Platform engineering & Backstage · Lesson 4 of 5 · 4 min read
Building the platform as a product
Platform as a product
The most important idea in platform engineering: treat the platform as a product and developers as customers.
- Understand users: interview teams, watch them onboard, measure where time goes. Build what removes the most pain, not what's most interesting to build.
- Make adoption voluntary where possible: if the golden path is genuinely the easiest way, teams choose it. Mandates hide whether the platform is good, and breed workarounds.
- Documentation and support are product features: quick starts, examples, an office-hours slot, a support channel with response expectations.
- Roadmap and communication: publish what's coming, collect requests, announce changes and deprecations early, version interfaces and give migration paths.
- Reliability: the platform is production for every team. It needs SLOs, on-call, change management and upgrade testing like any critical service.
Enlarge diagram
Swipe sideways to see all of it, or tap to open it full screen.
Measuring success
| Measure | Examples |
|---|---|
| Delivery performance (DORA) | Deployment frequency, lead time for changes, change failure rate, time to restore service |
| Onboarding | Time for a new engineer to make a first production deploy; time to create a new service |
| Adoption | Share of services on the golden path, templates used per month, catalogue coverage |
| Developer experience | Regular surveys (satisfaction, perceived friction), as in the SPACE framework |
| Toil | Tickets to the platform team, time teams spend on infrastructure tasks |
| Platform reliability | SLOs for CI, clusters, deploy pipelines |
Combine numbers with feedback: a platform with high adoption but low satisfaction is being tolerated, not loved.
Anti-patterns
- Ticket-ops: the "platform" is a queue where developers request things and wait days. Self-service is the point.
- Building everything: writing your own CI system or deployment tool when mature products exist. Assemble, integrate and add the glue.
- Over-abstraction: a custom YAML format that hides Kubernetes completely, so nobody can debug, and every new need requires the platform team.
- Big bang: a year building in isolation, then a launch nobody adopts. Ship small, with a pilot team, and iterate.
- No owner for golden paths: templates rot, and new services start on outdated versions.
- Mandates without value: forcing migration to a platform that is harder than what teams had.
Enlarge diagram
Swipe sideways to see all of it, or tap to open it full screen.