Human Oversight

This is the single most important principle behind how AI operates on the Gabaltech Platform, so we state it plainly:

AI suggests. Humans approve.

Critical infrastructure changes require review

Any AI-generated Terraform, Pulumi or Kubernetes change goes through the normal review path — a diff a person can read, a plan a person can inspect, before it's applied.

Production deployments require approval

Deployments to production environments are gated behind explicit human approval in our internal developer platform. AI can prepare a deployment; it cannot approve its own request.

AI cannot independently take destructive production actions

Actions like deleting infrastructure, dropping data, or force-pushing to a protected branch are not actions an AI-assisted workflow can take without a configured, human-in-the-loop approval step. Where we've built AI-assisted automation, destructive operations are the specific case we've been most careful to gate.

What this looks like in practice

• A developer asks an AI assistant to draft a new ECS service definition — the assistant proposes Terraform, the developer reviews and applies it.
• A designer generates infrastructure diagrams and blueprints from a description — the resulting Terragrunt units still go through plan before apply, same as anything hand-written.
• A deployment to a production environment, whether triggered by a human or assisted by AI, sits behind an approval gate held by a person in the approvers group.