May 2026 Summaries

Cast AI's innovative project at their internal hackathon led to a significant overhaul of their Kubernetes cluster snapshot system, achieving remarkable improvements in storage efficiency and processing speed. The existing system, which took snapshots every 15 seconds, resulted in over a petabyte of data monthly and incurred high storage costs due to duplicated data. The hackathon-inspired redesign, named Snapshots V2, introduced a custom binary format with selective loading, dictionary-based differential compression, and lazy loading with smart memory management. These innovations allowed for a 93% reduction in storage space, saving over $300,000 annually, and an 82% decrease in snapshot processing time, enhancing service efficiency without disrupting existing operations. The new system not only reduces storage and compute costs but also maintains the high-frequency, detailed capture that is crucial for Cast AI's operations, including cost optimization, customer reporting, machine learning, and customer support.

Application Performance Monitoring (APM) and Application Performance Automation (APA) provide distinct yet complementary approaches to managing system performance, particularly in Kubernetes environments. APM tools, such as Datadog and Prometheus, offer visibility by monitoring response times, error rates, and infrastructure metrics, allowing teams to identify issues through alerts and dashboards. However, APM requires human intervention to address identified problems, which can be challenging when managing numerous performance signals. In contrast, APA systems, like Cast AI, automate responses to performance signals, offering real-time adjustments to Kubernetes workloads without manual input, thereby closing the operational gap left by APM. APA enhances efficiency by continuously rightsizing workloads, optimizing node usage, and managing Spot instance lifecycles, ultimately reducing costs and preventing issues like out-of-memory events. While APM remains essential for observability and compliance tasks, APA provides the necessary automation layer to act on performance data swiftly and effectively, ensuring optimal infrastructure management.

Agentic operations involve utilizing autonomous AI agents to manage Kubernetes infrastructure tasks such as detecting, diagnosing, and resolving issues without human intervention, significantly reducing the time from alert to resolution. This approach shifts the role of engineers from execution to approval, with AI agents handling routine operational work, including drift remediation, OOM prevention, and security tasks like CVE patching and RBAC drift detection. The use of SLO-driven automation allows agents to respond to error budget burn rates and correlated signals, acting proactively before user impact occurs. This model offers a reliable strategy that reduces mean time to resolution (MTTR), decreases cloud costs as a byproduct of optimized resource allocation, and addresses the operational complexity challenges highlighted by the CNCF 2023 Annual Survey. Cast AI's Application Performance Automation platform exemplifies agentic operations by integrating predictive model engines, agentic runbooks, and self-healing capabilities to provide a comprehensive system that improves application reliability and engineer productivity while ensuring compliance with security standards.

Agentic runbooks represent a significant advancement in automating Kubernetes operations by leveraging AI to observe cluster states continuously, make context-aware decisions, and execute multi-step recovery workflows without human intervention. Unlike traditional or automated runbooks, which require human involvement in executing predefined steps, agentic runbooks autonomously detect anomalies, determine appropriate remediations, apply fixes, and verify outcomes, thus minimizing operational overhead and closing the "alert-to-action gap." This closed-loop system enhances efficiency by handling known failure patterns, such as out-of-memory (OOM) events, Spot instance interruptions, and node consolidation, reducing the need for engineers to respond to alerts manually. Tools like Cast AI's Application Performance Automation platform implement agentic runbooks to optimize Kubernetes infrastructure by rightsizing workloads, managing Spot instances, and consolidating nodes, resulting in significant cost savings and operational efficiency. By continuously adapting to the cluster’s real-time state, agentic runbooks ensure that resources are used effectively, allowing engineers to focus on tasks requiring human judgment, ultimately transforming how Kubernetes environments are managed.

Application Performance Automation (APA) is a pioneering software category that autonomously connects real-time application performance signals to infrastructure actions, aiming to streamline operations for platform engineers and Site Reliability Engineers (SREs). Unlike Application Performance Monitoring (APM) which only observes and alerts, APA platforms, exemplified by Cast AI, autonomously act on performance data to rightsize resources, scale workloads predictively, consolidate nodes, and remediate anomalies while adhering to Service Level Objectives (SLOs). This automation reduces manual toil, enhances reliability, and achieves significant infrastructure cost reductions, as evidenced by Cast AI's platform trained on extensive data across major cloud providers. APA complements existing tools like APM and Kubernetes autoscaling by offering predictive, multi-cloud, and policy-driven solutions that operate beyond the capabilities of traditional monitoring and native Kubernetes tools.

Karpenter, a tool that improves node provisioning on Amazon EKS by reacting swiftly to pending pods and evaluating NodePool constraints, faces operational challenges in live production environments. Cast AI for Karpenter, now generally available, addresses these challenges by providing an intelligent layer that enhances visibility, optimization, and automation. This integration allows for continuous rebalancing of resources, accurate workload request alignment with real usage, and prevention of overprovisioning or interruptions, all without altering existing configurations. It enables teams to maintain optimized clusters, reduce manual work, and respond dynamically to changing demands by automating capacity adjustments and scaling decisions while preserving operational safety. Cast AI offers insights into costs and scale events and helps manage workloads effectively, providing a streamlined approach to cluster management without forcing changes to a team's existing processes.