Infrastructure as code solves the "nobody knows what is running" problem and creates a new one: nobody knows who is allowed to change it. Governance is the set of controls around the code: review, policy, state, approvals and evidence. It decides whether the platform is auditable or just automated.
This comparison covers four governance models used by mid-sized teams. It is written for the CTO or operations lead who has two to six engineers and a growing cloud bill, not for an enterprise platform organisation with fifty specialists.
The four models
1. Review-only
A pull request, one reviewer and a written convention document. The pipeline runs a plan and posts it as a comment.
- Strengths: cheapest to start, low friction, familiar to any team using Git.
- Weaknesses: the reviewer is the only control. A tired reviewer approves a production database deletion, and the plan output is often too long to read.
- Fit: one team, one cloud account per environment, few changes per week.
2. Policy as code in the pipeline
Static analysis and a policy engine evaluate the plan before apply. Rules block public buckets, unencrypted volumes, overly broad roles or untagged resources.
- Strengths: consistent enforcement, fast feedback, and a written rule set that doubles as documentation.
- Weaknesses: policy maintenance becomes its own backlog, and poorly written rules block legitimate work until people learn to bypass them.
- Fit: any team with more than one engineer and a shared cloud account.
3. Platform team and golden modules
A small platform group owns versioned modules, environment definitions and pipelines. Product teams consume modules rather than writing raw resources.
- Strengths: small blast radius, consistent architecture, far less duplicated code.
- Weaknesses: the platform team becomes a queue. If they are also on call, module requests wait.
- Fit: three or more product teams sharing infrastructure, or a regulated environment with a fixed architecture.
4. Managed run environment
Remote state, a controlled run environment, approval gates, drift detection and an audit trail are provided by a platform such as Terraform Cloud, Atlantis or Spacelift.
- Strengths: strong evidence trail, consistent runs, no state file on a laptop, built-in approval records.
- Weaknesses: another vendor, another bill and another access model to review. Exit requires exporting state and rebuilding the pipeline.
- Fit: auditor-facing environments, distributed teams, or teams that keep losing state.
Criteria that separate them
- Control strength: model 2 and model 4 are strongest because enforcement is mechanical. Model 1 depends entirely on the reviewer.
- Developer speed: model 1 is fastest for small changes. Model 3 is fastest at scale once modules exist, and slowest before they do.
- Blast radius: model 3 limits it through module boundaries and separate accounts. Model 1 limits nothing.
- Drift detection: only model 4 provides it out of the box; elsewhere it is a scheduled job you build.
- Audit evidence: model 4 records who approved what and when. Models 1 and 2 rely on Git history, which is good but does not capture apply-time context.
- Skills required: models 1 and 2 need Git and cloud literacy. Models 3 and 4 need someone who can own a platform.
- Cost structure: models 1 and 2 are mostly engineering time. Model 3 is engineering time plus a platform owner. Model 4 adds a subscription priced by runs, resources or users.
Decision guidance
Most teams should not pick one model. A practical combination is policy as code plus a small set of golden modules, with a managed run environment added when the audit requirement arrives.
- One team, one cloud, under ten changes a week: review-only with a mandatory plan comment. Add policy checks for the three failures that would hurt most.
- Two or three teams sharing accounts: policy as code, separate accounts per environment, and modules for the resources that are always built the same way.
- Audit or customer due diligence: managed run environment with approval gates, plus policy as code inside it, plus an evidence export reviewed quarterly.
- Regulated or multi-cloud: all four, with the platform team owning the module catalogue and the policy library.
The parts people forget
- State security. The state file contains secrets in plain text. Encrypt it, restrict access, and enable locking so two applies cannot race.
- Environment boundaries. Separate accounts or subscriptions are a stronger control than naming conventions in one account.
- Break-glass access. A documented, time-limited, logged path for the day the pipeline is broken. Without it, engineers create permanent admin credentials "just in case".
- Module versioning. Pin versions, publish a changelog and deprecate on a schedule. Unpinned modules make every apply unpredictable.
- Drift ownership. Someone must own the weekly drift report, or the report becomes decoration.
- Cost visibility. Tagging rules belong in policy, not in a spreadsheet reviewed at quarter end.
The local team reality
Most mid-market companies in Bangladesh do not have a platform team, and hiring one is not realistic in the short term. That makes the hybrid path the practical one: policy as code in the existing pipeline, two or three modules owned by the most senior engineer, and a managed run environment only when an auditor or enterprise customer requires the evidence.
A 60-day adoption plan
- Week 1-2: inventory repositories, state files, cloud accounts and who can apply to production.
- Week 3-4: move state to remote storage with locking, and remove local state from laptops.
- Week 5-6: add plan comments and three blocking policies: public storage, unencrypted volumes, wildcard permissions.
- Week 7-8: extract two or three golden modules and migrate one service as the reference implementation.
Metrics
- Percentage of infrastructure changes applied through the pipeline.
- Policy violations caught before apply, and violations found after.
- Drift detected per month and time to reconcile.
- Number of standing production credentials.
- Lead time from pull request to applied change.
Where this fits in our work
If you want the controls designed around your current repositories, start with Managed Technology & Cybersecurity; the pricing page explains how platform work is scoped. For the commercial side of the same decisions, read how cloud vendor lock-in changes year by year.
Sources
- HashiCorp, Terraform documentation, accessed 17 November 2026.
- Open Policy Agent, documentation, accessed 17 November 2026.
- NIST SP 800-53 Rev. 5, Security and Privacy Controls, accessed 17 November 2026.
- Center for Internet Security, CIS Benchmarks, accessed 17 November 2026.
- FinOps Foundation, accessed 17 November 2026.
Responses (0)
No comments yet. Add the first one below.
Leave a response
Your email address will not be published. Required fields are marked *.