AutomationOps, Explained for Charter Operations Desks

AutomationOps, Explained for Charter Operations Desks

AutomationOps sounds like vendor jargon, yet the problem it names is familiar to anyone who has worked a charter desk. Underneath the human work of dispatchers and schedulers sits a shelf of scheduled jobs — roster reminders, licence renewals, fuel-ticket syncs, passenger emails — each written once by somebody who understood the situation at the time, and each repeating that understanding ever since with nobody present. On a good week the shelf is invisible. On a bad week, one of the jobs misfires at four in the morning and the first human to notice is a customer.

That shelf is the subject of this piece. The Ops Atlas uses AutomationOps as an editorial label — RPA vendors attach the word to their own products, so the meanings differ — but the substance matters more than the naming dispute: treat the scripts that act on your systems the way an operator treats an aircraft. You know what each one touches, you know what happens when it fails, you can halt it from the desk, and somebody signs for it before and after every run.

The Job AutomationOps Actually Names

The label covers run-time actions: the restarts, clean-ups, failovers, credential rotations and fleet-wide configuration runs that fire on a schedule or in response to an event. These jobs share a family trait. They were written once, by a person who understood the situation at the time, and they now repeat that understanding — or that misunderstanding — with nobody present. AutomationOps is the practice of running that fleet as a production system: chosen deliberately, guarded before release, contained when it misbehaves, and owned by a named person until the day it is switched off for good.

Notice what the definition leaves out. It does not say “automation is good” or “automate more”. Deciding what not to automate is half the discipline. A repeat that is rare, poorly understood, or dependent on judgement does not become safe because somebody scheduled it. It becomes dangerous, because now it happens faster and nobody is watching.

Where the Boundary Sits

Visitors often arrive holding a neighbouring term, so the edges are worth drawing plainly. The Atlas keeps these disciplines separate on purpose:

  • Infrastructure code builds and reshapes the platforms themselves — servers, networks, environments. AutomationOps assumes the platform exists and governs the jobs acting on it while it runs.
  • Workflow engines chain steps into long business processes. The concern here is narrower: each individual action inside the chain, and how it behaves on its tenth repeat.
  • Chat-triggered work lets a person launch a job from a messaging window. Convenient, but the conversation is not the control; the guardrails around the job are.
  • Toil measurement belongs to site reliability practice, and it feeds this one — the toil ledger is where candidates come from. Controls flow the other way, out to the neighbours.

Keep the edges in view and the subject stops blurring. The question is never “which tool wins” but “which neighbour owns this piece, and what does the piece owe the whole.”

Five Controls, Explained Without Jargon

The Atlas describes five controls that turn a risky repeat into a boring one. None of them requires new software. All of them require intent.

  • Rehearse before you fly. A dry run plays the whole job end to end while changing nothing. Treat the rehearsal honestly: some tools skip the probing steps unless you flag them, so a green rehearsal can hide a job that would behave differently in earnest.
  • Never let an empty list mean “everything”. If a job meant to touch forty servers receives zero names, it must stop and complain — not shrug and act on the whole estate. Google learned this expensively: a cleanup routine once took an empty answer as “all of it”, and data that should have survived was erased.
  • Cap the blast radius. Roll out in batches. One target first, watch it, then a small group, then the rest. Stop the run when a slice of the batch fails instead of pressing on.
  • Give every job a failure budget. Decide in advance how many failures a single run may rack up — often just one — and halt there. A job that fails twelve times in a row is not persistent; it is out of control.
  • Fit a stop button and drill it. Each automation needs its own kill switch, with a global one above it. The drill is the point: for anything risky, practise the stop twice a year so the motion lives in the fingers before the bad night arrives.

Grading Each Job by Its Stakes

Not every job deserves the full ceremony, and pretending otherwise just breeds approval fatigue. The Atlas proposes four tiers, and they read like a weight-and-balance sheet:

  1. Tier 0 — reads only. Reports, checks, listings. If a job cannot change anything, it needs little beyond a named owner.
  2. Tier 1 — changes one target, reversibly. A restart here, a cache flush there. Rehearsal and a stop control are enough.
  3. Tier 2 — changes many targets, reversibly. Now the job must rehearse, move in batches, and pass health checks between waves.
  4. Tier 3 — destructive or irreversible. Deletions, overwrites, anything without an undo. A second person approves, deletions are staged so they can be recovered, and the rollout is spread over time.

The strictest tier mirrors a principle auditors already enforce elsewhere: two people for irreversible acts. Charter crews grasp this instinctively — a single signature rarely releases fuel, cargo or passengers — and the same arithmetic applies to a script that can erase a database.

Giving the Desk an AutomationOps Map

Every operations desk eventually learns the same lesson: the estate of jobs needs an owner’s map before it needs more tooling. Walk the shelf. Write down what runs, what it touches, who wrote it, and what it would cost if it ran wild. Retire anything whose purpose nobody can state in one sentence. Neighbouring disciplines reward the same housekeeping — the site’s walkthrough of RobotOps and robotics automation applies identical thinking to physical robots, while the overview of mastering modern IT operations places each practice in the wider frame. Even a finance desk feels the pull: the piece on automated invoicing describes exactly the kind of repeat that deserves governing rather than neglect.

The Atlas page that defines AutomationOps sets out the five controls and the tier model in full, and it is the reference behind everything above.

What Stays Human

Here is the part that surprises newcomers: the discipline is not a plan to remove people. Errors do not vanish when a job is automated; they migrate to the moment of design, where the script is written and its assumptions harden. That is precisely why a person stays in the loop at the edges — reviewing what the rehearsal showed, reading the health checks between batches, and holding standing authority to stop anything without asking first. The last point deserves a line in team policy, verbatim: the engineer on call may pull the stop on suspicion alone, and the review afterwards asks whether the suspicion was reasonable, never whether the pull was permitted.

Back at the charter desk, the shelf keeps humming. The difference after a season of this treatment is quiet but real: the dispatcher knows which jobs exist, the scheduler knows who owns each one, and when something misbehaves at four in the morning, the paged engineer knows exactly where the stop lives — and has already pressed it in practice. Boring, repeatable, owned. That is the whole promise, delivered in plain language.