engblogs

Misconfigured routing policy automation caused a 25-minute IPv6 BGP route leak at Cloudflare’s Miami data center, exposing prefixes to unintended peers and workloads, followed by rapid rollback, incident response, and security-focused mitigations.

Concise overview of Cloud Composer's Airflow 3.1 preview, detailing a decoupled architecture, DAG versioning, managed backfills, event-driven scheduling, HITL workflows, and multi-language extensibility to advance workflow orchestration.

Shows how Datadog LLM Observability automatically instruments Google ADK agentic applications to trace agent decisions, monitor token usage and latency, evaluate response quality and safety, and run offline and online experiments to optimize performance before production.

A scalable, self-learning generative AI system for catalog enrichment that uses multiple lightweight worker models and a supervisor agent on Amazon Bedrock to extract and improve product attributes at scale.

Demonstrates provisioning and orchestrating production-grade AI agents with Amazon Bedrock AgentCore via AWS CloudFormation and Infrastructure as Code, highlighting modular templates, observability, and end-to-end weather-focused workflows.

A technical look at how GPT-5 enables workplace AI, with deployment patterns, integration approaches, and real-world business use cases.

Explores how to architect, evaluate, and govern production-grade AI agents on Databricks MLflow (with LangGraph) to ensure responsible, calibrated performance in telecom use cases through guardrails, monitoring, and guideline-driven evaluation.

An end-to-end guide to building, testing, and deploying a Gemini 3 app on Google Cloud Run using Google AI Studio's Build mode, turning ideas into a shareable URL with API access.

A concise exploration of scaling PostgreSQL to support 800 million ChatGPT users, focusing on performance, reliability, and operational scalability.

How Agentforce 360 unifies Salesforce’s data, agent, and applications into a single real-time, AI-driven platform powering enterprise experiences across service, sales, and marketing.

Gemini CLI and Gemini 3 empower Google SREs to orchestrate AI-assisted, human-in-the-loop outage response—from paging to postmortem—dramatically shortening Mean Time to Mitigation.

Scaling WideEP MoE inference on Google Cloud's A4X (GB200 NVL72) with NVIDIA Dynamo to deliver rack-scale, disaggregated MoE serving that balances throughput and latency via GPUDirect RDMA, KV caching, and GKE orchestration.