# Shan Valleru's Technical Blog

> Technical blog by Shan Valleru covering Kubernetes, distributed systems, AI, and business strategy.

For full article content, see: https://svalle.ru/llms-full.txt

## About the Author

Shan Valleru is a Software Engineer and Wharton MBA. This blog covers Kubernetes and distributed systems, AI and emergence, and business strategy.

## Topics Covered

- Kubernetes operators and controllers
- Kubernetes scheduler internals
- GPU scheduling and Dynamic Resource Allocation (DRA)
- WebAssembly (Wasm) on Kubernetes
- Kubernetes garbage collection (OwnerReferences, finalizers)
- Leader election in distributed controllers
- CPU throttling and cgroups performance
- Multi-cluster Kubernetes with KubeFleet
- Kubernetes networking (iptables, overlay networks, Services)
- CoreDNS and NodeLocal DNSCache
- eBPF and Cilium for kube-proxy replacement
- client-go patterns (Informers, work queues, rate limiting)
- etcd and watch mechanics
- Admission webhooks and policy engines (Kyverno)
- Container internals (namespaces, cgroups, containerd, runc)
- Linux primitives for containers
- Go concurrency patterns
- Python parallelism and multiprocessing
- Production debugging and observability
- ArgoCD at scale
- Infrastructure economics and depreciation
- Platform engineering ROI and investment
- Open source total cost of ownership
- Infrastructure as growth constraint
- Scaling costs and unit economics
- Multi-cloud strategy and portfolio theory
- Platform bundling and microeconomics
- AI/LLM industry competitive analysis, pricing strategy, bundling
- Running AI/LLM inference on Kubernetes: GPU scheduling mismatches, model loading cold starts, token-aware routing, autoscaling limitations
- AI-native infrastructure: Gateway API Inference Extension, llm-d, Kueue, KAI Scheduler, Ray Serve, agent orchestration
- LLM pricing microeconomics, versioning, price discrimination, commoditization vs differentiation
- OpenAI vs Anthropic vs Google pricing comparison, ChatGPT Go, Claude Code costs
- AI bundling strategy, subscription economics, API pricing trends
- Exploration vs exploitation, innovation management
- Disruptive innovation, creative destruction, Christensen, Schumpeter
- Agentic coding, AI agents, Claude Code, developer productivity
- Building AI agents from scratch: the agent loop, request/response vs agentic loop, tool calling, context engineering, agent memory, evals, guardrails, non-determinism, OpenAI Agents SDK and Claude Agent SDK
- The one pattern behind vertical AI startups: intent extraction and routing, context assembly, bounded agent loops vs open ReAct, policy gates and guardrails, sandboxed execution and browser agents, verification, human handoff, trace streams and evals, RL environments, sovereign AI, and the startup landscape per component (Vapi, ElevenLabs, Deepgram, LangGraph, E2B, Browserbase, Langfuse, Pinecone)
- Sandboxes as the compute primitive for AI agents: microVM isolation lineages (Firecracker, gVisor, Kata), snapshot-restore and fork, cold starts, four demand curves (inference, RL training, evals, serving AI-generated apps), BYOC and sovereign AI, browser sandboxes, pricing compute by the attempt (E2B, Modal, Daytona, Steel, Browserbase)
- Emergence, complex systems, consciousness, phase transitions
- Scaling laws, Chinchilla, compute-optimal training, AI economics
- Adverse selection, technical debt economics
- Consumer surplus, developer productivity measurement
- Developer productivity measurement, throughput accounting, activity vs progress metrics
- Lean thinking, Toyota Production System, platform waste
- Inventory theory, cloud capacity planning, FinOps
- Infinite games, James Carse, infrastructure maintenance vs building, platform team burnout
- Portfolio theory for team composition, skill diversification, hiring strategy
- Exogenous vs endogenous change, transformation theater, top-down vs bottom-up innovation, punctuated equilibrium, organizational decoupling, skunkworks

## Articles

### Kubernetes

- [Signs You're Operating Kubernetes at Scale](https://svalle.ru/posts/kubernetes/scale-symptoms/)
- [Scaling Beyond 5,000 Nodes Per Cluster: Anatomy of Kubernetes Limits and Multi-Cluster Architecture](https://svalle.ru/posts/kubernetes/scaling-beyond-5k-nodes/)
- [Building a Kubernetes Operator from Scratch](https://svalle.ru/posts/kubernetes/building-operator-from-scratch/)
- [Bottlenecks in Large-Scale Multi-Cluster Kubernetes](https://svalle.ru/posts/kubernetes/multi-cluster-bottlenecks/)
- [Eventual Consistency and Stale Caches in Kubernetes Controllers](https://svalle.ru/posts/kubernetes/stale-cache-controllers/)
- [Kyverno at Scale: Multi-Cluster Policy Without the Pain](https://svalle.ru/posts/kubernetes/kyverno-multi-cluster/)
- [Admission Webhooks at Scale: Diagnosis, Hardening, and Multi-Cluster Consistency](https://svalle.ru/posts/kubernetes/admission-webhooks-at-scale/)
- [From etcd to Watch: How Kubernetes Watches Actually Work](https://svalle.ru/posts/kubernetes/etcd-to-watch/)
- [Kubernetes Networking Demystified: Tracing the Magic (and Debugging the Nightmare)](https://svalle.ru/posts/kubernetes/kubernetes-networking-demystified/)
- [From kubelet to Process: How Kubernetes Actually Runs Your Container](https://svalle.ru/posts/kubernetes/kubelet-to-process/)
- [Beyond kube-proxy: eBPF Service Routing in Kubernetes](https://svalle.ru/posts/kubernetes/beyond-kube-proxy-ebpf/)
- [client-go Patterns: Informers, Work Queues, and Rate Limiting](https://svalle.ru/posts/kubernetes/client-go-patterns/)
- [Garbage Collection in Kubernetes: OwnerReferences and Finalizers](https://svalle.ru/posts/kubernetes/garbage-collection-finalizers/)
- [How the Kubernetes Scheduler Actually Works](https://svalle.ru/posts/kubernetes/kubernetes-scheduler-deep-dive/)
- [CPU Throttling in Kubernetes: Why Your Limits Are Lying to You](https://svalle.ru/posts/kubernetes/cpu-throttling/)
- [Leader Election in Kubernetes Controllers](https://svalle.ru/posts/kubernetes/leader-election/)
- [CoreDNS Under Pressure: How We Fixed DNS Bottlenecks with NodeLocal DNSCache](https://svalle.ru/posts/kubernetes/coredns-nodelocal-cache/)
- [WebAssembly on Kubernetes: The Next Evolution Beyond Containers](https://svalle.ru/posts/kubernetes/wasm-on-kubernetes/)
- [GPU Scheduling in Kubernetes: From Device Plugins to Dynamic Resource Allocation](https://svalle.ru/posts/kubernetes/gpu-scheduling-dra/)
- [Running AI on Kubernetes: What Breaks and What's Being Built](https://svalle.ru/posts/kubernetes/running-ai-on-kubernetes/)
- [Why Your Kubernetes Cluster is a Depreciating Asset](https://svalle.ru/posts/kubernetes/cluster-depreciation/)

### Business & Strategy

- [Portfolio Theory for Managing Engineering Teams](https://svalle.ru/posts/business/portfolio-theory-for-teams/)
- [Activity vs Progress: The Visibility Trap in Engineering Productivity](https://svalle.ru/posts/business/activity-vs-progress/)
- [Lean Thinking for Platform Engineering: Finding the Muda in Your Platform](https://svalle.ru/posts/business/lean-platform-engineering/)
- [Adverse Selection in Technical Debt: Why Bad Code Drives Out Good](https://svalle.ru/posts/business/adverse-selection-technical-debt/)
- [Consumer Surplus and Developer Productivity: The Hidden Value of Your Platform](https://svalle.ru/posts/business/consumer-surplus-developer-productivity/)
- [Inventory Theory for Compute Capacity: How Much Buffer Should You Hold?](https://svalle.ru/posts/business/inventory-theory-cloud-capacity/)
- [Commodity or Moat? Three Bets on the Future of LLMs](https://svalle.ru/posts/business/llm-pricing-microeconomics/)
- [A Flashlight in a Hurricane: What Disruption Frameworks Don't Cover](https://svalle.ru/posts/business/flashlight-in-a-hurricane/)
- [Exploration vs Exploitation: The Hidden Cost of Cutting Innovation Slack](https://svalle.ru/posts/business/exploration-vs-exploitation/)
- [Porter's Five Forces for LLM Builders—And Why It's Not Enough](https://svalle.ru/posts/business/porter-five-forces-llm/)
- [Bundling vs Unbundling Platform Services: A Microeconomic Framework](https://svalle.ru/posts/business/bundling-unbundling-platform/)
- [Multi-Cloud: Hedging Strategy or Wasted Optionality?](https://svalle.ru/posts/business/multi-cloud-hedging/)
- [Platform & Infra Teams: Cost Center or Profit Enabler?](https://svalle.ru/posts/business/platform-infra-cost-center/)
- [Scaling Infrastructure ≠ Scaling Costs: Economies, Step Functions, and Leverage Points](https://svalle.ru/posts/business/scaling-costs/)
- [The J-Curve of Platform Investment: Why Platform Teams Feel Expensive Before They Pay Off](https://svalle.ru/posts/business/platform-j-curve/)
- [The True Cost of Free Open Source](https://svalle.ru/posts/business/true-cost-of-open-source/)
- [Infrastructure as Growth Constraint: When Systems Become the Bottleneck to Revenue](https://svalle.ru/posts/business/infrastructure-growth-constraint/)
- [The Infinite Game of Infrastructure](https://svalle.ru/posts/business/infinite-game-of-infrastructure/)
- [Imposed vs Chosen: Why Most Transformations Are Theater](https://svalle.ru/posts/business/exogenous-endogenous-change/)

### AI

- [Scaling Laws: What They Are and Why They Matter](https://svalle.ru/posts/ai/scaling-laws/)
- [How This Blog Gets Written](https://svalle.ru/posts/ai/how-this-blog-gets-written/)
- [Emergent Phenomena: From Ants to Transformers](https://svalle.ru/posts/ai/emergent-phenomena/)
- [Agentic Coding: From Single Agents to Agent Teams](https://svalle.ru/posts/ai/agentic-coding-patterns/)
- [Agent Building for Software Engineers: It's Mostly a While Loop](https://svalle.ru/posts/ai/agent-building-for-engineers/)
- [One Pattern Behind Every Vertical AI Startup](https://svalle.ru/posts/ai/anatomy-of-a-vertical-ai-agent/)
- [Sandboxes Are the New Servers](https://svalle.ru/posts/ai/sandboxes-are-the-new-servers/)

### Go

- [Go Concurrency: Beyond Goroutines and Channels](https://svalle.ru/posts/go/async-programming/)

### Linux

- [What Is a Container, Really? Namespaces, Cgroups, and the Linux Primitives Behind Docker](https://svalle.ru/posts/linux/what-is-a-container/)

### Python

- [Python Parallelism: The GIL, Multiprocessing, and When Each Matters](https://svalle.ru/posts/python/multiprocessing-idioms/)

## Optional

- [About the Author](https://svalle.ru/about/)
- [Books & Reading](https://svalle.ru/books-etc/)
- [Photo Galleries](https://svalle.ru/photos/) — Travel photography from Japan, London, Tibet, Vancouver, and paragliding

## Contact

- Website: https://svalle.ru
- Twitter: https://twitter.com/svalleru
- GitHub: https://github.com/svalleru
- LinkedIn: https://www.linkedin.com/in/svalleru/

## Location

All articles are available at https://svalle.ru/posts/