S5 Bonus: Slashing the Cloud Bill in Half: Automated Kubernetes Optimization with Leon Kuperman of CAST.ai

Episode Summary

Executive Summary: Leon Cooperman, co-founder/CTO of cast.ai, explains how repeated cloud-bill pain led him to build an autonomous Kubernetes optimization platform that reduces cost without manual DevOps work. The conversation covers the company’s pivot from multi-cloud Kubernetes to EKS-first, its product philosophy of radical 10x value, the engineering culture behind rapid delivery, and lessons on hiring, scale, and frugal fundraising.

Main Topics: Origin of cast.ai from cloud cost pain: Leon describes recurring frustration with rising AWS bills at prior startups and at Oracle, where resource allocation was opaque and optimization required manual intervention. This became the motivation for an autonomous cloud optimization product. Autonomous Kubernetes optimization as the product thesis: cast.ai focuses narrowly on Kubernetes and uses first principles to automatically rebalance workloads, optimize cost and performance, and avoid manual recommendation-based tooling that customers often never act on. MVP design and early technical strategy: The team started with cloud economics benchmarking and even built a custom Kubernetes distribution/multi-cloud concept as the first MVP, aiming to understand compute cost and placement choices across providers. Engineering culture and operational discipline: Leon emphasizes blameless postmortems, strong CI/CD, no QA/DevOps silos, full engineer ownership, extensive automated testing, and frequent production deployment as core to reliability and speed. Product pivot and customer validation: The company initially believed multi-cloud was the future, but customer feedback showed most users wanted help inside existing EKS/GKE/AKS environments. This led to a pivot to Amazon EKS and a low-friction savings-estimator hook. Hiring strategy and team composition: cast.ai built its team in Lithuania, hired Go-oriented engineers, and looked for entrepreneurial drive. Leon values people who want to learn fast and eventually build their own companies. Scale challenges and future roadmap: The platform now faces data/storage and combinatorial optimization challenges as it samples clusters frequently. Future expansion includes more Kubernetes platforms and adjacent domains like security, storage, compliance, and disaster recovery.

Key Arguments: Customers usually ask for incremental improvements, not the disruptive leap they actually need, so product vision must lead customer interviews. For cloud cost optimization to matter, it must be fully autonomous; recommendation-only products are too manual to deliver real savings. Kubernetes was a deliberate focus because the problem space is defined enough to let the system act automatically and effectively. Building with production discipline early prevents technical debt from compounding later, even for a tiny team. A strong product hook must deliver immediate value in seconds, or prospects will not stay engaged long enough to appreciate deeper features. The initial multi-cloud thesis was ahead of market readiness, so the company pivoted to meet customers where they already run workloads. Rapid, small production releases reduce risk more than large, infrequent batches because rollback is easier and impact is smaller. Frugal early financing helps founders avoid excessive dilution before product-market fit is proven.

Data Points: Cloud bill growth at prior startup: from a couple thousand dollars/month to close to $200,000/month - Leon used this experience as a major motivation for cast.ai Cloud bill increase multiple: ~100x growth - Describing the AWS spend trajectory at his previous startup Team size at early engineering stage: 5 or 6 people - cast.ai’s early team that established robust engineering processes Production delivery cadence: 10 to 15 changes/day; about 30 changes/day at the time of recording - Leon cites frequent automated deployments as evidence of CI/CD maturity Rollback requirement: within 30 minutes - Operational rule for every production change Production incident rollback example: 26 minutes - Time from deployment to impact to rollback in a specific incident Production scenario test interval: every 5 minutes - Full end-to-end cluster lifecycle tests run in production Cluster snapshot frequency: every 15 seconds - cast.ai takes frequent snapshots for analysis and optimization Data size per snapshot: a few megabytes - JSON blob produced per cluster snapshot Potential placement permutations: 10^16 possible combinations - Combinatorial scale of pod-to-node placement decisions MVP customer value hook: one-line script - Creates a read-only agent and quickly reports waste and recommendations Kubernetes market focus: EKS first, with AKS, Rancher, OpenShift planned - Current and future platform expansion strategy Cash/cloud decision tradeoff: multi-cloud pivot - Customer feedback showed they wanted in-place optimization rather than a startup-run control plane

Pivotal Quotes: "If you ask customers what they want, they will ask you for a better mousetrap." — Leon Cooperman: Explaining why product vision must go beyond incremental customer requests "Everything that we do has to be a fully automated or autonomous approach." — Leon Cooperman: Defining cast.ai’s core product philosophy "We didn't start the product management process by going and interviewing a bunch of customers... There needed to be some kind of radical shift in thinking that would create a 10x benefit." — Leon Cooperman: Describing the company’s approach to finding a disruptive wedge rather than a minor feature

Implications: The episode underscores that infrastructure startups win by delivering immediate, autonomous value and by narrowing scope to a domain where automation can truly act. It also suggests early customer discovery should test bold hypotheses, not just gather feature requests.