ScaleOps CEO Yodar Shafrir on autonomous cloud orchestration for agentic AI
Key points
Key takeaways from a theCUBE interview at KubeCon + CloudNativeCon NA 2026 with Yodar Shafrir, founder and CEO of ScaleOps, on autonomous cloud-resource orchestration for AI workloads (November 2026):
- The problem: Four and a half years ago, teams struggled to allocate resources to microservices — too little causes reliability failures, too much wastes cloud spend; ScaleOps automates the decision end-to-end, claiming 70–80% infrastructure cost reduction.
- Agents change the economics: Agentic workloads sit idle then spike, load memory continuously and crash up to 80% more often than traditional microservices when under-provisioned, making static human-driven scaling obsolete.
- Context-aware orchestration: ScaleOps considers each application's connections, dependencies, latency and business need, deciding resources, replicas and node placement in real time.
- Trust journey: Customers install ScaleOps read-only on their hardest environments, watch its decisions, then run load tests with automation enabled before full rollout; the platform is self-hosted, keeping data on customer infrastructure.
- Agent cost efficiency: A new feature lets agents share idle resources, claiming 90% more cost-efficient agent workloads and up to 10x more agents on the same infrastructure, Kubernetes-native with no migration or SDK.
- GPU optimisation: For companies hosting their own models, ScaleOps distinguishes real-time inference from batch workloads to maximise time-to-first-token or throughput, claiming 60–70% GPU cost reduction.
- Customers: ScaleOps works with Adobe, Salesforce and Coinbase on critical production environments.
