Day 2 of the Google × Kaggle 5-Day AI Intensive focused on extending what AI agents can do, not just what they can say. We explored how to: Add external tools to agents to enable real-world actions Design tool interfaces thoughtfully to avoid ambiguity Use Model Context Protocol (MCP) to make tools interoperable across agents and platforms Handle long-running tasks and asynchronous workflows reliably Our hands-on lab involved integrating new tools into our existing agents and testing how they collaborate and delegate tasks in a multi-agent setup. Amazing to see how agents evolve from reasoning systems → into actionable, goal-driven digital teammates. Excited for Day 3! 🚀 #AI #AgenticAI #MCP #Gemini #Kaggle #Google #5dayAiAgentIntensive

“Prompt sets are the new PRDs,” says Aparna Chennapragada, CPO for AI Experiences at Microsoft. In the pre-AI world, product-building began with long specs and stakeholder reviews. In the AI world, it starts with prompts. In the latest episode of #DecodingAI, she shares how product thinking itself is being rewritten when you build AI-first. Watch the episode on our YouTube channel to understand how foundation models are changing every layer of product-building: https://lnkd.in/gzSqA36z Anand Daniel • Accel

Remember when we thought chatbots were impressive? Today I learned we're already way beyond that. 🚀 Day 1 of Google's Gen AI Intensive opened my eyes to AI Agents - systems that don't just respond to prompts, but actively plan, reason, and take action to solve problems. I built two hands-on projects: 1️⃣ From Prompt to Action: My first autonomous agent using Gemini and ADK 2️⃣ Agent Architectures: A multi-agent system where specialized AI agents work as a team The difference? Traditional LLMs are like asking someone a question. AI Agents are like having a colleague who can figure out what needs to be done and actually do it. The key components that make this possible: → Models (the brain) → Tools (the hands) → Memory (the context) → Orchestration (the coordination) → Evaluation (the quality control) This is just day 1 of 5, and I'm already seeing why agents are called "the next frontier of AI." 📂 My notebooks: https://lnkd.in/eEs4T9Re https://lnkd.in/e458U2XV Who else is taking this course? Let's connect! 👇 #GenerativeAI #AIAgents #MachineLearning #Kaggle #Google #LearningInPublic

Wrapping up Day 3 of the Gen AI Intensive Course with Kaggle & Google — it’s all about memory and context! From short-term session memory to long-term persistence, today was a deep dive into making AI agents truly stateful and personalized. #GenAI #Google #Kaggle #AIInnovation #ContextEngineering #DeepLearning #AIJourney #TechLearning

Day 4 of the Google × Kaggle AI Agents Intensive 🚀 Today’s sessions moved beyond simply building agents and focused on something most people overlook — making agents measurable, debuggable, and production-ready. This is the part that separates toy projects from real-world AI systems. Key Takeaways: • Set up observability using logs, traces, and metrics to understand exactly how an agent behaves internally. • Inspected logs to study chain-of-thought transitions and tool usage patterns. • Used trace analysis to identify why an agent selected a particular action or step. • Explored metrics for evaluating accuracy, stability, responsiveness, and tool performance. • Learned agent-evaluation techniques using LLM-as-a-Judge and HITL scoring frameworks. Core insight: Building an agent is just the starting point — measuring, explaining, and debugging its decisions is what makes it trustworthy and ready for real deployment. Up next: Day 5 — combining everything and applying evaluation at scale. 🔥 #GoogleAI #Kaggle #AIAgents #AgentOps #AIObservability #MachineLearning #LearningJourney

It all started with a simple curiosity — what if AI models could not just think, but act? This single question opened the door to one of the most fascinating learning experiences I’ve had recently, the 5-Day “AI Agents” learning journey by Google & Kaggle. I walked in with a solid understanding of LLMs and left with a deeper appreciation for Agentic intelligence, systems that observe, decide, and evolve. Each day peeled back a new layer of how the next generation of AI is being built. Here’s a quick recap of what I learned: Day 1: Agent Fundamentals - How do we shift LLMs from thinking to acting? → Through agent components, hierarchies, and structured workflows. Day 2: Advanced Tools & Integration - How do we solve the huge N agents × M tools integration challenge? → Using the Model Context Protocol (MCP). Day 3: Sessions & Memory - How can agents go from knowing “facts” to knowing you? → By managing current context and long-term memory. Day 4: Observability & Evaluation - What happened? → Logs (events). - Why did it happen? → Traces (stories). - How well is it working? → Metrics (performance). Day 5: Production Deployment - How do we go from a cool demo to a reliable system? → Through AgentOps and the Observe-Act-Evolve loop. Resources: The course provided incredible whitepapers + hands-on codelabs. 👉 Course Guide: https://lnkd.in/gffiYVGk https://lnkd.in/gr_nSNhQ A huge thank-you to all the brilliant authors, speakers, and organizers who made this journey possible — Kanchana Patlolla, Anant Nawalgaria, Antonio Gulli, Dr. Sokratis Kartakis, Ran Li, Huang Xia, Saurabh Tiwary, Lavi Nigam, Kimberly Milam, Julia Wiesinger, Kristopher Overholt, Laxmi Harikumar, Maddula Sampath Kumar and many others This experience has truly amplified my agentic AI mindset — inspiring me to keep building intelligent, self-evolving systems that bridge human goals and machine autonomy. #AI #AgenticAI #MachineLearning #Google #Kaggle #LLMs #AIEngineering #LearningJourney #AgentOps

I pulled together some thoughts on the AI dev tools I am using in my day-to-day life and how they fit into my workflow. Do give it a read: https://lnkd.in/dh4iyXts

🤖 Day 2 — Intensive AI Agent Course with Google × Kaggle 🌍✨ Today’s session focused on how AI agents think, connect, and collaborate through modern tools and protocols. Here’s what I learned: 🔹 Explored different agent architectures — how they sense, reason, and act. 🔹 Understood agent tools that help in building and managing intelligent systems. 🔹 Learned about interoperability with MCP (Model Context Protocol) — how it enables agents and models to communicate seamlessly across platforms. 🔹 Applied these concepts through hands-on experiments in Kaggle notebooks, observing real-time agent interactions. Each day is deepening my understanding of how AI agents can work together intelligently to solve real-world problems. 🚀 #Google #Kaggle #AI #AIAgents #ArtificialIntelligence #MachineLearning #MCP #LearningJourney #Day2 #ContinuousLearning

🌟 Day 3 Complete: #Kaggle x #Google 5-Day AI Agentic Course Today’s focus was on something that makes artificial intelligence feel a little more human: memory. I learned how to help an AI agent remember information, continue conversations naturally, and use past context to make smarter decisions. 💡 What I Learned Today: ✅ Context Engineering I explored how to organize information for the model so that it always knows what is most important in a conversation. Instead of repeating everything each time, the system learns to keep only what truly matters. ✅ Sessions and Events I learned how to maintain a continuous conversation with the agent. This means it can remember what was said earlier and respond in a more natural and connected way instead of starting fresh every time. ✅ Persistent Storage This part was really interesting. I saw how conversations can be saved and restored even after the system is restarted. It feels like giving the AI a memory that stays with it between uses. ✅ Session State Here I learned to keep track of small but important details, such as user choices or information mentioned earlier. This helps the agent make consistent decisions later on. ✅ Building Memory I gave the agent a real memory system using a memory service. It can now save information from previous sessions, search through what it has seen before, and bring back relevant details whenever needed. There are two ways it can do this: • In a reactive way, where it decides when to recall something. • In a proactive way, where memory is automatically loaded at the start of the conversation. ✅ Memory Consolidation This is where the agent summarizes what it has learned from many conversations and keeps the most useful knowledge. We used tools that act like an organized memory bank, helping the system grow more capable over time without forgetting older information. By the end of the day, my agent could hold meaningful, continuous conversations, remember earlier discussions, and respond more intelligently based on past experience. This step made me realize that true progress in AI is not only about reasoning or coding. It is about creating systems that can learn, remember, and understand context just like we do. #GoogleAI #Kaggle #AIAgents #AgenticAI #Gemini #VertexAI #AIInnovation #MachineLearning

Back in the day, Dennis wrote about how we built a Genkit plugin for Deepseek, a tool that helps developers and businesses make their AI workflows smarter and smoother. It started from curiosity, turned into a project that helps others build better and faster. We highly recommend you to read Dennis Alund's complete writing here! https://lnkd.in/g7aDqUMF #oddbit #deepseek #firebasegenkit #opensource #softwareengineering #innovationpartner