Landing zones
A landing zone is the multi-account foundation every workload runs on: the top of the cloud's resource hierarchy, the accounts, subscriptions or projects under it, the network that connects them, the security services that watch them, identity for people and pipelines, and the state and delivery machinery that manages all of it. The BuiltForProd Baseline is a landing zone delivered as code in one repository per edition: acme-aws-platform-baseline, acme-azure-platform-baseline and acme-gcp-platform-baseline.
Why it matters in production
A single account, or an ad-hoc sprawl of them, has no isolation between environments, no clear blast radius and no governance model that grows with the company. Each cloud offers one hard boundary for access, billing and blast radius: the AWS account, the Azure subscription, the Google Cloud project. An access mistake, a leaked credential or a runaway workload in one does not reach another. One isolation unit per stage gives every environment its own access and billing boundary; one per shared concern (logs, security tooling, network, identity) keeps those concerns out of the units that run workloads and out of the root of the hierarchy.
How each edition implements it
Every edition has the same 15 roles. The core group holds the shared foundations: root, audit, security, identity, network, dns, artifacts, auto, corp, public and backup. The plat group holds one workload unit per stage: sandbox, dev, staging and prod. Every isolation unit receives the same baseline from one unit definition, and the guardrails attach above the units, where an administrator of a single unit cannot remove them.
| Aspect | AWS Enterprise Baseline | Azure Enterprise Baseline | GCP Enterprise Baseline |
|---|---|---|---|
| Top of the hierarchy | One AWS Organization | Management group mg-acme under the Tenant Root Group | The Google Cloud organization node |
| Grouping | Organizational units core and plat | Management groups mg-acme-core and mg-acme-plat | Folders core and plat, and a folder backup of its own |
| Isolation unit | 15 accounts | 15 subscriptions | 15 projects |
| Per-unit baseline | units/account-baseline | units/subscription-baseline | units/project-baseline |
| Where guardrails attach | Service control policies on the organization | Azure Policy at mg-acme | Organization policies and an IAM deny policy at the organization node |
| Global versus regional | global/ folders, units named acme-glob-<stage>-* | global/ folders, units named acme-glob-<stage>-* | global/ folders for organization-wide units |
In every edition a unit's folder tells you which account, subscription or project it deploys to and whether it exists once per organization (global/) or once per region. Global units deploy to the home region.
AWS Enterprise Baseline
The organizations unit in environments/core/root/global creates one AWS Organization with two organizational units (OUs) and fifteen accounts: the management account plus fourteen members.
| Account | Responsibility |
|---|---|
core-root (management, 111111111111) | The Organization, OUs, service control policies and tag policy; the state bucket; the organization CloudTrail; delegation of security. |
core-security, core-audit, core-identity | Delegated security services and remediation; the audit bucket and Config aggregator; IAM Identity Center. |
core-network, core-dns | IPAM, the hub VPC, the Transit Gateway, private DNS and Client VPN; public Route 53 zones. |
core-artifacts, core-auto | Container registries with immutable tags; GitHub OIDC federation and the CI roles: the Baseline's deployer and reader, and one per blueprint repository and stage. |
core-corp, core-public | Reserved for internal tools and intentionally public assets; baselined only. |
core-backup | The organization's backup target: locked vaults that hold the copies of the workload accounts' recovery points; no workloads and no VPC. |
plat-sandbox, plat-dev, plat-staging, plat-prod | One workload account per stage, identical in shape, each with a spoke VPC attached to the Transit Gateway. |
Every account receives units/account-baseline: default EBS encryption, an account-wide S3 public-access block, AWS Config recording, the acme-terraform-deployer role CI assumes to write and the read-only acme-terraform-reader role Baseline plans and drift checks assume. Core accounts get it through stacks/includes/account-common, workload accounts through stacks/plat-account.
The management account is not subject to service control policies and is the highest-value target, so it holds Organizations, billing and state only. Every security and governance service is delegated to the account that owns it: GuardDuty, Security Hub, Inspector and Macie to core-security, AWS Config to core-audit, IAM Identity Center to core-identity. The offerings note describes the result as conceptually similar to Cloud Posse architectures and AWS Control Tower Account Factory for Terraform, delivered as a turnkey customized product. The AWS Enterprise Baseline overview has the full account table.
Azure Enterprise Baseline
The organizations unit creates the management group mg-acme under the Tenant Root Group, with mg-acme-core and mg-acme-plat below it: three management groups in all.
The 15 subscriptions are acme-core-* and acme-plat-*. Each is either created through a billing scope or adopted from the subscription_ids map when your organization creates them first. acme-core-root sits directly under mg-acme, outside the two groups, and holds the tenant-level units and the state storage account.
Every subscription receives units/subscription-baseline: explicit resource-provider registration, the Microsoft Defender for Cloud plans and security contact, the Activity Log export to the audit workspace and storage, Network Watcher, and a baseline resource group with a CanNotDelete lock. Core subscriptions get it through stacks/includes/subscription-common, workload subscriptions through stacks/plat-subscription.
Azure Policy is assigned at mg-acme, so no subscription Owner can remove it. The security services are tenant-wide and administered through roles at mg-acme: Security Admin for the Platform Leads (eligible through Privileged Identity Management when it is switched on) and Security Reader for the auditors, so no delegation step exists. Inside a subscription, every unit owns one resource group, <prefix>-<unit>-rg, which is the scope of its locks, its role assignments and its cost view.
GCP Enterprise Baseline
The organizations unit creates the folders core, plat and backup under the organization node and the 15 projects acme-core-* (11) and acme-plat-* (4). Ten core projects sit in core; acme-core-backup, the protected backup vault, sits alone in backup, so access granted on core does not reach it. The seed project acme-core-root exists before OpenTofu runs, holds the state bucket, and is adopted and moved into core.
Project IDs are globally unique and permanent: a deleted project's ID can never be used again. project_id_suffix in common.hcl handles an ID that is already taken and is fixed once used. Every core project and the staging and prod projects carry deletion_policy = "PREVENT"; sandbox and dev carry DELETE, which a switch turns into PREVENT as well.
Every project receives units/project-baseline: the APIs from one list, a de-privileged default compute service account, the deployer sa-acme-terraform-deployer (Owner of its own project only), the _Default log bucket retention, OS Login, Essential Contacts and an optional budget alert. Core projects get it through stacks/includes/project-common, workload projects through stacks/plat-project.
The organization policies and the IAM deny policy are set once at the organization node and inherited by every folder and project. Units whose resources sit above one project, such as folders, organization policies, the Shared VPC host and the aggregated sinks, are organization-scoped: 15 unit definitions carry a marker file org-scoped.hcl and run as the Baseline's CI service account instead of a project deployer. Security Command Center and the aggregated log sinks work at the organization, so no per-project enrollment exists.
Naming
| Thing | AWS Enterprise Baseline | Azure Enterprise Baseline | GCP Enterprise Baseline |
|---|---|---|---|
| Isolation unit | acme-core-network, acme-plat-prod | acme-core-network, acme-plat-prod | acme-core-network, acme-plat-prod |
| Stage | core: the name; plat: prd, stg, dev, sbx | core: the name; plat: prd, stg, dev, sbx | core: the name; plat: prd, stg, dev, sbx |
| Resource prefix | acme-usw2-prd | acme-eus2-prd | acme-usw1-prd |
| Root email | plus-addressing: aws+acme-core-audit@company.com | none | none |
Worked example: adding a stage
On the GCP edition, a new workload stage such as qa touches four places, and no unit definition changes:
- A
qaentry in theplat_projectsmap of theorganizationsmodule, with its stage slug and deletion policy. The slug is a value of thestagesecure tag, which holdsprd,stg,dev,sbxandcore, so a new slug also extends the tag values and the map's validation. - The stage in exactly one isolation domain in
network.hclisolation_domains,prodornonprod. - A
/16for the stage innetwork_map.yaml, inside the plat/13, which holds eight, with its subnets and ranges. environments/plat/qa/with anaccount.hcl, a region folder withregion.hcl, and a stack file that instantiatesstacks/plat-project.
The project ID acme-plat-qa is permanent once the organizations unit creates it. See hub-and-spoke networking for what the new stage connects to and least privilege for who can reach it. On the AWS and Azure editions the same change is an entry in the plat_accounts or plat_subscriptions map, an address-plan entry, an isolation domain, and a stage folder that instantiates stacks/plat-account or stacks/plat-subscription.
Common mistakes
- Running workloads in the root unit. The AWS management account, the Azure
acme-core-rootsubscription and the GCP seed project hold the hierarchy and the state; keep workloads out of them. - Copying a global unit into a second region. Global units exist once per organization; a copy means a duplicate or a conflict error. They live in
global/. - Reusing a closed account's email (AWS). AWS keeps a closed account suspended for 90 days and its email cannot be reused for a new account.
- Assuming a deleted project ID can be reused (GCP). It cannot; choose
namespaceandproject_id_suffixbefore the first apply. - Making a bucket or blob container public outside
core-public. On AWS, automated remediation re-blocks public buckets everywhere else. On Azure, the guardrail initiative denies public storage access, and onlyacme-core-publicholds theacme-allow-public-blobsexemption. On GCP, thestorage.publicAccessPreventionconstraint refuses public access everywhere butacme-core-public.