Internal Platform Rebuild
A two-year rebuild of the internal developer platform — fewer surfaces, named owners, and a deploy story product teams could actually use.
- Role
- Platform Lead
- Year
- 2025
- Stack
- TypeScript · Go · Kubernetes · PostgreSQL
Outcomes
- Reduced average deploy time from 38 minutes to 4.
- Cut on-call incident volume by 41% over two quarters.
- Consolidated three deployment pipelines into one team-owned interface.
Context
The platform had grown by accretion. Each new product team added one more deployment story to the pile. By 2024 there were three live pipelines, two shadow ones, and an unspoken convention about which projects used which. Nobody owned the seam.
What we changed
- Named a single platform interface: one CLI, one config schema, one deploy command. Everything else became transitional.
- Made the transition real with a deprecation calendar that survived three reorgs because it was published, dated, and stapled to a runbook.
- Built operator-facing UX with the same review standard we applied to product UX. A new SRE could recover a production incident in under ten minutes from a cold start.
What I'd do differently
I'd start the deprecation calendar earlier. We spent six months building the new thing before we publicly committed to retiring the old ones. During that window every team made one more workaround.