
Introducing Gemini 4 Hydrogen–bringing cutting-edge intelligence with speed and efficiency to everyday coding and agentic enterprise work. Learn more → goo.gle/4zTuFDK
✓Follow along for how-tos, demos, product news, and more. For company updates, check out @GoogleCloud.

Introducing Gemini 4 Hydrogen–bringing cutting-edge intelligence with speed and efficiency to everyday coding and agentic enterprise work. Learn more → goo.gle/4zTuFDK

Best practices for GKE—all in one place. This document provides a consolidated overview of best practices for designing, building, and operating apps on GKE. Follow these recommendations to optimize for cost, performance, security, and reliability → goo.gle/4j0JtJR

Apigee Hybrid 1.17 now supports MCP! Expose your existing APIs as governed MCP tools across on-premises and multicloud environments without writing or maintaining custom MCP servers → goo.gle/4iPhdti

48 hours until Project Masar Season 2 goes live! 🚀 Season 1 brought hundreds of developers across MENA together to deploy serverless apps and foundational agents in Arabic. This Wednesday, November 18 at 7:00 PM KSA (GMT+3) / 6:00 PM Cairo, we are taking the next step moving from single agents to Multi-Agent Systems with the Agent Development Kit (ADK). Here is what we are building live over the next 5 episodes: 🔹 Ep 1: Empowering AI Agents with Tools 🔹 Ep 2: Stateful & Personalized Agents (Memory) 🔹 Ep 3: Interoperability with MCP & A2A 🔹 Ep 4: Multi-Agent Apps with MCP Toolbox 🔹 Ep 5: Multi-Agent Workflow Patterns 100% Arabic, 100% hands-on live coding, zero slides. RSVP now: 👉 g.dev/cloud/masar #MasarByGoogle #GoogleCloud #ADK #MultiAgent #DevRel #MENA

Get hands-on with the Google Cloud Developer Plugin in Antigravity ↓ skills.google/focuses/167163…

Check out the full article ↓ cloud.google.com/blog/topics…

5. Encode operational habits, not documentation dumps Re-evaluate your skills every time you upgrade the underlying AI model. If a new model performs just as well without the skill, delete the skill to keep code clean.

4. Separate efficiency and security from correctness so agents can't reward-hack Don’t let agents game the system. Isolate your grading environment and keep your evaluation metrics separate to avoid reward hacking.

3. Measure paired WITH vs. W/OUT ablation, not vibes When a test case shows the same success rate with no efficiency gains, stop optimizing. Your custom skill isn’t speeding up the run. It's just adding context bloat. Delete it.

2. Never trust an eval that runs on a developer laptop Running skill benchmarks directly on your workstation? Your numbers are lying to you. Slower file scans and subprocess timeouts can cause the agent to abandon search paths prematurely.

1. Gate trigger collisions and security risks before spinning up containers No eval benchmark improvement after adding a plugin? Before rewriting SKILL.md instructions, check if the agent actually loaded the skill.

Testing agent skills files? Don’t just paste a prompt, watch it pass once on a laptop, and merge to main. Here are five rules for evaluating skills before they bloat your context window ↓

How to build an AI harness around your coding agent in four steps: 1. Context and memory (sessions, state) 2. Tools and workflows (MCP, ADK) 3. Sandboxes and prompt injection defense 4. Evals (code, LLM judge, human) Watch the crash course → goo.gle/3XtB36o

Level up your security posture with Certificate Manager (2nd gen), now in preview. Centralize observation and automate management of SSL/TLS certificates across your org: 1. Overview → goo.gle/4ev6djM 2. Compare Certificate Manager versions → goo.gle/3WGCYnY

DYK? In the BigQuery Data Transfer Service, you can view granular error details and progress for individual tables during your data lake migrations. Dive into the docs → goo.gle/4kaA3h1

Knowledge Catalog, Google Cloud's universal context engine, turns knowledge into shared business context so agents answer and act with higher accuracy and trust ↓