When my mother had her hip replaced, the surgeon’s parting words were not about the operation at all. “The surgery was the easy hour,” she said. “Everything that matters now is the walking plan.” Software teams discovering PlatformOps get the same lecture, eventually. The platform you announce is the easy hour; everything that matters is how it gets walked into daily use, one supported step at a time. Here is the six-step plan we followed at a regional insurer last year, told the way it actually felt — including the wobble in the middle that nearly ended it.
Step One: Diagnose Before You Prescribe Anything
Nobody books a hip replacement off a hunch, and nobody should book a platform off a conference talk. We started with five interviews across the engineering teams, asking one question: where does your time actually go? The answers overlapped uncomfortable amounts. Every team maintained its own build scripts. Three had independently re-implemented the same deployment approval flow, unknown to each other — a small, private echo of what Spotify once called rumour-driven development, where teams rebuild similar tools because nobody knows what already exists. We wrote the inventory down, capability by capability, and marked who owned each one. Half had no answer.
Step Two: Agree the PlatformOps Charter in Writing
With the diagnosis on paper, we drafted a one-page charter and cut the ambition to the thinnest viable platform we could defend: the smallest set of shared capabilities that removed genuine repetition, and nothing more. Two rules went into the document. First, adoption would be pulled — teams opt in because the platform makes their week easier, never because a director said so. Second, every published capability would carry an owner, a lifecycle stage, and a service level we were prepared to be judged against. The physiotherapist framing helped here: this is your exercise plan, in writing, with milestones. Signing it was what turned six sceptical seniors into participants.
Step Three: Take the First Supported Steps
Rehabilitation starts with two steps to the kitchen and a hand on the rail. Our rail was a single golden path: one template, one pipeline, one route from commit to production, built with a design partner — the payments team, who had volunteered and had the most boring, well-understood service to trial it on. For four weeks their engineers used nothing else, and every snag became a backlog entry rather than a grievance. That is the design partner’s real function: a hand on the rail while the path finds its footing, close enough to catch you and honest enough to say when it wobbles.
Step Four: Widen the Circuit Carefully
Next came the longer corridor: a beta with a second team, deliberately different in shape — event-driven, older codebase, different deployment rhythm. The payments team’s quiet endorsement did more persuading than any slide deck; engineers believe engineers, not memos. This is also where Team Topologies earned its keep. Early on we worked in collaboration mode, side by side with each adopting team. As the path stabilised, the relationship settled into X-as-a-Service: teams consumed the platform without our hands in their repos. Watching that handover happen without ceremony was the quietest, most satisfying milestone of the year.
Step Five: The Wobble, and What It Taught Us
Week nine was our setback. Adoption flattened, two teams quietly reverted to their old scripts, and a director asked why we were still paying for the experiment. The metrics explained it: we were sitting in the dip of the J-curve that DORA’s research predicts — delivery performance typically sags while teams migrate, then recovers and improves. We presented the curve, marked our position on it, and asked for six more weeks instead of defending the plateau as success. Whoever runs your program needs that argument rehearsed in advance, because the dip always arrives, and unprepared sponsors cancel good platforms in bad months.
Step Six: PlatformOps Discharge Day, Rehab Plan Included
General availability is not the finish line; it is discharge day. Before declaring it, four items had to be true:
- a runbook written and rehearsed against real failures
- an on-call rota covering the platform’s own components
- a lifecycle policy covering every stage a capability passes through, from first proposal to final retirement
- a first capability actually retired — an old shared script nobody had loved in years — with engineers assigned to help migrate its users
That last item is the step everyone skips. Retiring a capability takes hands, not announcements: passive notices almost never move anybody, a lesson Google’s own engineers recorded after watching advisory migrations succeed where banners failed. Then, like any sensible discharge, came the follow-up appointments: service-level attainment reviewed monthly, adoption and satisfaction tracked, and the Ops Atlas PlatformOps guide circulated to the wider team as the reference for the practices we had just lived through.
The Recovered Gait, One Year On
A year later the numbers read plainly: self-service requests up, time to first deployment for a new service down from eleven days to under two, and the platform team answering for its components while application teams answer for theirs. Nothing about it required heroics — only the discipline of small, supported, measured steps, the kind every good recovery plan is made of. If the parallels have you thinking about your own team’s aches, the way we separate symptom from cause in the everyday causes of hip pain rhymes with how those five interviews separated real friction from folklore. And choosing where to have the work done matters in both worlds; our checklist for picking an orthopedic hospital for chronic hip pain argues, step by careful step, for evidence over advertising — which is also the whole philosophy of a golden path.
