AI coding agents are moving off developer laptops and into the enterprise. That shift raises hard questions for platform teams. Where do agents run? What can they reach? Who controls the data, and who pays for the tokens? Many organizations are also rethinking whether AI workloads belong in a hyperscaler or on infrastructure they control.
Coder Developer Relations Engineer Dave Ahr examines that challenge with Alan Shimel at KubeCon 2026 in Salt Lake City. Ahr came to Coder after years at HashiCorp and VMware.
From Remote Workspaces to Agents
Ahr explains that Coder began as a way to build workspaces developers could connect to and shut down when done. Adoption was slow at first, since developers rarely want to give up their laptops. Then large customers asked Coder to put agents inside those workspaces.
Coder is self-hosted rather than SaaS. That appeals to regulated industries and government agencies that need to control their own data. Ahr sees the same pull toward local control in how companies think about IP, AI and token spend.
Hardware Isolation on AWS
AWS released Nitro Sandboxes during KubeCon week. These hardware-isolated microVMs give each workload its own boundary. Ahr spent the week building a joint demo with AWS on how Coder fits.
Coder builds its workspaces from Terraform templates. That lets teams orchestrate the full environment, including network policies and pod identity. Ahr says the result is a workspace that can be locked down as tightly as needed. It works for human users, automated processes and agents alike.
Platform Teams Own the Bubble
Ahr argues the problem sits with platform engineering, not security or developers. He sees two shifts underway. First, organizations need isolated workspaces that put AI in a bubble. Second, networking is becoming a priority again.
Teams now have to consider where agents reach and how large the blast radius could be. Data loss can happen without any malicious intent. The goal is to contain AI coding agents without slowing adoption. Ahr also points to non-developers building apps of their own, a group many companies struggle to support.
Up Next in KubeCon 2026 Salt Lake City
-
Data Management Should Build Down Fro...
Models Come and Go. Data Stays.
Agentic data management is quickly becoming one of the hardest problems in enterprise AI. Models can be rented, swapped and replaced each year. The data underneath them is where long-term value lives. Agents also generate far more queries than humans ever did. Mach...