Two Voice Devs

In Part 1 of a two-part series, guest host Sam Witteveen joins Allen to catch up and dive deep into the rapidly evolving world of AI models. Sam shares his fascinating journey from being a successful pop songwriter to becoming a Machine Learning Google Developer Expert (GDE) and running the massive Machine Learning Singapore meetup.

The conversation shifts to the latest AI developments, exploring the "Big Three" model builders—Anthropic, OpenAI, and Google. Sam and Allen discuss the frenetic pace of new model releases, changes to the Gemini 3 API, and how developers navigate the trade-offs between intelligence, latency, and cost.

Finally, they pull back the curtain on how these models are actually trained today. Discover why models are no longer trying to be "fact machines" and how post-training breakthroughs, code execution sandboxes, and Reinforcement Learning (RL) environments are dramatically improving AI capabilities. Stay tuned for the end of the episode, where they hint at what's coming in Part 2!

Timestamps:

[00:00:00] Introduction and catching up

[00:01:33] Sam's fascinating journey from pop music to machine learning

[00:05:23] Running the massive Machine Learning Singapore meetup

[00:07:42] Stumbling into YouTube and teaching AI with Google Colab

[00:12:38] Analyzing the "Big Three" AI models and rapid release cycles

[00:17:52] Gemini 3 API updates, Flash models, and thinking levels

[00:22:00] Tool use, knowledge cutoffs, and why LLMs aren't fact machines

[00:26:00] How post-training and code sandboxes revolutionized AI

[00:32:00] Scaling Reinforcement Learning (RL) environments for design

[00:34:04] Structured outputs and the return to predictable rules

[00:36:43] Tune in next time for more! And where to find Sam online

Hashtags:

#TwoVoiceDevs #AI #MachineLearning #DeepLearning #LLM #GoogleGemini #Gemini #OpenAI #ChatGPT #Anthropic #Claude #ReinforcementLearning #RAG #Developers #SamWitteveen

Allen has been busy! This week, he unveils the new `@langchain/google` package for LangChain JS. This major update consolidates five previous libraries into a single, standardized, and powerful tool for developers working with Gemini and Vertex AI. Allen walks Mark through the motivation behind the change, the focus on backward compatibility, and the exciting new features like simplified multimodal input/output and text-to-speech support. If you're building with Google AI and JavaScript, this is the update you've been waiting for.

[00:00:57] The confusion of previous packages

[00:02:52] Creating a unified package

[00:03:45] Introducing @langchain/google

[00:04:35] Backward compatibility

[00:06:48] Multimodal inputs

[00:07:54] Standardizing output and image generation

[00:08:58] Text-to-Speech support

[00:11:29] Simplifying parameters and reasoning

[00:14:55] Future roadmap

#LangChain #Gemini #NanoBanana #TextToSpeech #GoogleAI #JavaScript #TypeScript #VertexAI #OpenSource #AI #WebDevelopment #TwoVoiceDevs

Ever wonder how "Two Voice Devs" goes from a raw recording to a finished episode? In this episode, Allen Firstenberg takes Mark Tucker on a deep dive into his production workflow. They discuss how Descript’s text-based editing revolutionized their process, how Allen uses a custom Gemini CLI agent to automate show notes and descriptions, and the technical (and ethical!) journey of creating AI-generated thumbnails using Google's Nano Banana. It’s a candid look at how AI can act as a force multiplier for creators while keeping the "human in the loop."

[00:00:01] Introduction and Check-in

[00:01:27] Behind the Scenes: Why We Use AI

[00:03:42] Descript: Text-Based Video Editing

[00:05:24] Building a Knowledge Database from Transcripts

[00:08:13] Editing Video Like a Document

[00:12:34] Exploring Descript's AI

[00:13:36] Automating Show Notes with Gemini CLI

[00:14:10] The Power of System Instructions (GEMINI.md)

[00:19:30] AI Thumbnail Generation with Nano Banana

[00:26:10] The Ethics of Synthetic Media and Artistic Style

[00:28:40] Keeping the Human in the Loop

[00:33:00] Evolution of the Two Voice Devs Workflow

#TwoVoiceDevs #PodcastProduction #GeminiAI #Descript #GeminiCLI #NanoBanana #Automation #ContentCreation #Ethics #GenerativeAI #AIWorkflow #PodcastEditing

Mark and Allen dive into the emerging world of Agent Skills, an open standard for extending the capabilities of AI coding assistants like GitHub Copilot, Claude Code, and Gemini CLI. They explore how these skills work, how they compare to the Model Context Protocol (MCP), and walk through creating and installing a custom skill using the `skills` CLI. They also discuss the skills.sh website by Vercel, which acts as a registry and leaderboard for the ecosystem. The conversation touches on the potential for standardization, the current fragmentation in the ecosystem, and critical security considerations for these powerful new tools.

More Info:

* https://agentskills.io

* https://skills.sh

* https://cra.mr/mcp-skills-and-agents

[00:00:00] Introduction & Context: AI Agents and Tools

[00:02:18] Getting Information into Context (Instructions files)

[00:06:50] What are Agent Skills? (AgentSkills.io)

[00:09:55] Agent Skills vs. MCP Servers

[00:16:35] How Skills Work: Progressive Disclosure

[00:19:50] Mark's Example: List Global NPM Skill

[00:22:56] Installing Skills with skill.sh and the Skills CLI

[00:26:55] Demo: Installing on GitHub Copilot

[00:30:58] Demo: Installing on Gemini CLI

[00:37:37] Discussion: Discovery, Standardization, and Security

[00:43:05] Conclusion

#AgentSkills #AI #GitHubCopilot #GeminiCLI #CodingAssistants #MCP #ModelContextProtocol #DeveloperTools #TwoVoiceDevs

Allen and Mike discuss Google's new "Personal Intelligence" feature for Gemini. They explore how it connects to your personal data like Photos, Gmail, and Docs to provide context-aware answers. The conversation covers real-world use cases, privacy concerns regarding training data, and the importance of transparency and granular control in AI systems. They also touch on the "blackmail" scenario found in other AI research and what developers can learn from Google's implementation.

More Info:

* https://blog.google/innovation-and-ai/products/gemini-app/personal-intelligence/

[00:00:30] Google's Gemini Personal Intelligence announcement

[00:01:48] Connecting personal data sources to Gemini

[00:03:45] Google's unique advantage with user data

[00:06:40] Real-world use case: Tracking travel history

[00:07:30] Potential use case: Combining health data sources

[00:09:15] Privacy: Is your data used for training?

[00:12:40] The debate: Opting in vs. privacy concerns

[00:16:30] AI safety and the "blackmail" scenario

[00:18:50] Lessons for developers: Granular permissions and transparency

[00:20:30] Verifiability and user trust

[00:23:50] Conclusion

#Gemini #GoogleAI #PersonalIntelligence #Privacy #MachineLearning #Developer #TechPodcast #AI #TwoVoiceDevs

Allen Firstenberg welcomes back guest host Mike Wolfson, an Android Google Developer Expert, to discuss the shifting landscape of User Experience (UX) in the age of Artificial Intelligence. As we move toward autonomous agents and multimodal interactions—incorporating voice, haptics, and environmental data—the reliance on traditional screens and touch interfaces is set to diminish.

Mike shares insights on why "context" is the new "queen," illustrating the challenges of current AI with real-world examples (like his Meta glasses failing to find a decent breakfast burrito). The conversation tackles the critical question: "Who does the agent truly work for?" and explores how developers can avoid "notification fatigue" while ensuring users remain informed and in control. From the "creepy" factor of hyper-personalized data to the "Beverage Butler" concept, this episode dives deep into designing for a future where the best UI might be no UI at all.

[00:00:00] Welcome and Introduction

[00:01:21] The Evolution of UX: Beyond Touchscreens

[00:03:52] The Importance of Context: A Meta Glasses Example

[00:06:21] Privacy, Creepiness, and Agent Loyalty

[00:08:43] Application Development in an AI World

[00:11:34] Avoiding the "Seinfeld Assistant" of Notification Fatigue

[00:17:15] Feedback Modalities: Tones vs. Speech

[00:19:39] Discoverability of Features in Non-Visual Interfaces

[00:23:00] The "Beverage Butler" and Future Outlook

#AI #UX #UserExperience #ArtificialIntelligence #AndroidDev #ContextAwareness #VoiceUI #NoUI #TechPodcast #TwoVoiceDevs #GDE #GoogleDeveloperExpert #BeverageButler

In this episode, Mark introduces Allen to Parlant, an open-source framework for building agentic AI. They explore how Parlant differs from other frameworks like LangChain and LangGraph by using concepts like "journeys" for flexible conversation flows and "guidelines" for conditional rule application. Mark walks through the key features, including the ability to define glossaries, tools, and the engine's matching process. They also discuss the recent version 3.1 updates, such as linked journeys and behavior criticality levels. Finally, they dive into "Emcie," Parlant's managed NLP service that utilizes a teacher-student model architecture to optimize performance and cost using Small Language Models (SLMs).

[00:00:00] Welcome and Introduction

[00:00:43] Introduction to Parlant

[00:02:03] Journeys: Flexible Conversation Flows

[00:03:36] Guidelines: Conditional Rules

[00:06:27] Motivations and Compliance

[00:08:44] NLP Services and Providers

[00:11:42] The Balance Between Rigid and Loose Conversations

[00:18:43] Parlant 3.1 Updates: Linked Journeys and Behavior Levels

[00:22:20] Custom Matchers

[00:23:14] Emcie: Parlant's Managed NLP Service

[00:25:05] Model Tiers: Jackal and Bison

[00:26:14] Teacher-Student Architecture and SLMs

[00:30:26] Cost and Optimization with Student Models

[00:35:19] Conclusion and Wrap-up

#Parlant #AI #AgenticAI #LLM #SLM #OpenSource #SoftwareDevelopment #TwoVoiceDevs #Emcie #NLP

Happy New Year! Allen and Mark kick off 2026 by looking back at the whirlwind of AI developments in 2025. From the explosion of agentic frameworks like LangGraph, Google's Agent Development Kit, and the Microsoft Agent Framework to the emergence of protocols like MCP and A2A, it was a year of rapid evolution. They discuss the rise of "vibe coding," the state of voice assistants like Alexa Plus and Gemini, and the challenges of monetization and discovery in a model-centric world.

What lies ahead in 2026? The duo shares their big predictions! What do we need on the hardware side? Perhaps something new from Google, Microsoft, OpenAI, Amazon, or someone else? What companies are we rooting for? What's next for agents? Tune in and find out!

Timestamps:

[00:00:00] Introduction and Happy New Year

[00:01:13] Reflecting on a busy 2025: Google's weekly announcements

[00:02:05] New terms: MCP (Model Context Protocol) and A2A (Agent-to-Agent)

[00:03:52] The shift to Agents and Agentic solutions

[00:05:40] Framework evolution: Microsoft Agent Framework and LangGraph

[00:07:54] The rise of Coding Assistants and "Vibe Coding"

[00:09:25] State of Voice: Alexa Plus and Gemini on smart devices

[00:11:15] Smart Glasses and the future of ambient AI

[00:12:40] MCP: Challenges with discovery, monetization, and security

[00:15:10] Microsoft Foundry and low-code agent building

[00:20:25] Live Streaming Models: Video, Audio, and Text

[00:22:00] 2026 Predictions: Voice Flow acquisition

[00:23:05] Prediction: Moving from Chatbots to Autonomous Agents

[00:25:20] Prediction: The role of Small Language Models (SLMs)

[00:27:10] Closing thoughts and 2026 outlook

Hashtags:

#AI #GenerativeAI #Agents #AutonomousAgents #MCP #A2A #LangGraph #Gemini #VoiceAssistant #SmartGlasses #VoiceFirst #SLM #VoiceFlow #TwoVoiceDevs #YearInReview #2026Predictions #VibeCoding

Get ready for the ULTIMATE SHOWDOWN of holiday cheer and artificial intelligence! In this SPECIAL HOLIDAY EPISODE, Mark and Allen aren't just exchanging pleasantries—they're exchanging MIND-BENDING REBUS PUZZLES generated by some AI models themselves!

It's a battle of wits, a clash of code, and a festive face-off as Microsoft Copilot takes on Google's Gemini (and the famous "Nano Banana Pro") to solve visual riddles that will have you shouting at your screen. Can our hosts decipher the scribbles of silicon brains? Or will the AI stump the humans once and for all?

Grab your eggnog, put on your thinking cap, and play along! It's Two Voice Devs like you've never seen (or puzzled) them before! HAPPY HOLIDAYS!

[00:00:00] Intro: The Rules of Engagement

[00:02:48] Puzzle 1: A Nipping Chill

[00:03:33] Puzzle 2: Going for Gold

[00:07:00] Puzzle 3: Escaping the Cage

[00:10:00] Puzzle 4: The Silent Mouse

[00:11:00] Puzzle 5: A Knightly Gesture

[00:12:15] Puzzle 6: Sweet Ballerina

[00:13:30] Puzzle 7: Listen Closely to the Animal

[00:15:30] Puzzle 8: A Holiday Wish

[00:16:15] Puzzle 9: The Grand Finale Challenge

[00:19:30] Happy Holidays from Two Voice Devs!

#TwoVoiceDevs #HolidaySpecial #RebusPuzzles #LLMBattle #AIShowdown #Copilot #Gemini #NanoBananaPro #ChatGPT #ArtificialIntelligence #MachineLearning #JackFrost #WinterWonderland #Freeze #SugarPlumFairy #Nutcracker #NewYears #Gnu #MerryChristmas #FestivalOfLights #Hanukkah #Kwanzaa #Unity #HolidayFun #Games #Puzzles #TechHumor #DevLife #HappyHolidays #SeasonsGreetings #Fun #Creative #OverTheTop #Podcast #Developers #SoftwareEngineering

Mark Tucker explores Microsoft Azure's "Voice Live" feature within the newly rebranded Microsoft Foundry. He demonstrates how to take a standard text-based AI agent—in this case, one that talks like a pirate—and instantly give it a voice using WebSockets to bridge speech-to-text and text-to-speech. Mark walks through the differences between the "Old" Foundry (V1) and the "New" Foundry (V2), shows the configuration steps, and dives into a Python code example to connect it all together.

Learn more:

* https://github.com/rmtuckerphx/voicelive-agents-quickstart

[00:00:00] Intro and Holiday Plans

[00:00:46] Introducing Microsoft Azure Voice Live

[00:02:00] Microsoft Foundry Overview

[00:02:46] Creating a Pirate Agent in Foundry

[00:04:46] Enabling Voice Live in the Playground

[00:07:46] Demo: Speaking with the Pirate Agent

[00:08:46] Comparing Old and New Foundry

[00:13:46] Code Walkthrough: Voice Live Quick Start

[00:15:46] Connecting Version 2 Agents

[00:18:46] Conclusion

#MicrosoftAzure #VoiceLive #MicrosoftFoundry #AI #VoiceFirst #GenerativeAI #SpeechToText #TextToSpeech #Python #Coding