Leading a Platform Team Is Mostly a People Problem

Most engineering leadership content is written as if the job is a technical problem with people attached. In practice, running a small team against a 24/7 production platform is the other way round: the technical decisions are usually the easy part. What’s hard is that everything — an incident, a scope disagreement with another team, someone’s rota swap request, someone having a genuinely bad week — arrives in the same inbox, at the same priority, and it’s on you to sort out which is actually urgent.

With eight people covering round-the-clock production ownership, here’s what actually holds up.

On-call has to be designed, not just scheduled

A rota is not an on-call practice. The rota is the easy 10% — a calendar with names on it. The actual practice is everything that determines whether being on that calendar is sustainable:

Alert quality is a leadership responsibility, not an SRE nice-to-have. Every page that isn’t actionable is a tax on whoever’s carrying the pager, and it compounds — a noisy on-call week doesn’t just cost sleep, it teaches people to under-react to real pages. If your team is silencing or snoozing alerts, that’s not a discipline problem, it’s an alert-quality problem, and it’s yours to fix.
Primary and secondary, always. A single on-call engineer with no backup means every rota week has a single point of failure that happens to be a human. Secondary on-call isn’t redundancy for its own sake — it’s what lets primary actually rest, and what stops a bad incident from becoming a bad week.
Compensate the disruption, not just the hours. Time-in-lieu, an explicit “quiet morning after a night page” norm, whatever your organisation supports — the point is that carrying the pager has a real cost, and pretending it doesn’t is how you lose people who’d otherwise stay.
Rotate the unpleasant slots too. If the same one or two people always seem to draw the worst on-call weeks, that’s not the rota “just working out that way” — check it.

Blameless retros only work if you mean it

Everyone says “blameless postmortem.” Whether it’s actually blameless shows up in one moment: what happens the first time someone admits, in the retro, that they made the call that made things worse. If the response is curiosity (“what did you know at the time, what would have told you otherwise?”) people keep bringing you the early, ambiguous signals. If the response is anything that sounds like blame — even mild, even well-intentioned — they start quietly handling things themselves before they escalate, and you lose visibility exactly where you need it most: early, while a problem is still small.

Triage the inbound before it triages you

Leading a small team that owns critical infrastructure means you’re the natural landing point for every adjacent team’s request, every escalation, every “quick question” — on top of your own team’s actual work and your own technical judgment being needed on the hard calls. Left unmanaged, this is the single fastest way to burn out a lead: not the incidents, the constant context-switching between them.

What’s worked:

A visible intake, not an open door. “Ping me anytime” sounds helpful and is actually a tax on your team’s focus time and your own. A shared channel or ticket queue that the whole team can see and pick up from spreads the load and stops everything funnelling through one person by default.
Office hours for the ambiguous stuff. Not every cross-team question needs an instant answer. A standing slot for “bring me the things that don’t fit anywhere else” protects the rest of the week for deep work, for everyone.
Learn to push back on scope, explicitly. “Yes, and here’s what drops to make room” is a complete sentence. A team that owns 24/7 production for a regulated platform has a legitimate, standing reason to say no to work that isn’t safety- or reliability-related — use it.

Upskilling has to be protected, not scheduled around

On a team this size, everyone ends up as the de facto expert on something — a specific integration, a gnarly piece of legacy automation, the one system nobody else wants to touch. That’s efficient right up until that person is on leave, or leaves the team, and it turns out you had a bus factor of one on something load-bearing.

The fix isn’t a training budget line item, it’s deliberate redistribution of ownership:

Pair on the on-call-critical stuff, not just the interesting stuff. The system with the worst documentation is usually the one that most needs a second person who can operate it under pressure at 3am.
Rotate ownership of recurring pain points, not just new projects. Handing someone the thing you’d normally just fix yourself is slower once and faster for the rest of the team’s life.
Treat “I had to write it down to hand it off” as the actual test of whether something is documented. Documentation written for yourself and documentation written to hand off a real on-call responsibility are different documents — only the second one gets tested under pressure.

The personal side doesn’t fit in a 1:1 template

Standard 1:1 templates are built around status and career growth, and those matter — but on a team that’s carrying production risk, people also bring you things that don’t fit that template: a rough on-call week that coincided with something going on at home, burnout that’s presenting as “fine” in standup and not fine at all in the 1:1, a request to swap a rota slot that’s really a request for you to notice something’s wrong. None of that shows up if the only question you ask is “how’s the sprint going.”

The practices that actually surface it are unglamorous: consistent 1:1 cadence that doesn’t get cancelled first when the calendar gets tight, starting with an open question instead of a status update, and paying enough attention over time to notice when someone’s normal baseline has shifted. None of it is technical. All of it determines whether the technical practices above — the rota, the alert quality, the blameless retro — actually hold up when it’s a real person carrying them at 3am.

The platform is the easy part. It doesn’t have a bad week.