Michael Levan👨🏻‍💻☕️

An SRE Agent needs 4 things: 1. A proper prompt/system message. This will give the direction for the Agent to take. 2. Agent Skills that are designed to turn your Agent into an SRE and observability expert. 3. A good LLM 4. Targets Number 4 is where the "magic happens". When you put together a solid Agent for a particular use case, the target can be anything in its speciality. For example, in an SRE Agent, the target could be anything from a cloud provider to a Kubernetes cluster. #agenticai #ai

Running Agents in production can't occur without three things. 1. Isolation 2. Governance 3. Scale In this podcast episode, Nathan Pearce and I break down how to accomplish said 3 things with kagent and agent-substrate #agenticai #ai #kubernetes youtube.com/watch?v=5PJRvzL7…

You forgot @cmux

“If I know what it’s going to be, it’s not going to be very innovative”

How certs work in kagent + Agent Substrate 👇 ✅ Actors Actors get certs from the internal CA with Substrate. Every time an Actor starts or resumes on a Worker, `atunnel` in that Worker Pod creates a private key and a signing request. The key never leaves `atunnel`. `ate-api` signs it with `actor-id-ca-pool`. The certificate is valid for one hour, and `atunnel` renews it with the same key. ✅ Workers Workers (Pods) are where Actors run. They certs from the PodCertificate API, which is new a API as of k8s v1.37. The PodCertificate API assigns X.509 TLS certificates to Pods automatically through projected volumes and the kubelet runtime Agent. It removes the need for external tools like cert-manager or sidecar proxies. #agenticai #ai #kubernetes

🤔 When does "sandboxing" not actually mean "sandboxing"? Sorta feels like "all of the time" is the appropriate answer at this point. The industry keeps talking about sandboxes and isolation, but when you dive into what people mean when they say it, it's almost like a "I put a workload in a container" situation. That's not a true, isolated sandbox. I break down what that actually means in my latest blog on @thenewstack thenewstack.io/how-to-sandbo… #agenticai #ai #kubernetes

EVERYTHING in your Agents depends on how secure ingress and egress is. - What an Agent can hit - What an Agent has access/permissions to - What can talk to your Agent And the 45324342 things that fall into those three categories. In this blog post, I break down securing the egress path (what your Agents have the ability to talk to, access, and reach) and how to observe/secure that traffic. cloudnativedeepdive.com/agen… #agentai #ai #kubernetes

We need to all work together with our Agents. I know there's a lot of talk in the industry right now on "don't read the code" and "turn on auto-mode", but here's the thing... Agents suffer from the same exact problem as humans do: 👉 They only know what they know based on the info they have. That's why, as the architect/engineer/whoever else, you need to know what the Agent is doing and confirm whether it's valid or not. #agenticai #ai #kubernetes

Looking forward to speaking at SREDay in a few weeks on Agent Substrate 🎉 I’ll be diving into how every organization should be thinking about true isolation for agentic workloads. #agenticai #ai #kubernetes

🎉 3rd speaking session at KubeCon officially locked in 🚀 I'll be at Kube Auto Day (co-located event) to talk about all things Agent Substrate. The session will be focused on scale, performance, and efficiency when using Substrate as your sandboxed platform on k8s. #agenticai #ai #kubernetes

There was a time where places like Twitter and LinkedIn were for people that wanted to learn. Genuine conversations, engineers sharing ideas, and a massive community of like-minded individuals. Now it's "let's take the buzziest thing at the moment and write a post about it with AI to get likes/comments for literally no reason at all". Lets bring back the old social media.

Protecting against lateral movements in the AI era? Think "Agentic Intrusion Detection". The ability to use AI for detection, rule creation, and investigation of your systems (both agentic systems and the systems/environment that have zero AI). I break this down in the blog post below. agent-shield.dev/blog/2026/1… #agenticai #ai #kubernetes

Awesome keynote at AGNTCon + MCPCon from Solos CPO. youtube.com/watch?v=iNV5-YkP… #agenticai #ai #kubernetes

Claude Desktop has definitely come a long way @AnthropicAI

Was diving into the differences between OpenSandbox and Agent Substrate. OpenSandbox gives the Agent an isolated computer to use. Your Agent then connects to it to run commands, connect to repos, read files, etc. Standard client/server architecture. Agent Substrate is where the Agent itself runs along with the isolated compute. Everything is under one roof. You then connect to the Agents running in Substrate with a runtime, like kagent, to interact with the Agent. The concern for me with OpenSandbox is performance. Sandboxes perform a "cold reboot" when you spin it back up, whereas with Substrate, Actors go into a "warm state" and resume incredibly quickly. `BatchSandbox` in OpenSandbox is slower by design, as the resume recreates the Pod where the sandbox lives from scratch, and the app has to re-initialize. Has anyone tried it? #agenticai #ai #kubernetes

I did one conference talk this week. 5 more to go for the rest of the year. I’m truly grateful to be able to do stuff like this. Who knew some incredibly poor kid with two parents who were addicts could do this.

There was a time in tech where people created businesses (Google, Apple, etc.) and they didn’t want to sell. They wanted to build something big. Now, I feel like the mindset is “let’s build something cool then hopefully sell”. I wonder why that mindset shift occurred

🎉Looks like I now have 2x speaking slots at KubeCon 🤓 PUMPED to be speaking at Agent Substrate day in a few weeks. My session will cover everything you need to know about Agent Substrate from top to bottom. #agenticai #ai #kubernetes

How does `atenet-router` and `atunnel` work in Agent Substrate for Actor traffic? 👇 Tldr; ✅ atenet-router Finds the Actor's Worker, resuming the Actor onto a free one if it's suspended, and forwards the request to the Worker. ✅ atunnel On the selected Worker, atunnel delivers it into the sandbox on the Worker to the one Actor running. If you're thinking about egress as well, atunnel also carries the Actor's outbound traffic to egress. #agenticai #ai #kubernetes

When thinking about securing lateral movements of compromised env, what's important? The answer will change for some, but the lower-level primitives/systems are typically top of mind. Here's an example of agentshield covering detection, investigation, graph mappings for Agent Substrate. Latest project released at Agentic AI Foundation AGNTCon + MCPCon yesterday. agent-shield.dev/ #agenticai #ai #kubernetes