←
SAIM SAFDAR10,860 posts
SAIM SAFDAR avatar

SAIM SAFDAR

@cloudnativeboy

🎤 Host of @cloudnativefm podcast

◷ Joined December 2013▧ 2,358 Photos and videos
SAIM SAFDAR avatar
SAIM SAFDAR@cloudnativeboy·Nov 16

Apple is making Docker Desktop optional for K8s on Mac. apple/container runs Linux containers on Apple Silicon. With Apple’s K8s support: - Start a cluster with one command - Connect with kubectl - Add worker nodes - Share images with the cluster github.com/apple/container/b…

SAIM SAFDAR avatar
SAIM SAFDAR@cloudnativeboy·Nov 16

OpenShell controls the agent. Kubernetes and Kyverno help control the environment around it. That's an interesting direction for the future of agentic DevOps. Worth exploring: cncf.io/blog/2026/11/11/open… (9/9)

SAIM SAFDAR avatar
SAIM SAFDAR@cloudnativeboy·Nov 16

AI agents may become the next operators of our development and production environments. But before we hand them the keys to our infrastructure, we need to build the guardrails. (8/8)

SAIM SAFDAR avatar
SAIM SAFDAR@cloudnativeboy·Nov 16

A Kubernetes NetworkPolicy might allow a Pod to connect to GitHub, but it cannot decide whether curl can read an issue while being prohibited from creating one. OpenShell introduces that finer-grained level of control. (8/8)

SAIM SAFDAR avatar
SAIM SAFDAR@cloudnativeboy·Nov 16

What caught my attention is its approach to agent governance: 🔹 Filesystem isolation 🔹 Process isolation 🔹 Network governance 🔹 Credential protection 🔹 Kubernetes integration (7/7)

SAIM SAFDAR avatar
SAIM SAFDAR@cloudnativeboy·Nov 16

This is where OpenShell becomes interesting. In his recent CNCF article, "OpenShell on Kubernetes: A Guide for Platform Engineers," @JimBugwadia explores how platform teams can pair OpenShell with @kyverno admission controls to enforce least privilege (6/6)

SAIM SAFDAR avatar
SAIM SAFDAR@cloudnativeboy·Nov 16

Sounds like the future of autonomous DevOps? But here's the challenge: How much access should we actually give these agents? (5/5)

SAIM SAFDAR avatar
SAIM SAFDAR@cloudnativeboy·Nov 16

→ Deploy applications and manage Kubernetes workloads. → Analyze logs, metrics and traces to troubleshoot production incidents. → Recommend and execute remediation actions. (4/4)

SAIM SAFDAR avatar
SAIM SAFDAR@cloudnativeboy·Nov 16

Imagine an AI agent that can: → Understand engineering requirements and write application code. → Generate Terraform configurations and provision infrastructure. → Interact with GitHub, CI/CD pipelines and container registries. (3/3)

SAIM SAFDAR avatar
SAIM SAFDAR@cloudnativeboy·Nov 16

What happens when AI agents move beyond writing code and start operating your infrastructure? We are moving towards a world where AI agents won't just assist developers. They will actively participate in the entire DevOps lifecycle. (2/2)

SAIM SAFDAR avatar
SAIM SAFDAR@cloudnativeboy·Nov 16

AI agents may become the next operators of our dev & prod env. But before we hand them the keys to our infra, we need to build the guardrails. OpenShell controls agent. K8s & @kyverno help control the env around it. (1/1)

Post attachment
SAIM SAFDAR avatar
SAIM SAFDAR@cloudnativeboy·Nov 15

It’s not an agent SDK, and it’s framework-agnostic. It hosts LangChain, ADK, Claude Code and MCP servers as-is. It’s Apache-2.0 and runs on standard K8s. The honest caveat: it targets CPU-bound agent work. There’s no GPU passthrough, and connections drop on suspend. (8/8)

SAIM SAFDAR avatar
SAIM SAFDAR@cloudnativeboy·Nov 15

And the security model is the part I like most. Every actor is treated as hostile: 🔒 Runs in a gVisor or microVM sandbox, never a plain container 🔒 Gets its own short-lived SPIFFE identity 🔒 Holds zero credentials. An egress gateway injects them outside the sandbox, (7/7)

SAIM SAFDAR avatar
SAIM SAFDAR@cloudnativeboy·Nov 15

The result: → The repo’s demo runs ~250 stateful actors on just 8 pods → 10–30x oversubscription, so you pay for compute only while an agent is thinking → 1,500+ activations per second (6/6)

SAIM SAFDAR avatar
SAIM SAFDAR@cloudnativeboy·Nov 15

When an agent goes idle, Substrate snapshots the Actor to storage and hands the room back to the pool. When a request arrives, any free Worker picks it up and resumes the session in under half a second. Same conversation, same files, same state, maybe a different room. (5/5)