Ciao! 👋 Quick one for you this week, because it's a real opportunity rather than a think-piece: as a media partner of AGNTCon + MCPCon Europe, I've got my subscribers 25% off — and the event is in our backyard, Amsterdam, on 17–18 September. If you build agents, tooling, or the platforms they run on, this is the most relevant conference I've seen land this side of the Atlantic this year. Here's why, plus the code. 1) Three signals I'm watching Agentic AI is consolidating around open protocols. MCP went from "interesting" to "the default connector" in about a year. AGNTCon + MCPCon exists precisely because the agent stack is now a protocols-and-tooling problem, not just a model problem. The people writing the specs are on the stage. David Soria Parra, co-creator of the Model Context Protocol at Anthropic, is keynoting — alongside Clare Liguori (AWS, core MCP maintainer), Angie Jones (Agentic AI Foundation), and Dexter Horthy (HumanLayer, who coined "context engineering"). That's rare access to the people defining the standards. Certification is showing up next to the conference. The Linux Foundation is running the Model Context Protocol Associate (MCPA) certification on-site. If you're hiring or positioning yourself in this space, a vendor-neutral credential tied to the actual protocol matters more than another course badge. 2) My take: protocols beat frameworks for staying power I've watched enough "agent frameworks" appear and fade. What survives is usually the interface, not the implementation — and MCP is becoming that interface between models and the systems they act on. For platform and DevOps teams, the practical read is simple: spend your learning budget on the protocol, the identity model around agents, and how you'll observe and govern tool calls. That travels across every framework. The framework you picked in March probably won't. → AGNTCon + MCPCon Europe — about the event 3) From the field

August 2026 roundup: KubeCon Japan, Red Hat Summit, Hiroshima, Amsterdam meetups, and the AI platform trends I'm watching. This is the first newsletter-style roundup since July, so it covers a packed few weeks — from KubeCon Japan in Yokohama to Red Hat Summit in Atlanta, a personal trip to the Hiroshima Peace Memorial Ceremony, and two Amsterdam meetups. Here is what I have been building, speaking about, and writing, plus the trends I am watching in AI platform engineering. Latest initiatives & events I attended KubeCon + CloudNativeCon Japan 2026 — Yokohama. The Japanese cloud-native community is running real AI and Kubernetes in production at a scale worth learning from. Two takeaways stood out: - Photonic networks could reshape the AI data center by moving data optically — Fujitsu is leading this work in Japan. - Tuning Kubernetes for AI is the real story: it is not “more GPUs,” it is LLM caching and routing and the true GPU, memory, and electricity cost. Red Hat Summit 2026 — Atlanta. I delivered a lightning talk, “GPUs take flight: Safety-first multi-tenant Platform Engineering with NVIDIA and Red Hat OpenShift AI,” covering isolation patterns, platform engineering, and production-tested guardrails for sharing expensive GPUs safely. A short booth pass even turned into a Docker Captain announcement moment. Hiroshima Peace Memorial Ceremony 2026 — 6 August. I attended in person. The minute of silence, 120 nations, and two hibakusha testimonies were a powerful reminder of why we build the things we build. AI Builders Amsterdam — 19 August (StartDock) and BrowserStack Meetup Amsterdam — 20 August. Two hands-on meetups: validating side projects, writing docs for agents, running local LLM agents, and how Claude plus Playwright are reshaping QA. Coming up: EDGECASE 2026 — 24 September, Hilversum. The Netherlands’ biggest Kubernetes event (400+ cloud-native engineers, hosted by Bart Farrell). I will be there — save the date. Latest trends I am watching - GPU multi-tenancy is finally real. Fractional GPU allocation (the NVIDIA DRA driver), GPU-aware scheduling (KAI Scheduler), and llm-d for splitting inference into prefill and decode are landing in the CNCF. Sharing GPUs safely is the number-one cost lever for 2026. - Agentic AI is getting an identity layer. kagent (agents as Kubernetes CRDs) and kagenti (cryptographic SPIFFE/SPIRE identities for agents) turn agents from “rogue processes in a notebook” into governed, observable, RBAC-controlled resources. - Local and edge agentic models are shipping. Meta’s open-source Muse Glimmer (30B) runs under 20 GB on a consumer GPU via 4-bit quantization — private, offline agentic AI is no longer a lab curiosity. - Sovereign and compliant AI is now a platform requirement, not a nice-to-have. Data residency and audit-readiness are spreading globally.

Hello everybody! 👋 Mi name is Glendel Fyne, and I am very excited to join me to the "Luca Berton AI Cloud Lab" community. I am currently exploring how to build more reliable AI Agents for real-world contexts: roles, memory, permissions, governed execution, evidence, coordination, and clear boundaries between what the model reasons and what the system actually executes. I am very interested in learning from the community, sharing ideas, and comparing approaches with people who are building AI Agents in real-world scenarios. It is a pleasure to be here, and thanks again for welcoming me to this community! 🚀

How many AI coding tutorials have you watched — and how many real projects have you actually shipped? This course closes that gap. You'll internalize one loop that runs through everything you build — **Plan → Implement → Test → Review → Commit** — and use it to ship a real, tested feature end to end. Write prompts that read like well-scoped tickets, generate honest test suites, run Best-of-N to pick winners on evidence, and wire up professional git, MCP, hooks, and skills. You stay the engineer of record the whole time: Claude proposes, you decide what ships. What you'll learn: • Run the five-step loop on every change so AI bugs are caught before prod • Write spec-grade prompts and CLAUDE.md brain files • Pick the best solution with Best-of-N and rubric scoring • Generate honest tests and make Claude review its own code • Run safe git workflows and connect tools via the GitHub MCP server • Automate guardrails with hooks and skills, then score production readiness 🔴 Live cohort (5-hour intensive, limited seats): https://maven.com/luca-berton/claude-code-ai-coding-masterclass 🎓 Self-paced on Udemy: https://www.udemy.com/course/claude-code-masterclass/ 📘 Companion book on Leanpub: https://leanpub.com/claude-code Taught by Luca Berton — cloud architect, automation engineer, and bestselling author of *Ansible for VMware by Examples* and *Ansible for Kubernetes by Examples*. #ClaudeCode #AICoding #PromptEngineering #SoftwareEngineering #Anthropic

May was a packed month on the blog — and one theme kept showing up everywhere: AI is moving from experiments into production systems. I covered a lot this month, including: 🤖 Agentic AI on AWS - AWS Summit Amsterdam 2026 focused heavily on Amazon Bedrock AgentCore, managed agents, enterprise data platforms, and how companies are starting to run AI agents safely at scale. 🧠 Claude Code and AI-assisted development. I announced the Claude Code Bootcamp and shared practical workflows for using CLAUDE.md files, test generation, screenshot-to-UI workflows, and production-ready AI coding patterns. ☁️ GPU platforms and enterprise AI. At Red Hat Summit 2026, I presented “GPUs Take Flight,” focused on multi-tenant GPU platform engineering with OpenShift AI, NVIDIA, quotas, guardrails, and cost control. 🔐 AI security and observability. A big topic this month was how to secure, monitor, and govern AI agents — especially when they interact with tools, files, email, APIs, and production environments. 🛠️ Practical cloud engineering. I also shared hands-on posts about AWS account cleanup, SQL Server performance tuning, Vercel cost optimization, Rust in 2026, and local/open LLM deployment. The big takeaway? AI agents are no longer just cool demos. They now need the same discipline we already apply to cloud infrastructure: identity, observability, security, cost control, and deployment automation. That’s where the real work begins. 👉 Full May recap here:https://lucaberton.com/blog/ What was the most interesting AI/cloud topic for you this month?