Kubernetes Isn't Outdated. It's Evolving. And So Should You.
There's a narrative I keep hearing in tech communities across LATAM and beyond: "Kubernetes is too complex now. It had its moment."
I think that narrative confuses two very different things: the accidental complexity we engineers have built on top of it, and the obsolescence of the platform itself. Kubernetes isn't trending or out of trend. It's infrastructure. And infrastructure doesn't go anywhere.
What is changing — fast — is the role we need to play as engineers around it.
"We won the scalability war. But we are losing the simplicity battle."
The Problem Nobody Wants to Admit
The original DevOps promise was simple: "You build it, you run it." It was a philosophy of ownership. But as the CNCF ecosystem grew from 10 to over 200 projects, that philosophy became a cognitive load trap.
Today, in many teams, a backend developer is expected to be an expert in:
- Business logic — their actual job
- Containers and Kubernetes — manifests, Helm charts, CRDs
- Network security — NetworkPolicies, mTLS, Cilium
- Observability — OpenTelemetry, Prometheus, Grafana Loki
- Compliance & Governance — OPA, Kyverno, RBAC policies
- Continuous Delivery — ArgoCD, Flux, GitOps workflows
That is an absurd stack of responsibilities for a single person. The result: developers become bottlenecks, and delivery speed collapses.
The problem isn't Kubernetes. The problem is that nobody built the floor between Kubernetes and the developer.
Platform Engineering and the Golden Path
Platform Engineering didn't emerge to "run servers" on behalf of developers. It emerged to build an Internal Developer Platform (IDP) — a product that absorbs infrastructure complexity and delivers a self-service interface.
The goal is the Golden Path: a set of automated, pre-approved, secure, and observable-by-default workflows. If you follow the Golden Path, everything just works. If you deviate, you need to justify it.
💡 What does a 2026 self-service workflow actually look like?
A developer opens their internal portal and requests a new microservice. Within 5 minutes they have: a dev environment deployed on K8s, an isolated database provisioned via Crossplane, traces and logs enabled with OpenTelemetry, and GitOps sync active against their repo. Zero infra tickets. Zero manual YAML.
Your CNCF Toolbox for 2026
If you're a Platform Engineer, the CNCF is your toolbox. Here's how you connect the dots to build your software factory:
| Component | Key Technologies | Why It Matters |
|---|---|---|
| The Portal (Frontend) | Backstage | The "storefront" of your platform. A central hub for software catalogs and self-service templates. |
| Infra as Code / API | Crossplane | Manage cloud resources (RDS, S3, GCS) using the Kubernetes API. Turns your infra into K8s objects. |
| Networking & Security | Cilium / Istio | Fast, visible, and secure networking (mTLS) from day one. |
| Unified Observability | OpenTelemetry | A common language for metrics, logs, and traces. Essential for explainability. |
| Automated Policy | Kyverno / OPA | Governance as code. Blocks insecure deployments before they ever hit production. |
| GitOps Delivery | ArgoCD / Flux | The heartbeat of the platform. Git as the single source of truth. |
The Next Challenge: The Agentic Platform
This is where the conversation gets interesting — and where most Platform Engineers aren't looking yet.
This week, during KubeCon EU Amsterdam, the CNCF published a standards paper for agentic AI in cloud native environments. The document — produced by the AI TCG — doesn't talk about models or data science. It talks about how Kubernetes must orchestrate autonomous agents securely, observably, and with governance from day one.
📄 Reference · CNCF AI TCG — Cloud Native Agentic Standards (March 2026)
The paper defines four pillars for agentic systems on K8s: General (containers), Control & Communication (MCP, A2A, SPIFFE), Observability (OTel + token metrics + agent traces), and Governance + Security (agent identity, JIT access, tenancy isolation).
Read the full document at cncf.io.
The protocols emerging as standards — MCP (from Anthropic, now donated to the Agentic AI Foundation) and A2A (from Google for peer-to-peer agent communication) — are already vocabulary that a 2026 Platform Engineer needs to understand.
Why does this matter to Platform Engineering? Because AI Agents are not a separate layer living "in some vendor's cloud." They are workloads. They run in containers. They need ephemeral identities (SPIFFE/SVID), namespace isolation, NetworkPolicies, and observability of LLM inference tokens.
⚠️ What nobody is telling you
If your IDP today has no identity model for agents, if your observability pipelines don't track LLM tokens, if your Kyverno policies don't account for MCP tool access — you already have technical debt in the 2026 stack. This isn't science fiction. KubeRay and Volcano are already optimizing how K8s handles these workloads.
Kubernetes Isn't Going Anywhere. You Need to Grow.
The message here isn't that DevOps is dead. It's that it matured. Platform Engineering is the next logical chapter of the same story: less manual heroism, more self-governing systems.
Kubernetes is the operating system of modern infrastructure. Nobody says Linux "went out of style." The same should be true of Kubernetes. What changes is the level of abstraction from which we operate. And for those of us building platforms, that is exactly the challenge that makes this work fascinating.
Next year, the most mature platforms won't just be the ones with the best DX for human developers. They'll be the ones capable of orchestrating autonomous developers — agents — with the same rigor, security, and observability we apply today to any microservice.
Are you building that platform, or are you still debugging YAML?
Want to continue this conversation in person? On April 18th we're hosting KCD Guadalajara 2026 at the Holiday Inn Guadalajara Expo. Platform Engineering, agentic AI, cloud native security, and the entire CNCF community of Mexico — all in one place.
👉 community.cncf.io/kcd-guadalajara-2026
#kubernetes #platformengineering #cncf #cloudnative #devops #mcp #agenticai #idp #kcdguadalajara