Why Kubernetes Complexity Does Not Equal Cloud Native

▶ Watch (0:13)

Munjal opened by separating two terms the industry often conflates. Kubernetes-native means your app speaks Kubernetes: deployments, services, ingress, CRDs, operators. Cloud-native is broader and, per CNCF, lacks a rigid definition. At Groupon, 423 microservices run across AWS and GCP. That footprint makes even thinking about a cloud migration uncomfortable. The point is not which cloud you pick. The underlying stack is effectively the same on GKE and EKS. What actually matters is cost.

The Real Price of Running Kubernetes at Scale

▶ Watch (0:50)

Groupon runs GKE Standard, not Autopilot, managing roughly 400 nodes, 10,000 pods on average, and up to 50,000 at peak. Munjal said the team spends a high number of engineering hours just keeping Kubernetes running, what he called the “keep the lights on” tax. Ingress has been deprecated for three years. Nobody has migrated to Gateway API because backlogs, migrations, and peak traffic windows eat all available time. Adding more Helm charts and YAML files does not improve developer SDLC cycle time.

Benchmark Results: 3x Code for Kubernetes, Slower on Every Metric

▶ Watch (11:52)

Munjal built four versions of a Pac-Man game, Kubernetes-native, cloud-native, AI-native, and non-cloud-native, then ran 12 benchmarks using Claude to analyze all four codebases. Kubernetes-native produced 3x the lines of code and 3x the total files compared to non-cloud-native. Startup time was slower. Requests per second were lower. YAML, Docker, and configuration files were significantly higher. The non-cloud-native version was not production-ready, but the cloud-native middle path solved real problems without the Kubernetes overhead.

The CNCF Tool Chain as a Self-Reinforcing Problem Set

▶ Watch (16:24)

The CNCF ecosystem started with one infrastructure problem and generated a chain of new ones: key management, secrets, config maps, storage, security, CI/CD. Each new problem produced a new tool and a new vendor. Search “cost optimization for cloud” and you find roughly 100 companies scraping the same Prometheus metrics with a thin agent. Munjal’s framing: code is a liability, not an asset. More code means more scanning, more maintenance, more security surface. AI agents make this pressure sharper, not lighter.

Platform Engineering and the Developer Distance Problem

▶ Watch (21:38)

Munjal argued that developers should own infrastructure directly, not through a portal. Platform teams create ticket-ops friction: developers know nothing about CPU requests, topology spread constraints, or resource limits, while platform engineers know nothing about the specific microservice. Tools like encore.dev let developers manage infrastructure in TypeScript, removing the handoff layer entirely. His rule: if your platform team has more people than your feature team, priorities are inverted. Building a portal does not fix a cultural problem.

Notable Quotes

i personally believe Kubernetes is not meant for everybody and anybody Prerit Munjal · ▶ 09:54

lines of code for Kubernetes, it’s 3x the times that non-cloud native way works Prerit Munjal · ▶ 27:05

I I feel like these days code is not an asset. It is an liability. The more you have, the more you are in trouble. Prerit Munjal · ▶ 19:51

if your feature team has less number of people than platform teams, you are doing it totally wrong. Prerit Munjal · ▶ 21:43

Key Takeaways

  • Kubernetes-native code costs 3x more lines and files than non-cloud-native equivalents.
  • Portability arguments collapse when vendor commits unlock significant enterprise discounts.
  • Developers owning infrastructure directly removes ticket-ops friction that portals cannot fix.

About the Speaker(s)

Prerit Munjal is a Cloud-Native Platform Leader and Senior Technical Product Manager at Groupon. As the former CTO of KubeCloud, he built no-code solutions bridging Cloud, DevOps, and SRE, leading the company to a successful acquisition. He manages CI/CD, observability, cloud, Kubernetes, and database infrastructure at Groupon, and reaches roughly 100,000 followers with 20 million yearly impressions through YouTube and LinkedIn content.