TWiT AI User Group

I'm not affiliated with TWiT, but I'm a Club TWiT member and join the AI User Group discussion each month. These episodes explore AI tools, workflows, and real-world applications.

Club TWiT: AI User Group #20 - Agents on the Loose!

2026-10-03 · 1:50:53

Club TWiT AI User Group #20 turned Leo Laporte and a crowded panel, including Darren Oakey, Larry Gold, Timothy Ingalls and Anthony Nielsen, loose on the messy reality of running autonomous agents. The group compared local models through hands-on benchmarks and swapped stories of strange prompt injections, edge cases that came back to bite, and subagents that went off the rails. They also dug into harness design, agent orchestration and careful prompting, and took a hard look at where security gaps really hide. Throughout, they weighed how much human oversight is enough once agents start fixing, or breaking, systems on their own.

Watch on YouTube

Club TWiT: AI User Group #19.5 - A Hallucination of Agents

2026-09-26 · 2:29:53

Club TWiT AI User Group #19.5, A Hallucination of Agents, followed Leo Laporte as he benchmarked the latest local and cloud models, including GLM-5.3-Flash, Kimi K3, and Quen 3.8-Flash, across his home agent setup, comparing quantization tricks, speed, and agentic reliability. He and guest Anthony Nielsen swapped hardware tips on Sparks, Macs, and a 3090, debated voice models, and peeled back how his AI household ran daily tasks, assigned voices, and handled memory. Practical test results mixed with quirky agent banter, leaving the long unscripted session at its most hands-on.

Watch on YouTube

Club TWiT: AI User Group #19 - Leo Gets (DGX) Sparky

2026-09-12 · 1:41:35

Club TWiT AI User Group #19 followed host Leo Laporte as he got Sparky with a DGX, diving into the reality of running local AI with Darren Oakey, Larry Gold, and fellow Club TWiT members. The group debated dual Sparks versus souped-up Macs, wrangled old gaming PCs back to life, and traded benchmarks against epic energy bills while unpacking homegrown setups, harnesses, and agent orchestration. Real talk ranged across open models, subscription regret, workflow hacks, late-night AI obsession, and the slightly dangerous thrill of non-stop token burning.

Watch on YouTube

Club TWiT: AI User Group #18 - Hacking Your Workflow

2026-08-22 · 1:58:15

Club TWiT AI User Group #18 turned the floor over to a show-and-tell of AI-powered workflow hacks, as Leo Laporte hosted members including Darren Oakey, Danno, Michael (Alakazip), Craig McFarlane, and Anthony Nielsen. The group demoed projects spanning home automation, custom agent harnesses, ad-sales automation, smart-device scripts, and video-creation bots, mixing hardware debates with live problem-solving. Fast-paced geekery and memorable stories from tech, media, and enterprise filled the session as the crew compared notes on hacking everyday work with AI.

Watch on YouTube

Club TWiT: AI User Group #17 - Models, Memory & Maestro

2026-08-08 · 1:33:50

Club TWiT AI User Group #17 dove into models, memory, and Maestro as Leo Laporte led a roundtable of hands-on AI experimentation. Darren Oakey and the crew swapped stories and demos covering workflow automation, agent armies, AI-powered news pipelines, and the practical limits of current subscriptions. The group explored real-world successes, coding mishaps, and hacks pushing today's generative tools into daily production.

Watch on YouTube

Club TWiT: AI User Group #16 - Building Better Agents

2026-07-04 · 2:03:36

Club TWiT AI User Group #16 brought together host Leo Laporte and a panel of passionate tech explorers — including Bill, Darren Oakey, Craig McFarlane, Larry Gold, Juan Hernandez, and Anthony Nielsen — for a fast-moving dive into building better AI agents. The group dug into hands-on demos of Hermes, Claude, and DeepSeek alongside search platforms like Pulse and Serper, while debating cost strategies, local model deployment, and DevOps best practices. The conversation ranged from concrete hardware specs and pressing privacy concerns to the deeper questions of creativity, agency, and consciousness, culminating in a spirited look at the coming wave of agentic AI and the existential future of intelligent machines.

Watch on YouTube

Club TWiT: AI User Group #15 - Starting Smart with AI Agents - Hermes Agent, Claude Code, and More

2026-06-06 · 1:35:53

Club TWiT AI User Group #15 dug into the practical realities of running personal AI agents, walking through what it takes to get Hermes Agent up and running alongside tools like Claude Code and how the setup measures up against OpenClaw. Members traded real-world use cases spanning daily workflows, memory handling, and security concerns, alongside their favorite tools and practical hardware recommendations. The group also shared hard-won lessons from helping non-tech-savvy teammates connect the dots, making the session a grounded look at what local and cloud AI agents can actually accomplish in everyday life.

Watch on YouTube

Club TWiT: AI User Group #14 - Building Personal Agents & Local AI

2026-05-09 · 1:36:02

Club TWiT AI User Group #14 gathered host Leo Laporte with Larry Gold, Darren Oakey, Bill, Juan, and Alakazip for a deep dive into building personal agents and local AI, covering practical agent setups, ambient computing, home automation, and hardware choices. The roundtable saw members swap ideas on agent memory, rapid app development, and smart-device orchestration using Claude, Gemma, and Hermes, delivering real-world hacks, hardware insights, and honest lessons direct from the club's top tinkerers.

Watch on YouTube

Club TWiT: AI User Group #13 - OpenClaw, Agents, & Building Smarter AI Workflows

2026-04-13 · 1:11:41

Club TWiT AI User Group #13 explores OpenClaw, agents, and building smarter AI workflows with expert Larry demonstrating a homegrown agent framework he created after discovering security flaws in OpenClaw, revealing why Claude 4.6 is central to the agentic boom. The episode dives into Model Context Protocol (MCP) servers—their authentication and integration benefits balanced against context window consumption—and discusses when MCPs, skills, and direct API integration work best for real-world applications like Fastmail calendar automation and transcript processing. Viewers will learn how to build secure, practical agent implementations, understand the tooling tradeoffs shaping the agent ecosystem, and see how community builders like Aeronova are leveraging these systems for sophisticated projects.

Watch on YouTube

Club TWiT: AI User Group #12 - Opus 4.6 & OpenClaw Cold Feet

2026-03-07 · 1:24:31

In this Club TWiT AI User Group episode, the hosts showcase how they're using Claude Opus to automate repetitive technical tasks, with a detailed example of building Pixu Control—a custom daemon that intelligently manages LED display configurations without requiring advanced coding skills. The panelists explore programming language choices (Rust, Python, and Go) for AI-assisted development and reveal how Claude has fundamentally shifted their automation philosophy: rather than waiting to repeat a task three times before automating it, they now deploy AI to handle tasks they simply don't want to repeat even once. Viewers will learn practical strategies for rapid prototyping automation tools with Claude, discover the performance and library ecosystem tradeoffs between programming languages, and understand how modern developers are reimagining the threshold for when automation becomes worthwhile.

Watch on YouTube

Club TWiT: AI User Group #11 - AI Slop or Super Tool?

2026-01-31 · 1:13:37

Club TWiT members demonstrate how Claude AI serves as a practical assistant for technical tasks, with hosts showcasing real-world applications like configuring Linux tiling window managers with custom weather and calendar widgets, and automating the tedious app store submission process. The discussion reveals Claude's versatility beyond creative applications—spanning system administration, DevOps, configuration file management, and automating repetitive workflows that developers typically avoid. Viewers will learn how to leverage Claude strategically as a collaborative tool to handle complex technical challenges and time-consuming administrative work, allowing them to focus on the aspects of their projects they actually enjoy.

Watch on YouTube

Club TWiT: AI User Group #10 - Google Antigravity Crushes App-Building

2026-01-04 · 1:14:05

Anthony and Club TWiT members Larry and Darren explore Google's Antigravity, a newly launched VS Code-based agentic development platform built with Gemini 3 that directly competes with AI coding tools like Cursor and Windsurf. Larry delivers a compelling live demo where Antigravity builds a full Polymarket prediction market application from a single detailed prompt, revealing the platform's standout features including automatic task generation, implementation planning, and powerful visualization and image generation capabilities. Viewers will learn how Antigravity's multi-agent system enables simultaneous development across multiple applications and projects, and understand why Google's unique integration of Chrome browser functionality and media generation currently gives it an edge over competing AI development platforms.

Watch on YouTube

Club TWiT: AI User Group #9 - DIY MCP Server With FastMCP

2025-12-06 · 48:21

In this episode of Club TWiT's AI User Group, Darren Oakley, a regular AI super user, walks through building a custom MCP (Model Context Protocol) server from scratch using FastMCP, a Python library that simplifies creating programs that communicate with Claude via JSON protocol. The demonstration focuses on creating a practical task manager MCP server with essential functions like adding, listing, and completing tasks, showing how simple it is to extend Claude's capabilities with custom tools. Viewers will learn how MCP servers work locally on their machines, how to use attributes to expose functions as tools, and gain hands-on insight into rapid AI application development using Claude's code generation abilities.

Watch on YouTube

Club TWiT: AI User Group #8 - Framework Desktop & Local AI

2025-11-01 · 1:07:39

In this Club TWiT AI User Group episode, host Leo Laporte joins participants from around the globe (Darren from Australia, Larry from New Jersey, Juan from San Diego, and Allan from San Francisco) to explore the newly released Framework Desktop computer equipped with an AMD Ryzen AI Plus 395 processor and 128GB of unified RAM for running local AI models. The group demonstrates how to use LM Studio—a user-friendly platform for downloading and running large language models locally—and discusses practical performance metrics like token generation rates (targeting 40+ tokens per second) and which models work best for different tasks like code generation. Viewers will learn how to set up a high-performance local AI workstation, run open-source models like Qwen and Mistral locally, integrate them with development tools like VS Code through OpenAI-compatible APIs, and understand the security and performance advantages of running AI inference on local hardware.

Watch on YouTube

Club TWiT: AI User Group #7 - n8n Automation Deep Dive

2025-10-04 · 2:00:48

Club TWiT's AI User Group episode #7 features host Leo, participants Larry and Anthony Nielson, and a deep dive into n8n, a powerful visual no-code automation platform that demonstrates how nodal programming and AI integration are fundamentally reshaping how people approach productivity and workflow creation. The team demonstrates practical deployment options for n8n, from self-hosted VPS solutions (starting around $10/month with providers like Hostinger) and local Docker containerization to the platform's native execution model, while exploring how to connect LLM models directly into automated workflows. Viewers will learn how visual node-based systems—inspired by educational programming tools like Carnegie Mellon's Alice and Apple Playgrounds—are democratizing automation by enabling non-programmers to build sophisticated, AI-powered workflows without writing a single line of code.

Watch on YouTube

Club TWiT: AI User Group #6 - Vibe Coding Mishaps

2025-08-30 · 1:40:17

The host experiments with Claude Code to reorganize their Obsidian knowledge base using the PARA method from a Reddit prompt, completing the ambitious task in under five minutes but discovering dataview errors and misaligned organizational changes that require reverting to backup. Viewers learn the critical importance of disabling sync and creating backups before letting AI tools autonomously restructure personal systems, along with practical insights into how the host structures Obsidian with hierarchical daily notes organized by year, month, and day. The episode demonstrates both the potential and perils of "vibe coding"—trusting AI to interpret loose prompts and make significant changes—highlighting how careful planning, backup management, and sync controls can prevent disaster when AI reorganization goes awry.

Watch on YouTube

Club TWiT: AI User Group #5 - Vibe Coding to the App Store

2025-08-04 · 1:35:47

In this Club TWiT AI User Group episode, Bill, Monty, Darren, and Joe explore "vibe coding"—using AI assistants like Windsurf and Claude Code to build real applications from concept to the App Store. Participants share hands-on experiences creating projects including a TWiT mobile app and an iOS reminders app using SwiftUI and Expo/React Native, demonstrating how developers with different programming backgrounds leverage AI effectively. Viewers learn essential strategies for successful AI-assisted development, including the critical importance of providing rich context to AI tools, using planning modes for organized code generation, and practical techniques like screenshot error reporting and auto-execution features that accelerate development cycles.

Watch on YouTube

Club TWiT: AI User Group #4 -Vibe Coding

2025-07-05 · 1:50:54

In this Club TWiT episode, participants explore "vibe coding"—using AI tools like Cursor, Copilot, and Claude's Sonnet 4 to assist with programming—while tackling the practical challenges of API costs, token optimization through diffs, and the limitations of tools built for coding rather than critical thinking. Host and guest Dr. Doo, a multi-language coder transitioning to Python, share hard-won lessons including discovering that Sonnet 4's power consumed $29 per day in API credits and that most AI coding tools struggle with outputting usable raw diffs across multiple files. Viewers will learn practical strategies for managing AI-assisted coding workflows, cost optimization techniques, and how to integrate AI capabilities into Obsidian for seamless personal knowledge management and creative project organization.

Watch on YouTube

Club TWiT: AI User Group #3 - Leo's Sci-Fi Makeover

2025-05-24 · 1:06:17

In Club TWiT's AI User Group episode, a skilled AI content creator demonstrates their sci-fi makeover project to Leo, showcasing how they used Sora to transform still images into dynamic AI-generated videos with futuristic aesthetics. The discussion reveals Sora's unique advantages over tools like Midjourney and ChatGPT—including superior prompt adherence, ability to incorporate reference images, and pixel-by-pixel repainting rather than simple inpainting—while exploring the highly iterative nature of effective prompt engineering. Viewers will learn practical techniques for using multiple AI tools in tandem, discover that starting fresh is often more efficient than endlessly refining prompts, and gain insight into how modern AI image generators understand and execute complex visual concepts.

Watch on YouTube