Good clinicians do one thing first: they separate symptoms from disease. Coughing is not the illness; it is the body reporting one. Software organisations work the same way, and the presenting complaints sound familiar — releases crawl, tickets pile up, every new service takes weeks to stand up, and three teams just rebuilt the same tool. Managers usually prescribe effort: more people, longer hours, a sharper deadline. The honest diagnosis is usually different, and it has a name now: the organisation needs PlatformOps — the discipline of operating shared build-and-run capabilities the way products are operated. Let’s walk through it like a proper consult — history, examination, diagnosis, treatment plan, and a realistic recovery curve.
Taking the history: what patients report
Ask an engineering organisation what hurts and you will hear a surprisingly consistent complaint list. New developers take weeks to ship anything. Every project wires up its own build and deployment, each slightly wrong. Requests for environments or access disappear into an ops queue and emerge days later, changed in ways nobody documented. And everyone privately maintains a workaround — a script, a checklist, a sticky note — because the official route is slower than the detour.
Individually these look like personality problems. Together they are a classic pattern, and researchers have catalogued it for years. Spotify gave the duplication part its best-known name, “rumour-driven development”: teams rebuilding similar things because no dependable way existed to find out what already existed. DORA’s researchers, meanwhile, flagged the queue itself — the ticket-and-vending-machine arrangement — as one of the reliable anti-patterns of platform work. Neither is a people failure. Both are operating-model failures wearing people-shaped costumes.
The examination: patterns a clinician would recognise
A few quick tests separate an operating-model problem from a talent problem. Does the same request recur across teams — meaning the organisation is re-performing work it should have done once? Does knowledge about deployments live in two heads rather than in documentation? When something breaks at 2 a.m., does the same exhausted group always get paged, regardless of whose code caused it? Three yeses indicate the organisation has no platform in the meaningful sense: no curated, supported, self-serve route to production. It has favours, performed at scale, by tired people.
The diagnosis: why PlatformOps, not more hires
Here is the part that changes treatment. If the cause were talent, hiring would fix it. But hiring more engineers into a broken flow adds queue depth, not throughput — the same way an extra cook cannot speed a kitchen whose single oven everyone shares. The disease is structural: common capabilities that keep being re-performed as favours instead of existing once, maintained, as products with owners. Modern practice has converged on the cure’s name — PlatformOps — and on its core moves, which the Ops Atlas notes on PlatformOps document in detail.
The PlatformOps prescription
- Pave the common route. Build one supported path from code to production — the golden path — opinionated and maintained, with the explicit rule that teams may leave it but then carry their own support burden.
- Turn favours into self-service. Environments, pipelines and access become things teams fetch themselves, at the moment of need, without a human broker in the middle.
- Publish the promises. Every capability carries a declared service target and a stated tier of support, so consumers know what “working” means before they depend on it.
- Assign owners and lifecycles. Each capability has a named owner and a lifecycle from proposal through beta and general availability to eventual retirement — with the pager split clean: app teams for app services, the platform team for platform parts.
- Treat adoption as a product outcome. Usage is earned by making the platform genuinely easier, never decreed; the CNCF’s guidance keeps platforms optional and composable precisely for this reason.
Recovery: what improvement actually looks like
Set honest expectations with the patient. Delivery metrics typically dip during treatment — the J-curve that DORA’s research keeps finding — before climbing past the old baseline, and a botched rollout can leave things worse than it found them. So the plan includes follow-up measurements, not vibes: what proportion of fresh services start life on the paved template, how many days pass before a new engineer’s first production release, how often teams get what they need without talking to another human, and whether the old workaround collection is shrinking. If those move in the right direction over two quarters, the treatment is working; if not, adjust the plan like any good clinician would.
Knowing when to refer — or when rest will do
Good medicine includes knowing when not to operate. Below roughly three teams with genuinely repeated needs, the evidence favours conservative care: tidy documentation, a couple of shared scripts, and a review in six months. Forcing a platform onto a two-team organisation treats a bruise with surgery. If you recognise the escalation logic from telling ordinary respiratory trouble from something that needs a doctor, you already understand the decision tree — match the intensity of the intervention to the severity and persistence of the symptom. And when the symptoms persist, get a proper consult: the telehealth guide’s advice on seeking guidance early applies to platform troubles too, because early expert eyes are cheaper than late rescues. For readers building the underlying skills, the DevOps roadmap for remote workers is a sensible first prescription.
Follow-up appointment
Summing the consult up: the coughing never was the disease. Ticket backlogs, duplicated tooling and slow onboarding are the organisation reporting a structural problem — common needs being served as one-off favours. The PlatformOps treatment is old-fashioned in the best sense: do the common things once, keep them working, publish what you promise, and stop performing the same favour forever. Most organisations that follow the plan — with realistic expectations and honest measurement — find the chronic cough fades within a couple of quarters, and nobody misses it.