Kubernetes has revolutionized cloud infrastructure—but ask any platform or DevOps engineer and they’ll tell you it still takes too much time, YAML, and guesswork to operate. Mirantis is aiming to change that with Lens Prism, the first embedded AI assistant for Kubernetes, now built into the popular Lens IDE.

“With Lens Prism, we’re not just reducing complexity—we’re helping developers surface real insights and solve problems faster,” said Kyle Wheeler, General Manager for Lens at Mirantis, in a recent TFiR interview.

Unlike external copilots or ChatGPT-style wrappers, Lens Prism is integrated directly into the Lens UI. That means it’s context-aware, cluster-aware, and understands what’s happening in your environment. You can ask natural questions like “Why is my pod crashing?” and get not only explanations but suggested commands to resolve issues—without handing over control.

“We’re taking developers from commands to conversations,” said Wheeler. “And it’s all about giving them insight, not just output.”

To avoid overreach, Prism starts with a read-only posture. It doesn’t make changes itself but provides copy-pasteable commands developers can verify and run. Future iterations may support write actions—but always under user control.

Behind the scenes, Lens Prism uses a combination of pre-prompting, contextual awareness, and integration with LLMs like OpenAI and Azure. Mirantis has also added flexibility for organizations to inject custom prompts or guardrails—whether that’s corporate policy enforcement or even something as simple as language preferences.

The release also includes an AWS EKS integration that simplifies cluster onboarding. Instead of manually exporting kubeconfigs, users can SSO into their AWS accounts and instantly see their EKS clusters reflected in Lens—with auto-removal when permissions change.

“As clusters are added or removed from your AWS profile, they automatically appear or disappear in Lens,” Wheeler said.

Looking ahead, the team envisions more automation—such as scheduled reports or even auto-remediation. But the ethos remains clear: AI augments the operator, it doesn’t replace them. “We don’t think AI will take your job,” Wheeler added. “But someone who uses AI well might.”

For teams ready to try it, Lens Prism is included in the Pro, Plus, and Enterprise tiers of Lens—with a free trial mechanism in the works.

TRANSCRIPT

Swapnil Bhartiya: Kubernetes, as everybody knows, is one of the most amazing technologies, similar to the Linux kernel, which has transformed how we build and run modern infrastructure. But let’s be honest, it’s still notoriously complex to operate. Most engineers spend more time deciphering YAML and hunting down logs than actually delivering value, and that is the challenge that is becoming critical—to let developers and engineers do what they do best: create value for their organizations.

And that’s where Lens Prism comes in. It is the first embedded AI assistant for Kubernetes built directly into the Lens IDE platform. It is designed to streamline operations, surface insights, and support real-time decision making. Joining me today is Kyle Wheeler, General Manager of Lens at Mirantis, to talk about why they built Prism, what it can do for platform and DevOps teams, and how it fits into the broader evolution of AI-powered infrastructure that Mirantis is helping build. Kyle, it’s great to have you on the show.

Kyle Wheeler: Hey, great to be here. Thanks for having me.

Swapnil Bhartiya: It’s my pleasure. We have been covering Lens since it was acquired by Mirantis from the early days, but it has evolved over time, and today we are going to talk about Lens Prism. Let’s start with the core of it. Our audience knows about Lens—if you want, you can do a quick intro of Lens as well—but I’m going to focus more on what exactly is Lens Prism? What does it do? And how does it help Kubernetes users who are dealing with a lot of complexity?

Kyle Wheeler: As you may know, Lens is one of the most popular, if not the most popular, IDE for Kubernetes and cloud-native development. With that, we provide contextual awareness to your clusters, visibility to logs, reporting, etc., in a very user-friendly and beautiful UI. So we’re taking you from essentially the command line and terminal into something that’s graphically easy to understand, again reducing the complexity, especially when dealing with complex infrastructure.

Swapnil Bhartiya: Can you also talk about AI? I mean, at KubeCon and before that, Mirantis folks are doing a lot of work in the AI space. Talk a bit about the reason for embedding an AI system in something like Lens. What specific problem are you trying to solve in terms of reducing complexity in the cloud-native environment, and how does it further complement what Lens IDE already does?

Kyle Wheeler: So just quickly, what Lens Prism is—it’s our AI assistant inside of Lens. Following Lens’s value proposition and vision of reducing complexity, we looked at solutions and ways to really fight through the noise and give actionable data and insights to developers.

When we looked at different ways of implementing AI solutions, we found that putting this at the fingertips of developers takes it beyond just entering commands to make fixes, but gives you actionable insights and lets you have conversations with this AI assistant.

What we found is that our sweet spot with developers is going to be new developers to the Kubernetes space that maybe don’t know all the commands they can run, or don’t know how to fix problems. So if you have an embedded AI assistant inside of Lens, you get surfaced information a lot faster, and then you get insights and action items of where you should go to fix this, or how to fix this—and even commands to tell you what to fix.

Swapnil Bhartiya: When we talk about AI assistants, we’re talking about developers interacting with AI. With LLMs or generative AI, the prompts are key. It’s more like if you don’t have the right question, you won’t get the right answer. So how do you enable them? Do they need to know specific things to ask, or how does it become easy for their workflow?

Kyle Wheeler: Our strategy has been to have a middle layer of pre-prompting that we do ourselves, to give that context to the LLM—whether that’s OpenAI, Azure (those are the two main supported ones right now), or any local LLM that follows the OpenAI-compatible format.

We’ve tried to take the guesswork out of that as much as we can, but we’ve also given it access to everything that you see in Lens, so it’s context-aware. When you’re in a cluster, it understands that you’re talking about this cluster, this namespace, or this pod.

We’ve given some insights on the backend with pre-prompting, but we’ve also given examples of what can be asked. We expect this to evolve over time, but it’s as simple as the same question that a developer asks right now: “What’s wrong with my pods?” or “What’s wrong with this cluster?” You can start there and then start to see the information come through. So again, taking you from commands to conversations—if you have leadership asking you, “Hey, what’s wrong with this piece of infrastructure?” you can ask that same question.

Swapnil Bhartiya: We all know that Kubernetes, as powerful as it is, is also very complicated. How far does Lens Prism go to reduce that operational friction without hiding too much, without becoming too opinionated, without taking too much control and flexibility that users would want from Kubernetes?

Kyle Wheeler: Guardrails have always been really important to us at Lens. That starts with our principles of following role-based access controls and giving users only permission to see what they already have permission to see outside of that—through kubeconfig files and the tokens required to have access to that.

In the same way, Lens Prism only sees what you as an individual developer have access to. While it could be more powerful in the future, we’ve actually limited capabilities to only give access from a read perspective. So it’s understanding information, it’s not running around rampant. It sees what’s happening with your clusters and then just gives the developer insight, giving more control back to the developer to go and fix anything that’s wrong.

While it might surface a command for you, it’s not going to run that command for you. You would have to go and copy that command and paste it into a terminal session. But it’s at the user’s desire and control.

Swapnil Bhartiya: That’s very important to understand that AI is not in the driving seat—you are still the operator. It’s just a tool. When I was listening to you, you talked about pre-prompts already, but how does this AI assistant learn over time? Is it like whatever you ship it with is the only knowledge it has, or is it also context-aware? Since everyone’s environment is different, as teams use it more within a specific cluster, does it learn from that and get more fine-tuned and optimized for that specific organization?

Kyle Wheeler: I think it starts wherever the LLM is, but then what it has access to in terms of the desktop application—it has some contextual awareness. We go a little bit further than that by also having the ability for individuals and companies that adopt it to go in and put in specific rules or requirements for the AI.

That’s as simple as, if you want it to speak only to you in German, for example, it can do that. But if there are other more technical-related items that you want to pre-prompt it with, you can do that as well as an individual user. That can be at the organizational level. So if your company has some things that they want to codify—”this is how we do things”—then that’s what Lens Prism would abide by.

In the future, what we’re looking at is—because most of our users come to Lens not just for creating another terminal session, but for visibility and seeing these beautiful reports and navigating through different information pieces inside of Lens—we’re looking at hard-coded links. So as Lens Prism surfaces information, it would also have a link to that specific log, for example, and you can go and see that in context inside of Lens, or go into a specific dashboard, connecting the assistant with the UI.

Swapnil Bhartiya: How do you see it evolve beyond what you’re starting? When we look at cloud and Kubernetes, we talk about cost optimization, policy enforcement—there are geopolitical things changing, so we have to focus on heavily regulated industries. Then there’s observability, predictive diagnostics. How do you see Lens Prism evolve beyond what you have started right now?

Kyle Wheeler: What’s interesting about what we see in the future is that we can already do some of this today, but because of our understanding—even our internal cloud developers that have been using Lens Prism for some time—we said, okay, there’s a resistance to letting AI run wild. So we put in very specific guardrails on the front end.

But we’ve been in some demo environments where we’ve removed some of the guardrails to allow it to not just run commands for read access, but to actually change things on the cluster, implement code itself. What that looks like and how adoption of that looks—obviously, if I’m in the driver’s seat, I want to be in the driver’s seat. I don’t want someone going and doing the work for me. We like control.

So we’re trying to balance those two ends of “this could make your life a little bit easier” but at the same time, there’s some risk involved. Right now, our stance is we’re just going to surface the information, and you can take and do what you want with it. You can review the commands before they run.

But in the future, that might look like something like, “Hey, set up a daily report and just tell me what’s going on with my clusters, and also go ahead and fix those things that are wrong with them.” So it could get pretty exciting pretty fast. But again, we’ve got to balance that with company policies, individual users’ resistance to letting AI run wild.

With everything that we do, we start with this control plane of “these are the boundaries that AI can live in inside of my environment.” That can be on an individual developer standpoint—we can just not have it do any writing commands and only read—or we can open it up a little bit more. But that would be up to the user.

Swapnil Bhartiya: Every time we talk about AI these days—I was talking to someone at the Open Source Summit, and they said it’s the first time where tech people are getting scared about things, because we have been taking other people’s jobs in a way with tech, but now because of ChatGPT and generative AI, it’s our own jobs. They mentioned it’s at least the first time we are tasting our own medicine. When we look at generative AI embedding in Lens Prism, is it enabling and augmenting the team? Is it empowering them? Or is there also suspicion that it will take away jobs, as every time a new technology comes in?

Kyle Wheeler: I think this goes into the philosophy of where AI sits and where we’re at in this AI wave. We’ve had these discussions internally that I don’t think AI will take your job, but someone that uses AI to its full potential will take your job.

This is a tool that really starts with insights. It’s AI for insights. It lets you learn faster, more efficiently, and understand what’s going on with your clusters in a much more efficient way than just going and searching through logs, in the same way that Lens does that for a developer.

Instead of having to run commands and search, find logs and sift through them, you can quickly identify where these logs are coming from, point to specific pods, clusters, and navigate between them. In the same way, Lens Prism allows for faster insights right now. As far as what it can do in the future, that’s going to be up to the developers and how they want to use it.

Swapnil Bhartiya: Can you talk about what else is new in this release?

Kyle Wheeler: One other thing that came out in this release with Lens is that we’ve created a new connection mechanism to AWS EKS clusters. As we’ve seen, there’s a lot of complexity, and again, back to our value proposition of simplifying the complex. The traditional method is going and getting kubeconfig files, and a lot of times that requires some effort from the developer.

What we’ve done is created a connection with AWS, allowing users to go in and sign in with a single sign-on, and whatever is in their profile—any number of clusters that are in their profile—are automatically inputted into Lens. While it’s just a technical connection, we expect that to also grow in our ability to connect to other things inside of AWS infrastructure, and we also expect to expand this to other cloud providers.

The nice thing about it is that as a cluster is removed from your profile in AWS due to permissions or some tech policy change in departments, the cluster is also removed from visibility inside of Lens. So it stays up to date. And as clusters are added to your AWS profile, they’re also added into Lens. So you have up-to-date cluster information, simplified connection time, and then you have this AI assistant on top of this that can go in and find out all the information for you about what’s going on to find greater efficiencies in your infrastructure.

Swapnil Bhartiya: Let’s talk about the teams who want to get started, who want to try Lens Prism. How can they access it? Is it part of Lens? Is it something separate? Is there a trial version? How can they get started with Lens Prism?

Kyle Wheeler: Lens Prism is, while we talk about it as a standalone product because we do think it changes the way users will interact with not just Lens but with Kubernetes clusters in general, it is just part of Lens. Right now, it is included in the Pro, Plus or Enterprise subscriptions.

So if you have access to and already have a license for Lens in one of those tiers, you will have access to Lens Prism and the AWS connection, along with some other features that make navigating your clusters a little bit easier.

If you’d like to get started today, obviously go to k8slens.dev, download the latest version. We are working on a trial mechanism to allow users to get started quickly with this, see the value, and if they’d like to purchase, they obviously can. And again, it’s included with the standard Lens pricing. There’s been no pricing change with the addition of Lens Prism or the AWS connection. So we tried to keep this as simple as possible.

Swapnil Bhartiya: Kyle, thank you so much for joining me today and giving us an update on Lens and Lens Prism. As you rightly mentioned, the scope may go beyond, and sometimes that’s what happens with open source technologies. You start to solve one problem, and then it expands beyond what you initially thought. So that also means we’ll talk a lot about Lens Prism in the future. But I really appreciate your time today. Thank you.

Kyle Wheeler: Thank you so much for having me. Really appreciate it.