Toni Klopfenstein

I just quit my job at Google 😰 After 4 years, some dope projects, and even doper teammates — I'm going all in on my startup. Nervous beyond belief? Yes. Hyped beyond belief? Also yes. I'm building RevSet: GitHub for Jujutsu (JJ), the next-gen version control system. In plain English — the code review tooling Google engineers get to use internally, finally available to everyone else. If you've ever left Google and missed Critique, this one's for you. Send all the unsolicited advice you've got. I’ll need it. 👉 And most importantly: the waitlist link is in the first comment.

You can't make it better until you make it work (Akins's law 41 aka McBryan's law). I've wanted a 27" touch screen DAKboard for a while now, but at $800 I couldn't justify the cost. So for the last year, I just configured a custom screen to fit an old iPad I didn't use much and got some little sticker brackets to mount it to the wall. Then I bought the preloaded micro SD card for $25 that can be run with a Raspberry Pi, and got one of those from the Detkin Lab at Penn Engineering. I had an extra monitor sitting around, but it wasn't touchscreen, and I really wanted everyone in the family to be able to check things off their to do list on screen. After a lot of searching, I found a renewed 24" ViewSonic touchscreen on Amazon for $240. I finally got the chance to set it up today. I still have to mess with the screen layout and mount it to the wall, but the hard parts are done! This house will be functioning like a well oiled machine by the time my classes start this fall semester.

The Court needs reform. I support 18 year term limits with 2 justices appointed every presidential term. https://lnkd.in/g3U6YXyh

Visiting my old stomping grounds. SparkFun Electronics

Building Reusable Intelligence: Day 3 of the Google AI Agents Intensive with Kaggle As AI agents become more capable, a new challenge emerges: how do we prevent them from becoming overwhelmed by ever growing context while still enabling them to perform specialized tasks effectively? Day 3 of the AI agents Intensive with Kaggle focuses on Agent Skills — a powerful approach to making agents more modular, scalable and reusable. 💡 What You'll Learn Tomorrow's whitepaper, "Agent Skills," explores how to manage dynamic context and combat "context rot" by equipping agents with portable, reusable capabilities called Skills. Rather than loading every instruction, tool and workflow into a single system prompt, the framework introduces progressive disclosure, allowing agents to load execution details only when needed. At the center of this approach is a structured SKILL.md file, enabling agents to efficiently switch between specialized roles without overwhelming their context window. This architecture allows a single agent to flex into hundreds of expert roles while maintaining efficiency, consistency, and scalability. In tomorrow's codelabs, we'll: Explore how Skills work within Google Antigravity Learn how to create and manage agents using Agents CLI and ADK Use natural language prompts to create agents, lint code and test agent workflows Understand how reusable skills can accelerate agent development and maintenance 📚 Day 3 Content Unit 3 – Agent Skills 🎧 Summary Podcast: https://lnkd.in/d6YYe6cV 📄 Whitepaper: https://lnkd.in/d5WV9p8Y 🛠️ Codelabs Getting Started with Antigravity Skills by Romin Irani https://lnkd.in/dSbzuiXf Build Agents with Agents CLI & ADK by Toni Klopfenstein, Sita Lakshmi Sangameswaran https://lnkd.in/d9BNzA59 📋 Upcoming Livestream The next livestream will be held tomorrow at 11:00 AM PT / 8:00 PM CET / 11:30 PM IST. Hosted by Anant Nawalgaria and Smitha Kolan, the session will feature codelab author Polong Lin, alongside Google experts Debanshu Das, Gabriela Hernández Larios, Julia Wiesinger and Tanvi Singhal, who will share insights and discuss the assignments. 🎥 Livestream: https://lnkd.in/dXwK2i-H 📺 All livestream recordings and schedule: https://lnkd.in/dSujrvhH Looking forward to learning how modular skills can help build more scalable, maintainable and context-efficient AI agents. #GenAI #Google #Kaggle #ArtificialIntelligence #MachineLearning #AIAgents #AgentSkills #Antigravity #ADK #AgenticAI #VibeCoding #Innovation #AIEngineering

Just wrapped up the 5-Day AI Agents: Intensive Vibe Coding Course with #Google on #Kaggle. So far in my AI journey, this one stood out for its structured and practical approach, providing a clear picture of what it takes to move from prototype to production—covering not just tools and technologies, but also architecture, optimization, security, evaluation, and building an end-to-end AI ecosystem. The course combined daily podcasts, whitepapers, CodeLabs, and highly engaging live Q&A sessions, culminating in a capstone project. 🔹 Day 1 – Introduction to Agents & Vibe Coding Explored the shift from syntax to intent, the evolution from vibe coding to agent engineering, context engineering, harness engineering, AI-driven SDLC transformation, and the CapEx/OpEx economics of AI adoption. 🔹 Day 2 – Agent Tools & Interoperability Covered MCP, A2A, A2UI, AP2, and UCP, addressing the N×M integration problem, agent-to-agent communication, interoperability patterns, and secure identity-driven interactions. 🔹 Day 3 – Agent Skills Focused on reducing token costs, context rot, and complexity through Agent Skills, progressive disclosure, evaluation-driven development, DAG orchestration, skill packaging, and production deployment strategies. 🔹 Day 4 – Agent Security & Evaluation A deep dive into securing AI agents across Infrastructure, Data, Models, Runtime, IAM, Observability, and Governance. Topics included RAG poisoning, semantic attacks, LLM firewalls, SPIFFE identities, Zero Ambient Authority, Red-Blue-Green Teaming, and evaluation frameworks for intent-driven systems. 🔹 Day 5 – Spec-Driven Production Development Focused on specification-driven development, MCP integration, guardrails, sandboxing, human-in-the-loop controls, AI-driven testing, policy enforcement, context hygiene, and governance models for production-grade AI systems. 💻 CodeLabs Highlights Hands-on experience with Antigravity, Agents CLI, Google AI Studio, ADK 2.0, MCP integration, Agent Skills, Cloud Run deployments, LLM-as-a-Judge evaluations, STRIDE threat modeling, Semgrep security scans, and deploying AI agents on Google Cloud. The labs provided practical exposure across the entire lifecycle—from vibe coding and prototyping to secure, production-ready deployments. Special thanks to Anant Nawalgaria, Smitha Kolan, Franziska Hinkelmann, Ph.D., Polong Lin, Toni Klopfenstein, and Lavi Nigam for creating such a practical, insightful, and engaging learning experience, and Big thanks to all authors of whitepapers and the guests who contributed to the livestream Q&A sessions. #AIAgents #AgentEngineering #VibeCoding #GoogleAI #Kaggle #GenerativeAI #LLM #MCP #A2A #ADK #CloudRun #AISecurity #DeveloperProductivity #PlatformEngineering

From Prototype to Production: Day 5 of the Google AI Agents Intensive with Kaggle We've reached the final day of the Google Gen AI Intensive with Kaggle! After exploring agents, interoperability, skills, and security, today's focus is on taking AI applications beyond prototypes and building production-grade systems that are scalable, secure, and enterprise-ready. 💡 What You'll Learn Today's whitepaper, "Spec-Driven Production Grade Development in the Age of Vibe Coding" explores how to bridge the gap between rapidly built AI prototypes and robust production software using Spec-Driven Development (SDD). Instead of treating code as the source of truth, SDD promotes behavior-driven specifications (Gherkin) as the foundation of development. The paper also discusses: Treating code as disposable while specifications remain the source of truth Building zero-trust development pipelines Using automated AI code-review agents for continuous validation Leveraging Hybrid Policy Servers to enforce governance, compliance, and security Creating reliable workflows that scale from experimentation to enterprise deployment In today's optional codelabs, we'll: Deploy and host AI agents on Google Cloud for enterprise-scale applications Build a frontend application using Antigravity and deploy it to Cloud Run Connect the frontend with a cloud-hosted AI agent through an asynchronous event-driven architecture Although these codelabs require a Google Cloud billing account, they provide valuable insight into what productionizing AI agents looks like in real-world enterprise environments. 📚 Day 5 Content Unit 5 – Spec-Driven Production Grade Development in the Age of Vibe Coding 🎧 Summary Podcast: https://lnkd.in/dkDvx3d6 📄 Whitepaper: https://lnkd.in/drDAsuiB 🛠️ Optional Codelabs Deploy & Host AI Agents on Google Cloud by Toni Klopfenstein https://lnkd.in/dtC9KyXF Build a Frontend App with Antigravity & Connect it to Your Cloud-Hosted Agent by Sita Lakshmi Sangameswaran https://lnkd.in/dbzmZqiT 📋 Final Livestream Join the final livestream tomorrow at 11:00 AM PT / 8:00 PM CET / 11:30 PM IST, hosted by Anant Nawalgaria and Smitha Kolan. They'll be joined by codelab author Lavi Nigam, along with Google experts Ankur Jain, Antonio Gulli, Elia Secchi, Lee Boonstra and Omar Sanseviero to discuss the final assignments and share production best practices. 🎥 Final Livestream: https://lnkd.in/dxqmx5Ht 📺 All livestream recordings: https://lnkd.in/dSujrvhH It's been an incredible journey exploring the next generation of AI engineering from vibe coding and agent development to building secure, interoperable and production ready AI systems.

Vibe coding gets you a prototype in minutes, but going fast can also generate bugs faster. It was great joining Lee Boonstra, Anant Nawalgaria, Smitha Venkat, Smitha Kolan Antonio Gulli on stage to discuss what it actually takes to build coding agents for production. We also released two whitepapers detailing these patterns: Spec-Driven Development in the Age of Vibe Coding: 👇 https://lnkd.in/dTwr7zwQ Agent Prototype to Productization (Updated): 👇 https://lnkd.in/djnhkCU7 To build agents yourself and try the Agents CLI, this hands-on codelab authored by Toni Klopfenstein will get you started in minutes: 👇 https://lnkd.in/d-rf3uiK