The first time Meera said the word, she said it the way people say a diagnosis they have only half understood. “We’re going NoOps,” she told her small team of four, who between them kept a hospital group’s appointment portals, lab interfaces and backup tapes breathing. She had read that automation and rented cloud services had made operations departments obsolete, and her board wanted the salary line gone by summer. NoOps — the promise that buying and scripting can replace an ops function — was about to meet a four-person team, and what happened over the following ninety days is the most useful audit checklist I have ever been allowed to watch. This is that story, and the one-page audit it produced.
The NoOps Plan, as First Written Down
Meera’s original plan fit on a sticky note: migrate the remaining servers to managed services, switch on the automation the vendor demonstrated at a conference, thank the team, redeploy them. To her credit, she agreed to one condition before signing anything. Her senior engineer, Anand, asked for ninety days of “rounds” — a structured audit, the kind clinicians run without thinking — before any role disappeared. The board grumbled; the compromise held. What follows is the audit they actually ran, lightly disguised, because the details belong to them but the lessons belong to everybody.
Grading Ourselves the Way Cardiology Grades Function
Clinicians reading this will recognise the method before the content. Cardiology does not ask “is this patient well?” — it grades function in classes, from no limitation to symptoms at rest, and the grade drives the plan. The technology world has an equivalent scale for operations maturity, and it was the spine of Meera’s audit. Level zero, in the field’s blunt vocabulary, is “no ops by assertion”: the label is declared and nothing else changes, which is where her sticky note honestly sat. Level one means duties are at least registered and named. Level two means every duty has an owner who knows they own it. Level three adds rehearsed failure — restores and provider outages practised, not merely feared. Level four, “low ops by design”, is the realistic summit: the residue has been automated or consciously accepted, counted and priced. Nobody jumps classes; you climb them, and the grade you admit to determines what treatment makes sense.
What the Audit Actually Asked
The first week produced a single page of questions. They were not technical, which surprised the board; they were questions a matron would ask about a ward.
- Who is woken when the appointment portal fails at 2 a.m., and does that person know they are on the list?
- Which duties have already moved to vendors or automation — and is there a written list, or only a memory?
- Of the duties that moved, who on our side still quietly performs the residue: the checks, the reconciliations, the emails?
- When did we last rehearse a restore from backup, and did it finish inside the working day?
- Which of our critical services depend on which external providers, and what does the arithmetic of their promises add up to for a patient at 9 a.m.?
- If our main cloud vendor retired our key service, what is our exit estimate — in weeks, in rupees, in staff?
- Does every scheduled script and every automated agent have a named owner with permission limits?
- What is our unowned duty share, as a percentage — and is it zero, or is it a number nobody has dared to count?
Question three drew blood. When the team listed what they actually did each week, a third of their hours were residue from “automated” systems: correcting feeds that the automation mis-mapped, coaxing a vendor’s nightly job, reconciling bills. The operations had been purchased, all right — and it had stayed home, wearing different clothes. Anand’s dry summary entered team folklore: “We didn’t remove the work. We removed the job title.” He was echoing an older argument than he knew; when Netflix let its developers carry the deployment pagers back in 2012, one of the field’s most respected veterans observed that operations had simply been reorganised, and the serverless revival of 2016 to 2018 drew the same scepticism from people with long memories, including Charity Majors and the CNCF crowd, who pointed out that somebody still holds the pager somewhere, always.
Running the Same Audit in an Afternoon
You do not need Meera’s budget to copy her method, only her nerve about honesty. Block half a day. Bring the people who actually carry the pager, not the people who approve the budget, because the two groups hold different facts. Write the eight answers on a whiteboard without editing — half-answers and blanks are findings, not failures. Then grade yourselves against the four levels, and resist the urge to round up; every grade inflation at this stage becomes a 3 a.m. surprise later. Finally, pick exactly three findings to fix in the next month. Small shops fail at this step in both directions: teams that fix nothing stay orphaned, and teams that try to fix everything at once reorganise themselves into exhaustion. Three fixes, written where the team can see them, reviewed at the same hour every week — that is the whole governance model, and it is enough.
The Ninety-Day Prescription
The audit’s findings were uncomfortable, which is how you know an audit is honest. The team graded itself at level one: duties half-registered, ownership partial, rehearsals never. Meera, to her lasting credit, rewrote the board proposal instead of defending it. The new plan ran on the standard thirty-sixty-ninety rhythm. Days one to thirty: finish the register — every recurring duty written down with hours attached, including the residue nobody had counted. Days thirty-one to sixty: assign owners and trim permissions, so each script, agent and vendor contact answered to a named person. Days sixty-one to ninety: rehearse failure — a timed restore, a simulated provider outage, a walkthrough of the exit from their riskiest dependency — and put first estimates against leaving each critical vendor. The salary savings the board wanted did not materialise that quarter. Something better did: by day ninety the same four people were spending measurably fewer nights awake, the restore rehearsal had shaved eleven minutes off its own record, and the group’s clinical directors, who judge reliability for a living, stopped asking about the headcount and started asking about the method.
What NoOps Meant by the End of the Story
Ninety days later, Meera’s unit still existed, still had four people, and had earned its version of the label honestly. The chores that left — patching cycles, hardware refreshes, the tape library — left for good, purchased and proven. The judgement that stayed stayed on purpose: keys, configuration, access, the vendor relationships, the 2 a.m. phone. If you ask her whether she “did NoOps”, she will say she stopped doing ops by hand and started doing ops by design, which is the only version of the promise that survives an audit. Cardiology reached the same conclusion about its own hype cycles long ago: grading precedes treatment, treatment precedes discharge, and nobody sane discharges on the basis of a slogan. Readers who want the graded vocabulary in its clinical form will find it in our overview of NYHA functional classification, and the follow-up habit the audit imitates — measured reviews on a schedule, never “as needed” — is exactly what our note on cardiac follow-up describes.
The audit question that outlived the project: if a duty has no name next to it, it has not been automated — it has been orphaned. Count those first.
Two closing pointers. For the full biography of the term Meera’s board had fallen for — its 2011 origins, the Netflix chapter, the honest limits — the Ops Atlas hosts a thorough NoOps reference page that pairs well with a ninety-day plan. And for readers curious how disciplined engineering partners approach the Indian digital ecosystem this story sits inside, our earlier profile of how Cotocus approaches India’s digital transformation shows the register-owner-rehearse method applied at much larger scale.
