Platform Pulse: Stateless MCP, Database Agents, and TPU Subslicing

🔐 Zero-Trust Security, Identity & Threat Defence

Security architectures are actively evolving to combat automated threats, introducing advanced data protection for AI models, secure runtimes, and quantum-resistant cryptography.

1. Model Armor Doesn’t Redact PII. Sensitive Data Protection Does: An explanation of how to properly configure Google Cloud Model Armor and Sensitive Data Protection (SDP) to redact personally identifiable information (PII) from AI model responses instead of simply blocking them.
2. Eval Is Evil: How to Safely Execute Untrusted AI Code with Cloud Run sandboxes and ADK: Exploring how Google Cloud Run sandboxes provide a secure isolation boundary for safely executing untrusted user input and AI-generated code.
3. Future-proofing data integrity: Quantum-safe digital signatures in Cloud KMS: Extending the PQC digital signature algorithms suite available in Google Cloud Key Management System to include ML-DSA and SLH-DSA.
4. Mastering Agent Runtime Security Controls on Google Cloud Platform: A comprehensive architectural guide for securing enterprise AI Agent Runtimes on GCP, focusing on preventing unauthorised data exfiltration and shadow tool integrations.
5. Cloud CISO Perspectives: Why AI Threat Defense is the new boardroom baseline: Why boards and CISOs must focus on AI-era security governance and threat defence as a standard business baseline.
6. Best Buy scales AI workloads and secures access with Workforce Identity Federation: To support tens of thousands of users, Best Buy modernised its identity architecture using Workforce Identity Federation.
7. Advancing brain tumor research with privacy-first AI: Discover how Google Cloud and MedPerf use Confidential Computing to enable secure, privacy-first collaborative medical AI evaluation.
8. Batten Down Your Packages: Mitigation Guidance for Supply Chain Compromise: Trends observed in threat actor use of software supply chain compromise, complete with mitigation and hardening recommendations.
9. UNC6671 Rebrands: Multi-Brand Vishing Extortion Targets Cloud Environments: Tracking UNC6671's rebranding from BlackFile to REDACT as they diversify extortion operations targeting financial and cloud services.

📊 High-Performance Databases, Analytics & Open Knowledge

Data platforms are focusing heavily on performance optimisation, cross-cloud querying, and deploying autonomous agents for database operations.

1. Unifying public and private data: Scale knowledge graphs with Data Commons on Spanner: Create a knowledge graph combining public Data Commons data as well as your private data with the new Spanner Graph-native platform.
2. Introducing Database Operations Agents: The future of autonomous database management: A closer look at AI-powered database operations agents for setting up, onboarding, and managing AlloyDB, Bigtable, Cloud SQL, and Spanner.
3. The borderless Lakehouse: Bring AWS, Databricks and Snowflake data to your AI agents: The borderless Lakehouse provides secure, bi-directional access via BigQuery and Managed Service for Apache Spark to any Iceberg-compatible engine.
4. Databricks Iceberg — BigQuery query: A comprehensive technical guide on how to configure and query Apache Iceberg tables created in Databricks using BigQuery.
5. Zero-code, low-cost data ingestion: New BigQuery DTS capabilities: Automate low-cost data ingestion from diverse sources directly into your data warehouse with new BigQuery Data Transfer Service capabilities.
6. Building an Agentic DB Analyst: pgvector, Google Cloud Storage (GCS), and Deterministic SQL: Bridging relational databases and Cloud Storage without sacrificing mathematical accuracy using deterministic SQL and pgvector.
7. How Target is enhancing retail discovery with Spanner Graph: Target transitioned to Spanner Graph to overcome siloed data and build a generative AI-powered Shopping Graph.
8. Agentic Future Ready With BigQuery: Continually improving price-performance through autonomous query processing with zero effort required.
9. How Database Migration Service automates SQL server to PostgreSQL translation: Automating the translation of SQL Server Stored Procedures into PostgreSQL using Google DMS.
10. How to Build a Serverless Agent Observability Pipeline in BigQuery: Building a real-time observability pipeline for multi-agent systems using the ADK, Continuous Queries, and BigQuery ML.

⚡ High-Efficiency Compute, Networking & GKE Workloads

Cloud infrastructure engineering continues to maximise accelerator density, introducing new failover mechanisms and smarter workload scheduling.

1. GCP Cloud Run’s New Automated Failover Is a Direct Answer to a Real Outage: Exploring Google Cloud’s rapid launch of automated cross-region failover for Cloud Run, utilising standard readiness probes and application load balancers.
2. Stop Over-Provisioning for Startup: GKE’s New CPU Startup Boost: GKE's new CPU startup boost feature allows developers to temporarily increase CPU requests during heavy application initialisation phases.
3. How to build an elastic, scalable LLM Inference Platform on GKE using Fluid Compute: Architecting a fault-tolerant, cost-optimised LLM serving platform on Google Cloud using diverse GPU consumption types and multi-region GCS buckets.
4. Scaling real-time AI agents with session-aware load balancing: Implementing application-level session tracking directly within the runtime to accurately measure the committed concurrent workload of active conversations.
5. Inside the optimization of Mistral 3 large inference on Ironwood: Detailed strategies like hybrid sharding and SparseCore tree reductions that boosted Mistral 3 inference throughput by 48% on Ironwood TPUs.
6. How We Reduced GKE Costs by 33% Through Smarter Scheduling and Resource Governance: A DevOps case study demonstrating how migrating to Spot nodes and right-sizing resource requests transformed cluster efficiency.
7. Cloud Run vs. GKE Autopilot: A Comparison for ADK Agents: A practical evaluation highlighting differences in identity configuration, cold-start behaviour, and infrastructure complexity when hosting ADK workloads.
8. Do more with less: How GKE can reduce your cost per agent by 75%: GKE Agent Sandbox increases agent density up to 3.5x and cuts compute costs without sacrificing performance.
9. Bootstrapping OSS Kubernetes on GCE with TPU6 and Open-Source DRANET: Exploring how to deploy an open-source Kubernetes cluster directly on Google Compute Engine while integrating Cloud TPU v6e accelerators.

🛠️ Ecosystem, Agents & Platform Modernisation

The ecosystem is rapidly standardising agentic workflows, focusing on stateless protocols and unified model routing gateways.

1. Scaling AI Agent Infrastructure with the MCP Stateless updates: The 2026-07-28 Model Context Protocol (MCP) specification introduces a fully stateless core, enabling cloud-native horizontal scaling and serverless deployments.
2. Agent Plugins package your skills, tools, and more: Agent Plugins 1.0.0 provides a vendor-neutral directory specification for packaging Agent Skills and MCP servers into a single portable unit.
3. Model routing with Google Cloud API Gateway: API Gateway now allows developers to dynamically route traffic to models like Gemini, Claude, or OpenAI OSS-GPT directly within their OpenAPI 3.x specifications.
4. From repetitive queries to instant SQL: Building Carrefour’s internal data assistant: How retail giant Carrefour leveraged Google Cloud ADK, Gemini, and Cloud Run to build a scale-to-zero RAG assistant integrated directly into Google Chat.
5. Enable on-demand expertise with Agent Skills in Genkit Go: To prevent context window bloat, Genkit Go introduces Agent Skills based on a progressive disclosure architecture.
6. Finally — Hard Caps to Limit Your Google Cloud Spend: Google Cloud introduces native Cloud Spend Caps to hard-stop spend before your budget target amount is exceeded.
7. The anatomy of an AI agent on Google Cloud: a complete guide: A comprehensive guide breaking down agent architectures—covering models, frameworks, runtimes, and memory—onto Google Cloud products.
8. Behind the scenes: How we build, test, and scale Google Agent Skills: Discover how Google maintains quality for Agent Skills with standard layouts, automated CI/CD checks, and continuous evaluations.
9. Scaling agentic AI: How UiPath built its high-performance GPU platform on AI Hypercomputer: UiPath successfully transitioned toward agentic AI using a high-performance GPU platform powered by Google Cloud.
10. Automate data monitoring and root-cause analysis with Looker Agentic Workflows: Extending one-time questions to ongoing inquiries utilising background agents that surface results directly in chat.

📋 Essential Release Notes

A summary of critical platform updates, feature deprecations, and runtime enhancements rolling out across Google Cloud.

1. Cloud Run: Support for sandboxes is now available in Preview for all resources, including jobs and worker pools.
2. GKE: TPU Subslicing (Dynamic Subslicing) is now Generally Available for Ironwood (TPU7x), enabling incremental provisioning of node pools.
3. Cloud SQL: DNS automation and global write endpoint DNS are Generally Available on instances with Private Service Connect. Performance capture is also GA.
4. Cloud KMS: Cloud KMS now supports quantum-safe key import in Preview, including HPKE_KEM_ML_KEM_768 and 1024 methods.
5. Secret Manager: Automatic rotation of regional Cloud SQL database credentials is now available in Preview.
6. Load Balancing: Regular expression URL rewrites (regexRewrite) for route rules in URL maps are now available in Preview for Application Load Balancers.
7. Compute Engine: The maximum IOPS per GiB for Hyperdisk Balanced Storage Pools have increased to 30 IOPS/GiB for Standard performance and 6 IOPS/GiB for Advanced.
8. AlloyDB: BigQuery integration enables real-time access (lakehouse federation) and periodic data synchronisation (Preview). Best Matching 25 (BM25) indexes are also supported.
9. Bigtable: The Bigtable remote MCP server and execute_sql tool are now Generally Available. You can also utilise Bigtable as a remote storage backend for LMCache (Preview).
10. Confidential VM: The accelerator-optimised g4-standard-48 machine type (Turin, AMD SEV, NVIDIA RTX PRO 6000) is now GA.
11. Policy Intelligence: The Policy Troubleshooter MCP server is generally available to let agents and AI applications troubleshoot IAM issues.
12. Billing: Spend cap budgets are available in Preview, alongside the new Originating products filter in Cloud Billing Reports.