Standing up a platform team where there wasn't one

Forty engineers, a wiki nobody read, and a single staff engineer who had become the human service catalog. New services took days to scaffold and three different teams disagreed on what 'production-ready' meant.

1. 01

   Ran a two-week discovery: shadowed three teams, mapped how a change actually reached production, and wrote down every implicit rule that lived in someone's head.

2. 02

   Picked Backstage, but only after agreeing with the CTO that we would not customise the React layer for at least six months. Boring is a feature.

3. 03

   Built one Go service template and one TypeScript service template wired with CI/CD, OpenTelemetry, secrets via External Secrets Operator, and a runbook stub. Made it the only sanctioned way to start a new service.

4. 04

   Wrote three Crossplane compositions for the resources teams actually requested weekly: a Postgres database, an S3 bucket with sane defaults, and a Redis instance.

5. 05

   Ran the hiring panel for the platform team in parallel — JDs, screens, technical interviews — and handed over a working platform plus a documented six-month roadmap on the last day.

A platform team of two now owns what was delivered. Service creation became a non-event; the staff engineer who had been the bottleneck went back to product work. Six months on, the team had added two more compositions without my involvement.