Skip to content
On this page 10

Control Tower vs a code-owned landing zone

AWS Control Tower gives you the account and guardrail layer at no license fee. Here is where a code-owned foundation wins, and where it honestly does not.

AWS Control Tower is a managed service that sets up and governs a multi-account AWS environment: an organization with dedicated log archive and audit accounts, an account vending mechanism, and a library of controls applied to organizational units. A code-owned landing zone produces the same outcome as infrastructure code in your own repositories, reviewed and applied by your own pipeline. Both are legitimate. They differ in who holds the definition.

This post compares them honestly, including the cases where Control Tower is the better answer and the cases where migrating would be a waste of a quarter. Where a claim depends on a feature set that AWS revises frequently, it is labeled as something to verify rather than asserted.

What Control Tower and AFT give you

#

Credit where it is due. The account and guardrail layer is genuinely hard, and Control Tower does a real job of it:

  • An organization with the separation already drawn. A management account, a dedicated log archive and an audit account, with organizational units above them. This is the structure most teams would eventually arrive at, delivered on day one.
  • Account vending. Account Factory creates accounts to a template rather than by hand. This is the single most important property of any estate that intends to grow, and it is the one most self-built foundations lack. We argue that point in multi-account strategy for enterprises.
  • A controls library. Preventive, detective and proactive controls you select rather than author, applied at the organizational unit. For a team with no policy-authoring experience, choosing from a list beats writing from a blank file.
  • Landing-zone drift detection and enrollment. It notices when its own settings change, and existing accounts can be brought under management rather than rebuilt.
  • It is an AWS service. Supported through your AWS support contract, familiar to your AWS account team, and uncontroversial in a procurement review. That is worth more than engineers usually admit.
  • No license fee. You pay for the AWS services it enables. That is not the same as free, but it is a very different starting position from any purchased alternative.

Account Factory for Terraform adds a Terraform-based pipeline for vending and customizing accounts, which narrows the gap for teams that want their account definitions in code. It is also another pipeline to operate, and that cost is real.

Where a code-owned foundation wins

#

Four properties, each concrete rather than philosophical.

Drift visibility across the whole estate

#

Landing-zone drift detection tells you when the landing zone’s own configuration has changed. It does not tell you that someone widened a security group in a workload account, attached an inline policy to a deploy role, or changed a retention setting on a log group.

A foundation defined entirely as code can be planned against reality on a schedule, and every difference between code and cloud becomes a finding with a diff attached. In the AWS Enterprise Baseline, that runs daily at 01:00 UTC for critical accounts and weekly across all 14, and it opens a GitHub issue rather than reverting anything — because automatic reversion can turn a discrepancy into an outage. The reasoning is in drift is a fact, not a failure.

Reviewability

#

When the foundation is code, a change to it is a pull request with a plan attached. The reviewer sees the effect, not the intent: these three resources change, this policy gains this statement, this route disappears.

That single property does three jobs at once. It is the change control your auditor asks about, it is the mechanism that stops an unreviewed change reaching production, and it is how a second engineer learns the system. A change made in a console is reviewed by nobody and teaches nobody.

Portability

#

A definition in plain OpenTofu or Terraform can be read by any engineer you hire, moved between organizations, forked for a subsidiary, or used as the reference for a different cloud’s edition of the same standard. State that lives inside a managed service goes where that service goes.

This matters most in two moments that are easy to forget during evaluation: an acquisition, and a change of platform lead.

Customization without fighting the model

#

Every real estate has requirements the standard model did not anticipate: a network topology dictated by a regulator, a particular key custody arrangement, an unusual region combination, a workload that needs a bespoke exception. In code, that is a conditional and a comment. Inside a managed model, it is a customization pipeline layered on top of a service whose behavior you do not control.

The comparison that decides most evaluations

#

The real difference is not quality. It is scope. Here are the layers a production platform actually needs, and who supplies each.

LayerControl Tower + AFTCode-owned foundation
Organization, organizational units, core accountsYesYes
Account vendingYes, Account FactoryYes, if a factory was built
Preventive and detective controlsYes, from a libraryYes, authored by you and reviewed as code
Centralized logging and audit accountYesYes
Network topology, address plan, isolation domainsNot in scopeYes
Identity model and permission setsPartly, via IAM Identity Center setupYes, defined in code
Pipeline trust and deploy identityNot in scopeYes
Secrets management patternNot in scopeYes
Compliance control mapping and evidencePartly, via detective controlsYes, as conformance packs and evidence procedures
Estate-wide drift detectionLanding-zone settingsEvery resource defined in code
Workload architectureNot in scopeYes, where blueprints are included
Documentation and runbooks for the aboveAWS service documentationWhatever you or your vendor writes

Nobody should read that table as “Control Tower is deficient.” It is scoped, and it does what it is scoped to do. The mistake is in the business case, not the service: teams approve a plan in which Control Tower is the landing zone, and then discover that the network design, identity model, pipeline trust, secrets pattern and documentation are still theirs to build. That is the bulk of the work, and it is the work we put engineer-month ranges against in the true cost of building your own landing zone.

Figure 1
Platform layers covered by a managed landing zone and by a code-owned foundation

Two stacks side by side. The left stack shows a managed landing-zone service covering organization structure, account vending, a controls library and centralized logging, with the layers above it marked as not in scope. The right stack shows a code-owned foundation covering the same base layers plus network, identity, pipeline trust, secrets, compliance evidence and documentation in one repository.

managed landing-zone servicecode-owned foundationdocumentationsecrets · pipeline trustnetwork · identity modelabove the line: your team, a partnercontrols libraryaccount vending · loggingorganization · core accountsdocumentationsecrets · pipeline trustnetwork · identity modelguardrails · compliance packsaccount factory · loggingorganization · core accountsone repository, one review process
The same platform, split by who owns each layer. The managed service covers the base reliably; the layers above it are supplied by your team, a partner or a delivered foundation, whichever path you choose.

Where Control Tower is the better choice

#

This is not a courtesy paragraph. There are estates where a managed landing zone is the right call and a code-owned foundation would be an expensive indulgence:

  • You have no infrastructure-as-code practice yet. A code-owned foundation assumes pull requests, reviews, a pipeline and people who read plans. Without that culture, code is just a second place for configuration to rot.
  • You need governance to be an AWS service. Some organizations genuinely need the answer “it is AWS’s managed service, covered by our support agreement” for an internal risk review. That answer has value and a code-owned foundation cannot give it.
  • Account vending is the actual problem. If your estate’s pain is that accounts are created by hand and everything else is fine, Account Factory solves the thing you have rather than the thing a vendor would like you to have.
  • Your team is small and your topology is simple. One region, a handful of accounts, no unusual compliance scope. The managed model’s constraints will not bind.
  • You are already on it and it is working. Migration is a cost with no user-visible benefit. Spend that quarter on the layers above the line instead; you can always move the base later.

What migrating actually costs

#

If you decide to move, know what you are signing up for. This is the part that gets underestimated in both directions.

Migration stepWhat makes it expensive
Re-parenting accounts into a new organizational unit structurePolicy inheritance changes the moment an account moves; anything depending on the old attachment has to be found first
Moving logging and configuration aggregationThe new archive has to receive and be verified before the old one is stopped, so there is a period of double delivery and double cost
Re-expressing controls as service control policiesA library selection becomes a policy you author, review and test — see service control policies that hold
Cutting over identityPermission sets, group mappings and session settings all move together, and nobody can be locked out during the window
Decommissioning the old landing zoneThe riskiest step, and the one that should come last, after a full quarter of the new foundation being the source of truth
The parallel periodTwo systems believing they own the same settings is the most likely way this goes wrong. Keep it short and write down which is authoritative per layer

Two practical rules from doing this kind of work. Never decommission before the new foundation has survived a real change, including one urgent one. And write the sequence down as a rehearsed plan with a rollback at each step, because the failure mode is not a broken account — it is a half-migrated organization nobody can reason about.

The BuiltForProd Baseline ships with a documented migration guide from AWS Control Tower for exactly this reason, and the AWS Enterprise Baseline’s own verification sequence — an 84-item checklist and 17 verification checks mapped to SOC 2 controls, run twice — exists because “it applied cleanly” is not evidence that a foundation is finished.

The honest summary

#

Control Tower is a good answer to the account and guardrail question and a non-answer to the platform question, which is fine, because it never claimed to be one. A code-owned foundation is a good answer to both and asks more of your team’s engineering culture in return.

If your evaluation is stuck, stop comparing products and do this instead: list the twelve layers in the table above, mark who owns each one in your organization today, and count the rows where the answer is “nobody yet.” That count, not the product comparison, is what your next quarter is actually about.

If you would like that count produced by someone outside the argument, the BuiltForProd Assessment measures an existing estate against the BuiltForProd Standard — and if Control Tower is the right answer for you, we will say so.

Questions

What does AWS Control Tower actually do?

Control Tower sets up and governs a multi-account AWS environment: it creates the organization structure with dedicated log archive and audit accounts, vends new accounts through Account Factory, and applies a library of preventive, detective and proactive controls to organizational units. Feature sets change frequently, so check the current AWS documentation before making a decision on specifics.

Is Account Factory for Terraform the same as a code-owned landing zone?

No. Account Factory for Terraform adds a Terraform-based pipeline for vending and customizing accounts on top of Control Tower. A code-owned landing zone expresses the whole foundation — accounts, network, identity, guardrails, logging, pipeline trust and workloads — in one repository under your review process, with no managed service holding part of the state.

Is Control Tower free?

There is no license fee for Control Tower itself, but you pay for the AWS services it turns on, such as configuration recording, logging and storage, and for whatever engineering time the rest of the platform takes. "No license fee" and "no cost" are different claims, and the second is the one that surprises people.

Should we migrate off Control Tower?

Not for ideology. Migrate when you need the layers it does not cover to live in the same reviewed codebase as the rest, when its customization model is fighting you, or when you want the entire foundation portable and readable. If it is working and your constraints are inside its model, the migration budget is better spent elsewhere.

What does it cost to migrate from Control Tower to a code-owned foundation?

The expensive parts are re-parenting accounts into a new organizational unit structure, moving logging and configuration aggregation to your own archive account, re-expressing controls as service control policies you author, cutting over identity, and running both systems in parallel long enough to prove the new one. Plan it as a quarter of part-time work with a rehearsed sequence, not a weekend.

This is what we build.

BuiltForProd deploys a production foundation into your own cloud accounts, hands over the repositories and the documentation, and, if you want it, keeps running it with your team. Tell us what you are running and what is coming next. The first call is 30 minutes, with an engineer who scopes engagements, not a salesperson.

The calendar invite and the video link go to this address. Prefer to write? Send a message instead.

Drift is a fact, not a failure

Infrastructure drift is not a sign of a broken team. It is information about what your environment needs. Detect it, triage it, and never auto-revert it.