Blog

Kubernetes autoscaling, cost, and the craft of running lean.

Engineering deep-dives, customer stories, and product news from the team building CloudPilot AI.

Featured · Product

CloudPilot AI Now Supports MCP: Bring CloudPilot Insights to Your AI Assistant

CloudPilot MCP is now available in preview, giving Codex, Claude Code, and other MCP clients secure, read-only access to operational and optimization insights available in CloudPilot.

CloudPilot AI·Sep 2, 2026·4 min read

Recent posts

Case Study

How a Generative AI Company Cut AWS EKS Costs by 55% with Full-Stack Intelligent Autoscaling

A leading generative AI company reduced its real AWS EKS costs by more than 50% and sharply improved cluster resilience—without changing a single line of code. See how CloudPilot AI's full-stack autoscaling rightsized workloads and reshaped nodes on top of EKS Auto Mode.

CloudPilot AI·8 min
Karpenter

CloudPilot AI Now Supports Microsoft Azure: Autonomous Cost and Reliability for AKS

CloudPilot AI now supports Microsoft Azure. AKS already provisions nodes with Karpenter-based Node Auto Provisioning—CloudPilot AI adds the optimization layer on top: rightsizing, safe Spot, and graceful node replacement, through a read-only agent.

CloudPilot AI·5 min
Kubernetes Optimization

Hermes EKS Benchmark: 85% Faster Image Pull and 34% Faster First HTTP 200, No Image Rebuilds

EKS benchmark results for Hermes on OpenSearch, Solr, and Spark: image pull was 71-85% faster, and first HTTP 200 improved by up to 34%, using the original OCI images.

CloudPilot AI·6 min
Karpenter

CloudPilot AI Now Officially Supports Google Cloud: Smart GKE Autoscaling from the Maintainers of Karpenter

CloudPilot AI now officially supports Google Cloud, bringing just-in-time provisioning, workload rightsizing, and predictive Spot handling to GKE—from the maintainers of karpenter-provider-gcp.

CloudPilot AI·4 min
Kubernetes Optimization

Hermes: Cut Pod Ready Time from 4m35s to 14s, No Image Changes

Hermes validates policy-driven SOCI lazy loading on EKS and Karpenter, reducing the scheduled-on-node to Running/Ready path for a 10.8 GB vLLM image from 4m35s to 14s.

CloudPilot AI·4 min
Kubernetes

Up to 45% Less Memory, Zero Compromise: Smart Java Workload Optimization on Kubernetes

Learn how CloudPilot AI jointly optimizes JVM Heap and Kubernetes container resources to cut Java workload memory by 40–45%—without sacrificing stability or requiring manual tuning.

CloudPilot AI·8 min
Company News

CloudPilot AI Achieves SOC 2 Type II Compliance

CloudPilot AI has successfully achieved SOC 2 Type II compliance, validating its enterprise-grade data security, regulatory compliance, and operational excellence for its autonomous SRE-agent platform.

CloudPilot AI·2 min
Cloud Optimization

Beyond the Report: How CloudPilot AI Solves the Efficiency Gaps in Datadog’s 2025 Analysis

Datadog's latest data reveals a harsh reality: 75% of container resources go wasted. This post explores why Karpenter is only half the solution, how CloudPilot AI tackles the 'Long-Tail' Java problem, and the hidden reliability of ARM Spot instances.

JingKang Jiang·4 min
KubeCon

Karpenter at KubeCon + CloudNativeCon North America 2025: Visit Our Project Pavilion Booth

CloudPilot AI·3 min

The Kubernetes Autoscaling Brief

Once a month: what we see in the Spot market, one real cluster torn down, and what shipped.

No spam. Unsubscribe anytime.