Hangar talk is full of confident sayings: “if it passed the walk-around, it is fine”, “never touch a thing that works”. Some are wise and some are traps. Server rooms have their own folklore, and the discipline nicknamed InfraOps, the careful handling of infrastructure that is described in code, is crowded with it. So let us go through six popular myths, one by one, and see what the Ops Atlas and ordinary common sense say instead. Grab a coffee; this is a friendly one.
Why myths stick around
Folklore spreads when it saves time. A rule of thumb like “the plan was clean, ship it” works nine times out of ten, and the tenth time is remembered as bad luck instead of a flawed rule. Pilots call this normalisation of deviance, and it applies just as well to a team that has not been bitten yet. Naming the myth is the first step to replacing it with a habit that holds up under pressure.
Myth 1: “InfraOps is a tool you can buy”
It sounds sensible, because every discipline seems to come with software. But in the Atlas write-up of InfraOps, the word is described as informal shorthand with no standards body, exam or licence standing behind it. A 2025 commercial release shares the name, and the Atlas files that under coincidence rather than definition. Think of it as the word “airmanship”: a way of working that tools can support, not a thing in a box.
Myth 2: “A clean plan means we know what exists”
This is the big one. A clean preview only covers objects the tooling already tracks. Something created by hand during a late-night fix will never show up in it. The Atlas points to a vendor tutorial where a hand-made security group appeared merely as an identifier in a diff. It is like a pre-flight check that only inspects the items printed on the checklist; a loose panel not on the list sails right past.
Myth 3: “Immutable servers mean you never patch”
Teams that replace machines instead of repairing them are sometimes accused of ignoring security updates. In truth the patching has moved. It happens in a scheduled image build, then fresh machines replace the old ones. One cloud architecture framework even forbids in-place updates on production under this model. It is the difference between fixing a part on the wing and swapping in a freshly inspected unit.
Myth 4: “Auto-revert makes drift a non-issue”
If someone edits a server by hand, why not let the machine put it back automatically? Because an automatic revert is an apply job with the strongest credentials and nobody reading it. The Atlas suggests confining it to components that keep nothing durable, and then only while the team is not fighting a live incident. Everywhere else, it prefers to tell an owner and give them about three days. Imagine an autopilot that silently overrides the captain whenever readings deviate; sometimes the deviation is the captain’s right call during an emergency.
Myth 5: “A delete-protection flag protects our data”
Many tools offer a setting that refuses to destroy a resource. Handy, but the Atlas notes it stops working if the configuration block that carries it is removed. The safer plan is to pair it with deletion protection at the provider, so a second lock exists outside the code. In flying terms, do not rely on a single guard on a switch that the same hand can remove.
Myth 6: “Just import everything and be done”
It is tempting to pull a hand-built estate into code in one sweep. The Atlas cautions that imported code tends to copy historical accidents, and Google advises avoiding imports where possible. Its illustrative approach is to bring in the ten riskiest resources of a three hundred item estate first and build coverage toward roughly eighty percent over a quarter. Imagine re-typing every old logbook entry on day one when you could start with the safety-critical systems and let the rest wait. The same ordering rule applies to code: risk first, tidiness later. After the first batch is under control, the team can decide whether the remaining hand-built pieces deserve code at all or should simply be retired.
A bonus myth: “Our coverage number is high, so we are safe”
The Atlas works through a sample where 1,512 resources are under code and 1,840 sit in scope. Quote that as a percentage and it feels comfortable. Present it as 328 items nobody reviews, plus one drift left open for 17 days, and the conversation changes. Which framing do you think gets a budget request approved?
So what should a team do instead?
- Compare code, the tool’s records and reality on a schedule, not just once.
- Keep records small, because Google Cloud suggests a few dozen resources each and certainly no more than around a hundred.
- Read previews before approving and treat them as short-lived.
- Apply from a pipeline with narrow credentials, not from laptops.
- Rehearse a rebuild, so recovery is something you have actually seen work.
Keep learning
If you are curious about how larger organisations plan their cloud operations, the roadmap article on software engineering and cloud operations planning in Japan gives a regional view. And anyone eyeing a switch from the cockpit or the hangar into technology may enjoy this walkthrough on mastering tech careers with a training school.
The takeaway is the same one every instructor gives: trust the checklist, but know what the checklist does not cover.

