AI coding agents are quickly moving off the developer’s laptop and into the enterprise, and platform teams are being asked tough questions about where those agents run, what they can reach, and who is paying for them. Ahead of KubeCon + CloudNativeCon North America 2026 in Salt Lake City, VMblog spoke with Nicky Pike, Field CTO at Coder, about how the company provides secure infrastructure for AI-powered development. Pike explains how self-hosted, Terraform-based workspaces give developers and agents the same consistent, centrally managed environments, with model credentials kept in the control plane, network policy enforced where the code runs, and every action tied back to the person who started it.
In this Q&A, Pike also shares results from a U.S. defense intelligence platform team and DekaBank, offers the metrics he believes leaders should track (including token cost per story point), and previews what attendees can expect at Booth 195, from live demos of Coder Workspaces, Coder Agents, and AI Gateway to the new Coder Academy. He also outlines Coder’s conference-week schedule, which includes the AI SRE Meetup, a community-building session from Marco Martinez, a Gene Kim book signing, and a talk on governed AI agents at the AWS booth. His central message to platform teams is that your agents need a crew chief.
++
VMblog: What technical and operational bottlenecks does Coder address for cloud-native development teams, and how are those challenges changing as AI coding agents become part of everyday software delivery?
Nicky Pike: Let’s be honest, most of what slows cloud-native teams down isn’t new. It still takes days to get a developer productive on a big codebase, and “works on my machine” is alive and well. AI just turned the volume up.
Every developer now has three or four AI interns working for them. They’re stupid smart and never need a rest, and developers kick off several at once on jobs that run an hour or more. On a laptop, the fan’s screaming, the work stops when the lid closes, and those agents run with the developer’s credentials and full network access. Meanwhile, platform and security teams can’t answer simple questions: Which models are we using, what’s it costing us, and what did that agent just touch?
Coder enables the AI operating layer for enterprises: self-hosted development infrastructure that gives developers and their agents consistent, centrally managed workspaces on infrastructure you control.
VMblog: Where does Coder fit within a cloud-native platform engineering stack? How do its self-hosted workspaces and Terraform-based templates connect developers and AI coding agents to existing infrastructure and tools?
Pike: Think of Coder as a highway with guardrails: a fast, paved road inside your own cloud or data center so nobody’s tempted to go off-road with tools nobody approved. We don’t run your models. We take care of where the code runs.
Platform teams define workspaces as Terraform templates, and Coder provisions them on Kubernetes, VMs, or whatever cloud or data center you run. Every developer and every agent gets the same code, dependencies, and rules, whether you’ve got three developers or 3,000, without assembling the code themselves or filing a ticket.
VMblog: What changes when an AI coding agent needs an environment to write, run, and test code? How does Coder help platform teams provision those environments, manage their lifecycle, and support collaboration between developers and agents?
Pike: Once an agent stops suggesting code and starts running it, it needs what a developer needs: the repo, dependencies, a place to test, and access to the right systems. A laptop doesn’t scale for that, and it hands the agent the developer’s credentials.
With Coder Agents, each job runs in its own isolated workspace on your infrastructure, built from the same templates platform teams already use for developers. Model API keys never land in the workspace, every action is tied to the person who kicked it off, and idle workspaces shut themselves off.
Because developers and agents share the same environment, the handoff is clean. When the agent’s done, a developer can jump into that exact workspace and pick it up. For platform teams, it’s the same job they already do, just more of it.
VMblog: AI coding agents can execute commands and access sensitive code and services. How does Coder help organizations control credentials, model access, and agent network activity and maintain an audit trail without adding friction for developers?
Pike: Agent tools change every few months. If your security plan depends on approving one tool and banning the rest, you’ll be redoing it by spring. Let developers pick their tools, and govern the environment where the code runs instead.
Coder is self-hosted and air-gapped, if you need it, so code and agent activity stay on your network. Model credentials stay in the control plane, not the workspace. AI Gateway manages model access, attributes prompts, tool use, and spend for each user, and enforces budgets before the bill appears. Network policy in the workspace template controls what every agent and process can reach.
Developers don’t feel that as friction. The guardrails are already there when they open a workspace. The approved way to use AI has to be the easiest way.
VMblog: Enterprises already have IDEs, AI coding tools, and cloud infrastructure. What does Coder add, and how does it help organizations preserve their choice of tools, models, and deployment environments as they scale adoption?
Pike: Most teams think they’ve got two options: a locked-down laptop or somebody else’s cloud sandbox where they’re the renter waiting on the landlord. Coder runs on infrastructure they already own. Own the house, and you get control, choice, and consistency.
Choice is the big one. We don’t pick your tools or models: Claude Code, Cursor, Gemini, open-weight models on your own hardware, or whatever ships next month. Lots of companies already bought Claude Code or Cursor, but security won’t let those cloud agents execute on somebody else’s machines. With Coder’s Agent Relay, now in preview with select customers, developers keep the interface they know, the provider handles orchestration and inference, and execution happens in a Coder workspace on your infrastructure.
Most customers get there in three steps: Migrate off the laptop, modernize with AI governance, and then multiply by running agents side by side, each one isolated and logged. That last step is where the real productivity jump is.
VMblog: What measurable results are customers seeing from Coder, and which metrics should leaders use to assess developer productivity, infrastructure efficiency, and the value of AI coding workflows?
Pike: I like sharing results from a U.S. defense intelligence platform team. Before Coder, every 20- or 30-person mission team had to stand up its own Kubernetes cluster, wire up its own security controls, and go through its own Authority to Operate (ATO). That’s months of work before anybody writes a line of mission code.
This team put Coder in as one multi-tenant platform. The ATO gets done once, and every team inherits it. New teams get a pre-approved environment on day one, GPUs included. Today, that’s roughly 2,500 developers across multiple mission groups, and they’re exploring how to bring AI, including Amazon Bedrock, onto that same governed platform.
On the commercial side, look at DekaBank, one of Germany’s systemically critical banks. It runs Coder on its own on-premises Kubernetes cluster. New developers now write and deploy code on day one instead of waiting two to three weeks, builds dropped from six or seven minutes to under one, and waiting time from technical issues fell 60%. Next, it’s getting ready to pilot Claude Code on that same governed platform, and it’ll measure token cost per story point, exactly the kind of metric leaders should be watching.
For metrics, look past lines of code or AI license counts. Start with time to first commit, workspace startup time, utilization, and cost per workspace. As agents take on more work, add cycle time, defect rates, and cost per completed task, tokens included. Tokens are a CFO problem now.
VMblog: Looking toward 2027, how do you expect platform engineering to evolve as teams support both developers and AI coding agents? Which priorities are shaping Coder’s direction?
Pike: Once agents leave the laptop, somebody has to answer three questions: Where do they run, what can they get to, and who’s paying for it? Platform teams already do this for developers with templates, policy, and access. Covering agents is the same job with a wider scope.
That’s the idea behind the AI operating layer: Govern the environment where AI runs, not the individual tool. It shapes our priorities, including model choice, including the self-hosted open-weight models Coder Agents already supports; Agent Relay, so the cloud agents teams already bought can execute on their own infrastructure; and AI Gateway for model access, audit, and spend at scale.
Developer experience stays at the center. If the governed path isn’t the easiest path, people go off-road. Platform teams that make it the easiest path will turn AI pilots into how their company actually builds software.
VMblog: What can attendees see or try at Coder’s KubeCon booth, and what practical lesson should they take back to their platform teams?
Pike: Booth 195. We’ll have engineers there, so bring your hard questions.
- Coder Workspaces: A full dev environment on Kubernetes, ready in a couple of minutes.
- Coder Agents: Hand a job to an agent, and watch it work in its own isolated workspace, with every model call and action tied back to the person who started it.
- AI Gateway and workspace network policy: The rules and the record, live, so you see what your security team would see.
- Coder Academy: Brand-new, free, hands-on courses on cloud development environments, open to anybody, whatever your job is.
The lesson: Your agents need a crew chief. When I was turning wrenches on C-141s, nothing flew until the crew chief signed off. The pilots picked the mission, but somebody owned that airplane and answered for it. Let developers pick their tools and models, put the controls where the code runs, and let platform teams be the crew chief, building on the Kubernetes and dev environments you’ve already paid for.
VMblog: Beyond the booth presence, what’s your broader community engagement strategy for KubeCon NA 2026? How are you contributing to the cloud-native ecosystem’s growth?
Pike: For us, KubeCon is a chance to share what we’ve learned, not just run a booth. My teammate Marco Martinez, who built our developer community and co-hosts the [Dev]olution podcast with me, is giving a talk called “Building a Developer Community That the Business Can’t Ignore.” He’ll cover what worked and what didn’t across Discord, GitHub, meetups, and content and how to tie community to business results without turning it into a sales pitch. That’s how we contribute: sharing lessons openly, warts and all.
VMblog: With Monday, November 9 dedicated to co-located events like Cloud Native AI + Inference Day, ArgoCon, BackstageCon, CiliumCon, and WasmCon, are you participating in any of these, and why did you choose them?
Pike: While Coder is not formally sponsoring a co-located event this year, we’re participating in the broader programming surrounding KubeCon.
- AI SRE Meetup – KubeCon Edition | Monday, November 9, 5:30–7:30 PM MT
Dave Ahr will join SREs, platform engineers, and AI infrastructure practitioners to discuss perspectives on operating AI workloads reliably and securely. As AI agents move from experimentation into enterprise environments, platform and infrastructure teams are increasingly responsible for providing the environments, access controls, and operational visibility these workloads require. - Building a Developer Community That the Business Can’t Ignore | Wednesday, November 11, 11:00–11:30 AM MT
Marco Martinez, Coder’s head of community, will share lessons on building an authentic developer community, creating meaningful engagement, and demonstrating the strategic value of community within an organization. - Governed AI Agents on Kubernetes: Safe Autonomous Code Execution at Scale | AWS Booth
Jiachen Jiang will explore how organizations can securely execute autonomous AI coding agents on Kubernetes while maintaining the governance, isolation, and control required for enterprise adoption.
We chose these opportunities because that’s where the real conversation is. Once agents move from experiments into production, platform and infrastructure teams own the environments, access, and visibility those agents need.
VMblog: Which conference tracks, keynote themes, or technical topics are you most excited about at KubeCon NA 2026? Where do you see alignment with your strategic vision?
Pike: Four tracks line up with what we hear from customers: AI Inference + Agentic, Application Development + Delivery, Platform Engineering, and Security:
- AI Inference + Agentic. You’ll hear a lot about inference, GPUs, and models on the floor. That’s a different layer. We take care of where agents’ code runs now that developers hand off several long-running tasks at a time.
- Application Development + Delivery. The goal hasn’t changed: Get developers into a working environment fast. Now that environment has to work for agents too.
- Platform Engineering. Somebody has to decide where agents run, what they can reach, and how their work gets tracked. Platform teams already do that for developers.
- Security. Identity, credentials, multi-tenancy, and zero trust all apply to agents. Since tools change constantly, teams should enforce isolation, network access, and audit where the code runs, not tool by tool.
VMblog: Are you organizing any exclusive customer events, partner meetings, or community gatherings during conference week in Salt Lake City? How can interested attendees connect with your team outside the booth — and what prize drawings, giveaways, or engagement activities have you planned?
Pike: Come find us at Booth 195. I’ll be there most of the week with our engineers, so bring your hard questions. You’ll see live demos of Coder Workspaces, Coder Agents, and AI Gateway, and you can check out the new Coder Academy.
A few more ways to connect:
- AI SRE Meetup | 5:30-7:30 PM MT on Monday, Nov. 9, with Dave Ahr.
- Marco Martinez’s session | 11-11:30 AM MT on Wednesday, Nov. 11.
- Gene Kim book signing | 1-3 PM MT on Wednesday, Nov. 11, Booth 195: Get there early, as those go fast. Signed copies while supplies last.
- Jiachen Jiang at the AWS booth: governed AI agent execution on Kubernetes.
If you want a deeper conversation, stop by the booth, and we’ll set up time during the week.
