Self-Hosted AI Development Infrastructure for the Enterprise with Coder
13m
Coder's Jiachen Jiang, a staff product manager focused on AI governance, opened by explaining how a developer-workspace company became infrastructure for AI coding agents. Coder started in 2017, when two high school students who built and sold Minecraft servers together needed a better way to collaborate; they created Code-Server, an open-source project that grew into the company's foundation. Coder now has about 150 customers, more than 90 of them in financial services, other regulated industries, and the federal government. Jiang said two things have held constant since the start: Coder still maintains a free open-source product, and it avoids locking customers into any cloud provider, model, or ecosystem. Customers can host Coder on their own infrastructure or in the cloud of their choice and use any model, a flexibility Jiang called the main reason regulated enterprises choose it. The product's core is the workspace, a developer environment defined in Terraform.
Jiang said two events in early 2025 changed Coder's direction: Anthropic launched Claude Code, which introduced a new generation of autonomous, long-running coding agents, and Anthropic became a Coder customer. Anthropic recognized early that agents running on a developer's laptop inherit all of that developer's access, so they need isolated environments limited to what they should reach. Workspaces, once a quality-of-life improvement for human developers, became a requirement for agents. Jiang added that strict approved-tool lists fail because AI tools change too quickly and developers find workarounds, which leads to unmonitored shadow AI. Coder's alternative is to control the environment instead, governing what data tools can reach, which models they use, and how information moves, with a central view of what every tool is doing.
Jiang then walked through Coder's product portfolio, which connects back to workspaces. AI Gateway is a model proxy that sits between the model and the agent running in the environment, so administrators can see all AI traffic and apply policies, including traffic that doesn't run through a workspace. Coder Agents is a native agent harness that runs in Coder's control plane and spins up workspaces as scratch space it can share with others. Agent Relay, built in partnership with Cursor and Anthropic, lets their cloud-managed agents run code in workspaces on the customer's own infrastructure. Customers can also run any agent through its command line interface (CLI) inside a workspace. Jiang described one customer that skips Coder's UI and builds its own on Coder's API to provision environments for thousands of developers, and quantitative investment firms that give workspaces to analysts using AI on sensitive work. Jiang argued that Coder, unlike most governance tools, reduces friction because platform teams can offer developers a choice between a 12- or 13-page approval process for a coding agent or a ready-made workspace. Jiang closed the overview by noting that AI governance is now the top reason new customers come to Coder.
Presented by Jiachen Jiang, Staff Product Manager, Coder, and Dave Ahr, Developer Relations Engineer, Coder. Recorded live at KubeCon North America 20026 in Salt Lake City, UT, on November 10, 2026. Watch the entire presentation at https://techfieldday.com/appearance/coder-presents-at-tech-field-day-at-kubecon-north-america-2026/ or visit https://techfieldday.com/event/kubecon26/ or https://coder.com/ for more information.