Hospital systems teams hear a lot of confident sentences about cloud infrastructure, and several of them are wrong in ways that cost real time. This piece takes six common myths about InfraOps, the informal name for managing infrastructure as reviewed code, and sets each against what a public reference says. It is written for administrators and IT leads in cancer care organisations who want straight answers, and it contains nothing clinical.
One housekeeping point first. InfraOps is shorthand, not a standard. According to the xOps School’s Ops Atlas, no standards body or certification defines the word, and it also shows up as a product name from an unrelated company. Treat it as a useful label for a set of practices.
Myth 1: “The plan came back clean, so we know what exists”
A clean preview means the tooling found no differences among the things it manages. It says nothing about resources somebody created by hand. The Atlas notes that a well-known vendor tutorial on drift surfaced a hand-made network rule only as an identifier inside a comparison. If your confidence rests on a clean report, ask a second question: what is not on the list?
Practical response: keep a periodic inventory that starts from the provider’s side and checks each item against your code.
Myth 2: “Immutable servers mean we never patch”
Teams that replace servers instead of editing them in place sometimes conclude that patching has disappeared. It has not; it has moved. The Atlas explains that patching shifts into a scheduled pipeline that builds fresh machine images, and that it cites a cloud architecture framework that rules out editing production machines in place once you adopt the replace-don’t-repair model. The upshot is that image age becomes a number worth watching.
Practical response: rebuild images on a regular cadence and track what share of servers run on recent ones.
Myth 3: “Auto-reverting every difference keeps us safe”
The appeal is obvious: any time reality wanders from the code, a job pulls it back. The trouble is that you have built an apply pipeline with strong credentials and no human review. The Atlas would allow automatic reversal only where a module stores nothing durable, and only on days with no live incident. Everywhere else it proposes alerting and assigning an owner with a target of around three days.
Practical response: decide per module which kind of drift is allowed to heal itself, and write the decision down.
Myth 4: “A protective flag on a resource guarantees it can’t be deleted”
Many tools offer a switch that makes a destroy command fail for one resource. The catch, as the Atlas describes it, is that the protection lives in the configuration. Delete the configuration block and the guard disappears with it. Pair the setting with the provider’s own deletion protection for anything that holds records you cannot recreate.
Practical response: layer two protections at different levels, and test that they hold.
Myth 5: “Let’s pull the whole existing estate into code in one go”
Bringing a hand-built estate under code is a good aim, but doing it all at once produces enormous files that merely record historical accidents. The Atlas relays one cloud provider’s advice to avoid imports where possible, and gives an illustrative plan: take a three-hundred-resource estate, bring in the ten riskiest first, and aim to move coverage from nothing to about eighty percent within a quarter. That example is a sketch, not a benchmark.
Practical response: rank by risk, import in small batches, and reread the code each time rather than trusting generated output.
Myth 6: “InfraOps is a product, or just GitOps by another name”
It is neither. Per the Atlas, the label does not name any tool, and it keeps the topic apart from the topic from continuous reconciliation. Consider this: when a pipeline runs an apply command right after a merge, what you have is code-driven provisioning with some review automation around it. That differs from a controller that keeps comparing and correcting on its own. The distinction matters when a vendor describes a feature using borrowed words.
Practical response: ask suppliers what, exactly, runs when, and who approves it.
A few lines on why these myths survive
Myths like these last because each one is partly true. A clean report is a real signal, just a narrow one. Replacing servers does reduce hand patching, just not to zero. Automatic reversal does help in some modules. When a belief is half right, nobody feels the urge to test it, and the other half stays hidden until an outage arrives.
There is also a vocabulary problem. Words like “managed”, “immutable” and “compliant” sound final, yet each hides a boundary. Managed by what? Immutable until when? Compliant with which written baseline? In an administrative setting such as a cancer care organisation, where procurement teams and IT partners talk past each other, asking for the boundary behind each word is the single cheapest improvement available.
If you only have time for one experiment, run the first myth’s test. Ask your technical partner to produce a list of every cloud resource in the account straight from the provider, then compare it with the list generated from your code. Whatever appears on one list but not the other marks your true unreviewed area, and its size tends to surprise people. Share the numbers openly, assign each gap an owner, and revisit the comparison monthly so the trend, not the snapshot, tells the story.
A last practical tip: keep a one-page glossary that your team agrees on, listing what each term means in your own estate. Update it whenever a supplier introduces a new word. It sounds dull and it prevents expensive misunderstandings.
What to do with all this
A myth list is only useful if it changes a Monday. Choose one of the six that sounds most like a sentence you have heard in a meeting, and test it against your own environment this week. You can read the fuller discussion in the Atlas’s InfraOps reference page, which also covers state size, plan handling and a maturity ladder it labels as its own proposal.
Readers who want to see how a wider engineering mindset fits around these ideas may like our overview of how modern software delivery and cloud-native work come together. And on the non-technical side, our note on why families look for formal accreditation when comparing hospitals overseas reflects the same instinct: ask what is formally checked, not what is merely claimed.
That instinct is the whole of this article. A calm, specific question beats a comfortable assumption, in a server room and in a boardroom.
