Ephemeral Kuberentes environments with FluxCD

Platform Teams in the AI era

Various challenges:

  • Pipelines are hard to maintain
  • Shared staging environments are slow, flaky
  • Long review cycles
  • App teams want self-service but platform teams need governance
  • AI and Agentic coding make it easy to open a PR, but this needs good testing before merges

The setup

Three repos:

  • platform (flux operator conf, cluster bootstrap, infra, …)
  • appx (the end-user app)
  • charts (helm charts)

The plan

You create a PR, build-tag-push a docker image, then label your PR and the Flux Operator reads the PR. Then the Operator uses the output of the PR to create a new helm release. And then you’re done.

One cool thing though, they use Go templating with a different template char:

<< .Value.name >>

Creating a new PR does not update the previous ephemeral environment.

Oh and they have an MCP server. From here the talk evolved into an AI showcase.