devops · golden-paths · internal-developer-platform
By Shayan Ghasemnezhad3min read
DevOps gave teams ownership. Platform engineering gives them leverage. The distinction matters for how you structure teams and invest in tooling.
DevOps told engineering teams: you build it, you run it. The intent was good—break down silos between development and operations, speed up feedback loops, improve accountability. What actually happened in many organisations was that every team became responsible for its own CI/CD pipelines, Kubernetes manifests, observability stack, and incident response. The cognitive load shifted from a central ops team to every developer. Platform engineering addresses the unintended consequence.
A product engineer writing a payment service should think about payment logic, not about how to configure a Kubernetes NetworkPolicy or tune a Prometheus alerting rule. When every team owns its infrastructure end-to-end, the cognitive load of operational concerns crowds out the domain thinking that creates product value. Teams spend 30–40% of their time on platform tasks that are undifferentiated but necessary.
Platform engineering inverts this: a dedicated team builds the internal platform that product teams consume. The platform provides golden paths—paved roads for common workflows (deploy a service, set up monitoring, provision a database) that handle the operational complexity so product teams do not have to.
An internal developer platform is not a tool. It is a product—built by a platform team, consumed by product teams. It includes self-service capabilities (deploy, scale, observe), documentation (how to use the platform, not how to configure Kubernetes), and an interface (CLI, portal, or API—whatever reduces friction).
Critically, the platform is optional. Golden paths are the recommended way, not the required way. Teams with unique requirements can step off the path, but most workloads should not need to. If teams regularly bypass the platform, the platform is not meeting their needs—that is a signal to improve it, not to mandate its use.
You need a platform team when at least three of these are true: you have more than five product teams, each team is maintaining its own deployment pipeline, onboarding a new engineer takes more than two weeks, teams are solving the same infrastructure problems independently, or incident response varies wildly in quality across teams.
Below this threshold, a shared set of templates and runbooks may suffice. A platform team without sufficient consumers is overhead. A platform that serves 15 product teams is leverage.
Start with the pain. Survey product teams: what takes too long, what breaks too often, what requires too much specialised knowledge? The answers will cluster around three areas: deployment (CI/CD pipelines), observability (logs, metrics, traces), and environment provisioning (databases, queues, caches).
Build the golden path for deployment first. It is the most frequent workflow and the one where inconsistency causes the most incidents. A standardised deployment pipeline with built-in rollback, canary analysis, and notification reduces both deployment time and deployment risk.
Platform teams that measure success by the number of features they ship are optimising for the wrong thing. Measure by the outcomes your consumers experience: deployment frequency (should increase), lead time for changes (should decrease), time to onboard a new engineer (should decrease), percentage of teams on the golden path (should increase), and platform-related incidents (should decrease).
The biggest failure: building a platform that product teams are forced to use. Mandated adoption without product quality creates resentment and workarounds. Treat the platform like a product: talk to your users, prioritise their pain points, iterate based on feedback.
The second failure: staffing the platform team with infrastructure engineers who do not talk to product teams. A platform team that builds in isolation produces tools that are technically excellent and practically useless. Embed product thinking into the platform team: user research, adoption metrics, and regular feedback loops.
Platform engineering is the mature evolution of DevOps: keep the accountability, remove the unnecessary cognitive load. Build it when the organisation is large enough to justify a dedicated team, measure it by developer productivity, and iterate it like any other product.
