The short answer
If your infrastructure exists as code in version control, augmentation works like any other development work: changes are reviewed, tested and applied through a pipeline. If it exists as a console someone clicked through two years ago, fix that first.
The reason is not trust. It is that undocumented infrastructure cannot be reviewed, so nobody can tell whether a change is safe.
The precondition
- Infrastructure defined as code, in version control
- Changes applied through a pipeline rather than by hand
- Environments that can be recreated rather than repaired
- Secrets managed properly, not in configuration files
- A way to see what changed and when
Where these are missing, the first engagement is usually to establish them. That is valuable work regardless, and it is the thing that makes everything afterwards reviewable.
What hands over well
| Work | Fit | Why |
|---|---|---|
| Build and deployment pipelines | Good | Definable, testable |
| Monitoring and alerting setup | Good | Clear requirements |
| Environment provisioning | Good | Code, reviewed like any other |
| Cost optimisation review | Good | Bounded, measurable |
| Incident response | Poor | Needs system and business context |
| Production emergency access | Poor | Keep inside your team |
Keep production knowledge internal
The temptation with infrastructure work is to let the specialist own production because they know it best. That creates a dependency that is invisible until three in the morning.
Pair on anything production-facing, require runbooks as part of the work, and make sure at least one of your own people has done each operational procedure at least once. Reading a runbook is not the same as having run it.
Access, carefully
- Named individual accounts, never shared.
- Development and staging by default, production only where genuinely needed.
- Time-limited elevation for production tasks, with a record.
- Changes applied through the pipeline rather than directly.
- Removal on the agreed date, with secrets rotated.
Most infrastructure work does not need standing production access. Where it does, elevation with a record is a better answer than permanent permission.
The value that outlasts the engagement
The lasting benefit of good infrastructure work is not the pipeline. It is that your team can now change things safely, recreate an environment, and see what is happening when something breaks.
Judge the engagement on that rather than on what was built. A beautiful pipeline nobody internal understands is a liability with good intentions.