install/quick start for Hermes Agent and Workspace, and improving agent performance by giving it more personal context via a memory system. Julian prefers using a local Obsidian vault (optionally paired with Omi for daily memory capture) over built-in Hermes memory providers. He also shows an AI-driven website/content deployment process using Netlify and emphasizes iterating one useful automation at a time, while highlighting his AI Profit Boardroom resources, prompts, zip files, trainings, and coaching calls.00:00 Hermes Agent OS Overview00:36 Hyperframes Video Editing01:38 Installing Agent OS Dashboard03:00 Quick Start Setup Tips03:42 Memory With Obsidian05:05 Remotion Video Creation07:33 Replacing n8n Automations08:38 Choosing Models and APIs09:43 First Day Workflow Strategy11:25 Organizing Your Vault12:41 AI SEO Sites With Netlify15:53 Better Memory Than Defaults17:11 Paperclip and Workflow Migration18:31 Limits and Faster Automation21:42 Community and Final Wrap