Cardiology reads trends in traces: a rhythm strip tells you where a heart has been, where it is now, and what the next stretch probably holds, provided you know what a normal complex looks like. The idea of NoOps — information technology run so automatically that no separate operations team is needed — has produced a trace of its own over fifteen years, and 2026 is a useful moment to read it. Three waves of enthusiasm, two public rebuttals, and a fresh cluster of AI agents make up the recording. Hospital IT leaders, who live with the consequences of operational promises more acutely than most, will recognise the shape: every exaggerated claim on this strip eventually presented a bill, and every honest measurement paid it down.
The NoOps Trend Since 2011: Three Waves
Wave one: the provocation (2011-2013)
The term enters the record in February 2011, when a Forrester analyst argued that developers’ dealings with operations professionals could simply end — operations purchased rather than staffed. His own firm softened the position within a quarter, recasting the concept as DevOps extended rather than DevOps replaced, but the provocation had escaped. In 2012 the label resurfaced at a streaming company whose cloud architect pinned it on developers owning their deployments and alerts outright, and one of the respected elders of the operations world answered that the work had merely been reorganised. The first wave ended the way first waves do: with a vocabulary lesson rather than a transformation. Its lasting artefact was the question it forced into the open — which operational duties could, in principle, be purchased — a question worth keeping even after the prediction expired.
Wave two: the serverless revival (2016-2019)
Functions-as-a-service brought the claim back with better engineering behind it. If you rent code execution by the request, surely the machines — and the machine-tenders — have gone. The rebuttal came from two directions at once. Prominent practitioners, Charity Majors among them, declined the equation, and the Cloud Native Computing Foundation’s whitepaper stated that serverless adoption removes neither servers nor the people who understand them. Meanwhile a widely read 2019 Berkeley study credited the model with absorbing almost all system administration while flagging erratic cost behaviour at scale and a stronger tether to a single vendor. Both findings can be true, and both were: serverless is a real reduction in certain duties and a real transfer of certain risks — a change of hands, not an absence of hands. The wave’s residue in 2026 is everywhere: most hospital-adjacent systems now rent at least one function-by-request component, usually without anyone calling it by the old ideological name.
Wave three: the agent era (2024-present)
The current wave runs on machine intelligence rather than on pricing models. Amazon’s DevOps Agent went generally available this past March, with Azure’s SRE Agent following in the same season; both compress the diagnostic loop that used to eat an on-call engineer’s night. Vendor material phrases this as operations becoming autonomous, which is where the evidence checks begin. The pattern rhymes with the previous waves more than its publicity admits: a genuine capability, wrapped in a claim about disappearance that the capability’s own documentation declines to make.
Reading the 2026 Trace: What the New Wave Actually Contains
Look closely at the agents’ own documentation and the waveform is honest where the marketing is not. Microsoft’s guidance for its SRE Agent recommends keeping a person reviewing its work whenever production systems are involved — a recommendation that quietly concedes the entire argument: the most automated operations product of the cycle is designed around a human checkpoint. Someone owns each agent’s permissions; someone answers for its mistakes; someone notices when an automated investigation reaches a confident wrong conclusion faster than a careful right one. Hospital IT will meet this pattern first through patient-facing systems, where a wrong automated “fix” during clinic hours is not an abstraction.
The regulatory trace is moving too. From mid-January 2027, Europe’s Data Act wipes out the fees customers pay to switch providers or move their own data out — genuinely reducing the cost of moving data, while leaving the engineering labour of a migration untouched. And the commercial trace keeps repeating its old pattern: Google discontinued the IoT Core product line in 2023, handing subscribers a scheduled migration they had not budgeted. Trends in pricing and product availability deserve the same continuous monitoring as trends in automation, because they change what yesterday’s safe arrangement costs tomorrow.
What the NoOps Evidence Supports for Hospital IT
Read against that background, the defensible position for health systems is neither the banner nor the dismissal.
The evidence supports a measured reduction: hospital platforms with bursty, modest load — appointment finders, information portals, physician directories — can genuinely operate without a dedicated operations unit if four conditions hold:
- every surviving judgement — retention of records, sign-in rules, configuration, access — carries a named owner;
- restores are rehearsed quarterly, with the drill’s most recent date recorded somewhere an auditor can find it;
- exits are priced, in money and in staff-weeks, before contracts reach renewal;
- automated actors of every kind report to a person with the authority and the habit of reversing them.
Under those conditions practitioners grade the arrangement at the top of their maturity ladder — deliberately minimal, engineered rather than asserted. Remove the conditions and the arrangement falls to the ladder’s bottom rung, which exists in the literature precisely because so many organisations have announced the summit without climbing.
The displacement arithmetic is the part busy executives skip. Work that leaves an operations role reappears among providers, platform staff, product developers with pagers, automation pipelines, and contract management. The Snowflake intrusions offered the sector’s clearest demonstration that the resident portion matters most: the platform held; customer accounts without multi-factor sign-in, rotation, and allow-lists did not. Availability promises multiply across dependencies rather than borrow from the strongest one, so a portal’s real ceiling sits below its best vendor’s brochure. These are clinical-grade facts — checkable, dated, unforgiving of wishful reading.
One further reading of the trace deserves the board’s attention: the relationship between enthusiasm cycles and staffing. Each wave has pulled experienced operators out of organisations that believed it wholesale — people who left when their function was declared finished — and each wave has then hired consultants to reconstruct the knowledge those people carried. The cheaper path, statistically, has been retention through transition: keep the operators, redefine the role around the four conditions above, and let the automation remove the toil rather than the person. Health systems, which already understand what continuity of staff means for institutional memory, are better placed than most industries to run that play.
Leaders who want the full strip annotated — the history, both definitions, the metrics practitioners track — will find this NoOps timeline and reference the most complete single document on the subject, and its sourcing is careful enough to survive an audit committee. The monitoring theme runs through our own coverage as well: our survey of hospitals using advanced heart monitoring technology is, at bottom, about systems that watch continuously and escalate to humans on thresholds — the architecture every operations setup converges on. Our cardiac ICU guidelines for patients and families show the same ownership discipline in clinical form: every line item has a person, every escalation has a trigger. And the cost-reading habit transfers directly — our explainer on heart surgery costs and hospital comparison teaches the same separation of quoted price from total cost that the operations arithmetic demands.
Two readings that matter for the year ahead
First, the agent wave is the first one whose products ship with the human checkpoint built in rather than argued about afterwards. That is progress of a real kind: the industry stopped debating whether a person should review automated operations work and started designing where the person sits. Second, the regulatory calendar now shapes these arrangements as much as the technical one does — the 2027 European changes are the clearest example, and they reward exactly the organisations that keep portable, documented, exportable systems. Neither reading suggests standing still; both suggest standing on measurements.
The Trace, Summarised
Fifteen years of recording show one consistent waveform: the operations function compresses, relocates, and re-specialises — it does not flatline. The 2026 agents compress it further than any previous wave, and they ship with human checkpoints as standard equipment. For hospital IT, the reading is straightforward: adopt the automation, keep the ownership explicit, monitor the numbers you would monitor if the title still existed — because the duties certainly do.