Advertise on podcast: vBrownBag

Join us as Cobus Bernard takes us on a live tour of his sprawling homelab - and explains why Claude wrote most of the Terraform and Ansible running it. Cobus walks through a Proxmox setup running Traefik, Vault, Nomad, Keycloak, and a three-node Kubernetes cluster that doesn't quite work yet, all built by handing infrastructure-as-code tasks to Claude and slowly deciding how much of it to trust with real commands. You will learn how Cobus structures Terraform state and variables so an AI agent can safely extend it, why he still won't let Claude run terraform apply unsupervised, how MCP servers control what an agent is allowed to touch, and why confabulation might be a better word than hallucination for what these models actually do. Timestamps 0:00 Welcome & Introduction 8:50 What's Running in the Home Lab - Proxmox, Traefik, and Vault 13:48 A Kubernetes Cluster That Doesn't Work Yet 17:09 MCP Servers and What Claude Is Allowed to Run 21:46 Is It That Claude Wrote the Terraform? 23:06 Trusting Claude with Terraform Apply 27:43 The Home Assistant Confession - Copy-Paste vs. Actually Learning 36:57 Naming Things - Solving the Classic Kubernetes Problem 43:52 Confabulation - A Better Word for AI Hallucination 48:49 Building a Tax Calculator App With Claude How to find Cobus: https://www.linkedin.com/in/cobusbernard/ Links from the show:

Join us as Faye Ellis breaks down what responsible AI actually looks like in practice on Amazon Bedrock - live demos included, straight from her prep for the AWS Generative AI Developer Professional certification. Faye walks through the NIST AI risk framework's core characteristics of trustworthy systems, then demos two hands-on builds: a RAG pipeline using S3 Vectors as a low-cost knowledge base, and Bedrock Guardrails blocking financial advice, prompt injection attacks, and sensitive information in real time. You'll learn the difference between prompt engineering and guardrails, how contextual grounding checks catch hallucinations before they reach a user, why data cleaning and deduplication matter as much for cost and sustainability as for bias, and what Faye wishes she'd known going into one of AWS's hardest professional-level exams. Timestamps 0:00 Welcome & Introduction 9:28 What Makes an AI System Trustworthy - The NIST Framework 15:19 Famous AI Failures - From ChatGPT Code Leaks to Biased Systems 22:56 Building RAG with Bedrock Knowledge Bases and S3 Vectors 26:29 Live Demo - Setting Up RAG with S3 Vectors 31:19 Prompt Engineering and Bedrock Prompt Management 34:46 What Are Bedrock Guardrails and How They Work 41:22 Live Demo - Blocking Financial Advice, Jailbreaks, and PII 53:00 Preparing for the AWS Generative AI Developer Professional Exam 59:04 Resources and Wrapping Up How to find Faye: https://www.linkedin.com/in/fayeellis/ https://app.pluralsight.com/profile/author/faye-ellis Links from the show: https://aws.amazon.com/bedrock/ https://skillbuilder.aws/ https://aws.amazon.com/certification/

Join us as John Mark Troyer and Rakesh Gupta break down what AI observability actually means once agents leave the demo and hit production - and why the old playbook for monitoring doesn't cut it anymore. John Mark and Rakesh walk through why errors and latency are just the starting point for agents, how quality became a much harder thing to measure once bots went from answering questions to taking autonomous action, and why token-based costs are creating a confusing new economics problem for engineering teams. You'll learn the difference between online and offline evals, why a new engineering role has emerged just to build testing harnesses for agents, how trace data works differently when every prompt is its own trace, and what teams are doing to catch prompt injection and other AI-specific failure modes before they become expensive mistakes. Timestamps 0:00 Welcome & Introduction 3:20 Full Disclosure - Observe, Snowflake, and How This Conversation Started 7:07 From Developer Concerns to Boss's Boss's Boss - Spending Out of Control 8:29 What Actually Gets Measured - Errors, Latency, Quality, and Cost 10:30 The Casino Chip Problem - Confusing Token Pricing Models 13:47 Defining Quality When the Task Itself Is Nebulous 18:41 The New Role - Engineers Who Just Build Testing Harnesses 22:00 Non-Determinism and Why Testing Agents Is Expensive 32:10 Trace Data, Tool Calls, and What Observability Tools Actually See 55:08 Prompt Injection, Zero-Width Characters, and Real World Failures How to find John Mark: https://www.linkedin.com/in/johnmarktroyer/ How to find Rakesh: https://www.linkedin.com/in/rg0/ Links from the show:

Join us as Bob Belderbos breaks down how to actually learn new skills and languages in a world where AI can write the code for you before you've even finished the thought. Bob shares why he taught himself Rust the hard way, how keeping deliberate friction in your learning process protects you from skill atrophy, and why AI is incredible at explaining concepts but dangerous as a crutch for understanding them. You'll learn the difference between using AI to explain versus using it to do, how to structure a project-based learning path with tests as your guide, why coding autocomplete might be quietly hollowing out your skills, and how his Python and Rust cohorts are teaching professional engineers to use agents without losing ownership of their code. Timestamps 0:00 Welcome & Introduction 1:09 Bob's Background - From VBA to Python to Rust 4:39 Why Learn Rust When Python Already Works 11:43 AI as a Learning Assistant vs. a Socratic Teacher 12:30 The Slot Machine Problem - Agents and Skill Atrophy 17:24 Working Outside Your Expertise - The Fast LED Story 27:02 Structuring Prompts That Actually Teach You Something 33:56 Teaching Agentic AI in Production - The Expense Classifier Cohort 36:07 Autocomplete, Copilot, and the Line Between Helping and Hollowing Out 44:03 AI Slop, Coauthorship, and the Anti-Slop Engineer 53:49 What's Next - Rust, Haskell, and Bob's Upcoming Cohorts How to find Bob: https://www.linkedin.com/in/bbelderbos/ https://belderbos.dev/ Links from the show:

Join us as Du'An digs into the real mechanics of running AI locally and in production - from GPU memory math to multi-agent architectures, observability, and the economics of self-hosted inference. Du'An walks through how model weights and KV cache compete for GPU memory, why continuous batching matters when you have more than a handful of users, and how agent architectures like single-agent, workflow, graph, swarm, and supervisor patterns each solve different problems. You will learn how to instrument your agents with Langfuse for observability and cost tracking, when to use Ollama versus vLLM, how prompt caching can cut provider costs by up to 75%, and why GPUs should never sit idle. Episode two of three - the next episode covers deploying at scale. Timestamps 0:00 Welcome & Introduction 1:47 Du'An's New Role at Akamai Cloud 3:10 Data Privacy and the Case for Self-Hosted AI 7:21 Anthropic and OpenAI as the New Cloud Layer 12:48 Local Models for Specific Use Cases - Cancer Detection Example 15:02 GPU Memory Math - Weights, KV Cache, and Context Windows 19:32 Continuous Batching and GPU Time Slicing 20:03 Observability with Langfuse - Live Demo 27:44 Agent Architectures - Single Agent, Workflow, Graph, Swarm, Supervisor 36:36 Token Economics, Prompt Caching, and GPU Cost Planning 45:32 Ollama vs vLLM - Prototyping vs Production How to find Du'An: https://duanlightfoot.com https://www.linkedin.com/in/duanlightfoot/ Links from the show: https://langfuse.com/ https://github.com/akamai-developers/akamai-workshop-solution-architect-agent https://amzn.to/4bvHn1p https://vllm.ai/

Join us as Thorsten breaks down how everyday people - small business owners, retirees, hobbyists, and anyone who isn't a developer - can use AI to get real things done without writing a single line of code. Thorsten runs an AI roundtable for small and medium businesses in Germany and shares hands-on use cases from his own life and clients: a family recipe database built entirely on a mobile phone, 13 months of fitness data analyzed into actionable coaching, a personal AI project manager that reads his calendar and meeting transcripts, and real-world implementations for a dental practice and a tax advisory firm. You will learn how to think about AI as a team you never had, why prompts are the new apps, how to handle privacy and data regulations, and how to start this week with just one task. Timestamps 0:00 Welcome & Introduction 5:17 Who Is Thorsten and What Is Normal People AI 8:51 The Recipe Database Use Case 16:46 Fitness Tracking and Personal Coaching 22:27 Building an AI-Powered Personal Project Manager 34:23 Using Fireflies and MCP for Meeting Intelligence 41:14 Real Business Use Cases - Dental Practice and Tax Advisory 44:27 Privacy, GDPR, and When to Use Local Models 47:06 Getting Started - One Task This Week How to find Thorsten: https://www.linkedin.com/in/hoegertn/ Links from the show:

Join us as Dave walks through what it actually takes to build custom AI agents from scratch - not theory, but real projects he has shipped for his family, his work, and his community. Dave shares how he used Kiro and Claude to solve real problems: normalizing flood-damaged library inventory data, automating AWS well-architected review collateral, building a room-cleaning task agent for his 12-year-old, planning family menus with Apple Calendar integration, and post-processing live concert recordings. You will learn how agents reason and take action, when to reach for a Kiro power versus a simpler automation, how MCP servers connect agents to real-world tools, and practical strategies for keeping agents accurate without burning through tokens. Timestamps 0:00 Welcome & Introduction 7:57 Dave's Background and How He Got Started with Agents 13:00 The Library Flood Story - First Real-World Agent Use Case 16:00 AWS Well-Architected Review Automation 17:09 What Are Kiro Powers and MCP Servers? 22:13 Kiro Pricing and Bedrock Integration 28:13 Live Demo - Room Cleaning Agent with AWS Rekognition 41:24 Family Meal Planning and Apple Calendar Integration 44:27 Automating Live Concert Recording Post-Processing 52:31 Getting Started - Dave's Recommendations for Beginners How to find Dave: https://www.linkedin.com/in/dave-stauffacher/ Links from the show: https://kiro.dev/

Join us as Dale Orders (AWS Community Builder, four-time All Builders Welcome participant from Australia) walks through everything you need to know about getting to AWS re:Invent completely free - flights, hotel, conference pass, and more. Dale shares her personal journey from being rejected the first time to attending four AWS conferences through the All Builders Welcome grant program, including two as a mentor. You'll learn the exact eligibility criteria, what the grant actually covers (flights, accommodation, Uber vouchers, a prepaid Visa card, and a free AWS exam voucher), how to write an application that stands out, and the one thing that will get yours rejected immediately. Dale also covers what happens after you're accepted, how to handle the visa process if you're outside the US, and a full list of other tech conference grant programs beyond AWS. Applications typically open in late June - this episode is your head start. Timestamps 0:00 Welcome & Introduction 4:11 What is the All Builders Welcome Program? 8:03 Dale's Journey: Rejected Once, Accepted Four Times 11:07 Eligibility Criteria & Who Should Apply 15:49 The #1 Thing That Will Get Your Application Rejected 16:03 Everything the Grant Actually Covers 17:42 How to Apply & Timeline 18:41 Writing a Winning Application 35:18 Visa Process Warning: Don't Ignore This 38:41 Other Tech Conference Grant Programs & Wrap-up How to find Dale: https://www.linkedin.com/in/dale-orders/ Links from the show:

Join us as Dale Orders (AWS Community Builder, four-time All Builders Welcome participant from Australia) walks through everything you need to know about getting to AWS re:Invent completely free - flights, hotel, conference pass, and more. Dale shares her personal journey from being rejected the first time to attending four AWS conferences through the All Builders Welcome grant program, including two as a mentor. You'll learn the exact eligibility criteria, what the grant actually covers (flights, accommodation, Uber vouchers, a prepaid Visa card, and a free AWS exam voucher), how to write an application that stands out, and the one thing that will get yours rejected immediately. Dale also covers what happens after you're accepted, how to handle the visa process if you're outside the US, and a full list of other tech conference grant programs beyond AWS. Applications typically open in late June - this episode is your head start. Timestamps 0:00 Welcome & Introduction 4:11 What is the All Builders Welcome Program? 8:03 Dale's Journey: Rejected Once, Accepted Four Times 11:07 Eligibility Criteria & Who Should Apply 15:49 The #1 Thing That Will Get Your Application Rejected 16:03 Everything the Grant Actually Covers 17:42 How to Apply & Timeline 18:41 Writing a Winning Application 35:18 Visa Process Warning: Don't Ignore This 38:41 Other Tech Conference Grant Programs & Wrap-up How to find Dale: https://www.linkedin.com/in/dale-orders/ Links from the show:

Join us as Mike Fiedler (AWS Hero, PyPI Safety & Security Engineer, Python Software Foundation) makes the case for eliminating long-lived credentials from your release workflow - before an attacker does it for you. Mike walks through the real-world incidents that motivated Trusted Publishing, how OIDC-based short-lived tokens work under the hood, and the step-by-step process for setting it up in GitHub Actions. You'll learn how the 2024 Ultralytics compromise was forensically investigated thanks to Sigstore attestations, why that API token in your repo is just a password with a fancy hat, common pitfalls that will have you debugging for four hours, and why deleting your old token after setup is the step everyone forgets. PyPI went from 10% Trusted Publishing adoption in February 2024 to 36% today - this episode is how you become part of that number. Timestamps 0:00 Welcome & Introduction 4:00 Mike's PyCon US World Tour Recap 8:00 The Scale of PyPI: 13B Requests/Day & 36% Adoption 12:09 Why Long-Lived Tokens Fail: Four Attack Models 16:47 Case Study: The 2024 Ultralytics Compromise 21:44 What is Trusted Publishing? OIDC Explained 27:04 How the GitHub Actions Flow Actually Works 34:12 Other Registries: npm, RubyGems, crates.io, NuGet 36:34 Common Pitfalls & Debugging Tips 42:29 Provenance & Sigstore Attestations 44:22 The Step Everyone Forgets: Delete Your Old Token 47:06 Migration Guide & Getting Started This Week How to find Mike: https://www.linkedin.com/in/miketheman/ https://www.python.org/psf-landing/ Links from the show:

AI subscriptions are becoming as essential as internet bills - and just as expensive. The vBrownBag gang takes a hard look at the real cost of LLMs and what happens when the free ride ends. Chris, Shala, and Damian dig into the Anthropic pricing plot twist, why AI data centers consume 10x the power of traditional racks, the DeepSeek distillation controversy, and what happens when the first hit's free phase ends. You'll learn practical strategies for reducing token burn, why local models are becoming a viable cost escape hatch, how to pick the right model for the right job, and why blindly using Opus for everything is lighting money on fire. This is the unfiltered conversation every AI practitioner needs to have - before the subsidies disappear and the real bills arrive. Timestamps 0:00 Cold Open: Get These Darn Kids Off My Lawn 1:27 Chris's Big News: Leaving IBM for Six Feet Up 8:09 How Many AI Subscriptions Do You Have? 16:41 Stack Overflow Is Dead, Long Live Claude 17:12 Don't Just Blindly Copy and Paste (AI Edition) 31:00 Anthropic Gross Margin 2025: Negative 53% 35:30 When Token Costs Exceed a Junior Dev's Salary 42:02 Find the Model That Fits the Job 46:11 AI Multitasking Is a Lie (Just Like Humans) 49:05 We Are Uniquely Bad at Making Money Off This Show 53:19 Supply Chain Attacks and GitHub Actions 54:45 Did We Solve Anything? Yes. No. Maybe. 55:58 Grateful for Friends & Wrapping Up Links from the show:

Join us as Brian Hough (CEO & Founder of Tech Stack Playbook, AWS Hero) gets brutally honest about the state of tech hiring and what skills developers actually need to survive - and thrive - in the AI era. Brian walks through his frontline perspective on why tech layoffs aren't about skills - they're about market economics - and what that means for engineers trying to stay relevant. You'll learn which roles are actually hot right now (ML engineer, AI engineer, cloud architect, full stack dev), why companies want utility players who can build end to end, how to use social media and building in public to get quietly hired, and why the engineers who thrive will be those who can go from vision to deployed system. Brian also covers practical strategies for positioning yourself before the next wave hits, including using roadmaps as a personal curriculum and leveraging AI as a career accelerator rather than a threat. Timestamps 0:00 Cold Open 0:11 Welcome & Introduction 2:16 Taking Vibe Code to Production-Grade Systems 3:01 Brian's Update: Dog Feeding & Building Internal Tools 8:05 Mac Maximus: Building on AWS EC2 Mac 9:49 Let's Get Into the Presentation 10:10 Agenda Overview 11:11 Is Anyone Actually Working Less Because of AI? 12:52 What Happens When You Don't Understand What You Built 20:10 AWS Root Account Horror Story 23:24 The Skills You Need in 2026 24:09 Tech Scene Overview & Job Posting Divergence 26:19 What Companies Actually Want: Utility Players 28:00 Hot Roles: ML Engineer, AI Engineer, Cloud Architect 32:00 The Layoff Reality: It's Market Economics, Not Skills 40:49 Now Is the Best Time to Start a Startup 42:31 Roles & Salaries Breakdown 43:55 This Advice Is for Everyone - Not Just Job Seekers 48:01 What's Getting Replaced vs. What's Irreplaceable 49:14 How to Become an Irreplaceable Engineer 52:42 Maximum Viable Product 53:02 Building in Public & Social Media Strategy 55:32 Positioning Yourself Before the Next Wave 56:19 Brian's Closing Thoughts 57:03 AI on Your Resume = Getting Hired Fast 58:12 Using Brian's 30-Day Plan as a Claude Curriculum 59:55 Platform Engineering Hot Take 1:03:05 Wrap-up & See You in Seattle How to find Brian: https://brianhhough.com/techstackplaybook Links from the show: https://roadmap.sh/python https://roadmap.sh/ai-engineer https://roadmap.sh/machine-learning https://roadmap.sh/ai-agents

Andrew Brown (ExamPro) joins the vBrownBag crew to talk Gen AI skills, bootcamps, and whether vibe coding has made "learning to code" irrelevant. 🤖 Spoiler: it hasn't. Andrew walks through his upcoming Cloud Code Camp, demos agentic workflows he's built using Claude Desktop and Claude Code, and breaks down how to actually think about coordinator patterns, MCPs, and token-efficient development. 💡 The crew also gets into the "coders are obsolete" debate, AI-assisted JRPG game development, and whether AI certifications are worth your time in 2026. 🎮 Chapters: 00:00:01 - Andrew Brown, Classified Backstories, and Gen AI Fatigue 00:08:45 - Cloud Code Camp: Who It's For and What You'll Actually Build 00:15:38 - Building MCPs and Agentic Workflows with Claude Desktop and Claude Code 00:32:22 - Vibe Coding, Junior Devs, and Why Technical Skills Still Matter 00:54:03 - Are AI Certifications Worth It in 2026? #GenAI #CloudComputing #AIAgents #ExamPro #vBrownBag

AWS Hero Stephen Sennett joins the vBrownBag crew to argue that foundational knowledge matters more now, not less. 🧠 They dig into why communication skills have become a functional technical requirement, 💬 how to build a T-shaped career that AI can't commoditize, and where LLMs still fall completely flat. 🤖 There's a live car wash prompt test 🚗, an honest AGI reality check, and a look at how junior and senior expectations are shifting in the cloud careers era. ☁️ Worth the watch for anyone mapping out their tech skills strategy for 2026. Chapters: 00:01:35 - Why Foundational Knowledge Is Your Advantage in the AI Era 00:12:50 - Communication Is Now a Technical Requirement (AI Proves It) 00:20:07 - T-Shaped Skills, Learning in Public, and Standing Out in Cloud Careers 00:34:46 - What AI Still Can't Do: The AGI Reality Check and Car Wash Test 00:49:20 - How AI Is Reshaping Junior, Mid, and Senior Dev Expectations #CloudComputing #AWS #TechCareers #AISkills #vBrownBag

Join us for Part 1 of a 3-part series as Du'An Lightfoot (Senior AI Engineer at Akamai) breaks down everything you need to know to get started running AI models locally on your own hardware. Du'An walks through the fundamentals of local AI - from understanding why you'd want to run models privately (data ownership, air-gapped environments, IP protection) to the hardware concepts that make it possible. You'll learn how inference actually works under the hood, why GPUs matter for AI workloads, how to choose and quantize models for your hardware, and how to get up and running with tools like Ollama. This is Part 1 of a 3-part series - future episodes cover serving models via API and distributing inference at the edge with Kubernetes. Timestamps 0:00 Cold Open: Why Local AI? 0:26 Welcome & Introduction 1:30 Following Up on the Frontier Models Episode 2:25 Du'An's Background & AI Inference at Akamai 3:49 What If You Wanted to Own Your Data? 5:00 Local AI vs Cloud AI: A Different Layer of the Stack 5:47 Why GPUs Matter: The Nvidia Story 7:03 CPU vs GPU: Serial vs Parallel Processing 8:28 Model Weights & Quantization Explained 12:45 Choosing the Right Model for Your Hardware 18:22 Getting Started with Ollama 24:16 Live Demo: Running Your First Local Model 30:41 Hardware Recommendations & Requirements 36:52 Hugging Face & Finding Models 42:18 Performance Tips & Benchmarking 48:35 Use Cases: When to Go Local vs Cloud 1:01:30 Live Demo: Claude Put-in-Work Repo 1:11:17 Bonus: Building a Deck with Co-work Live 1:16:36 Preview: Episodes 2 & 3 1:17:17 Wrap-up How to find Du'An: https://www.duanlightfoot.com/ https://github.com/labeveryday/ Links from the show: https://ollama.com/ https://apxml.com/ https://localllm.in/ https://huggingface.co/ https://github.com/labeveryday/claude-put-in-work https://claude.ai/