DevOps promised that teams could own what they build and run. A decade later, many developers own a sprawl of YAML, dashboards, pipelines and cloud consoles. Platform engineering is the correction: give developers a paved road, and treat that road as a product.
- Platform engineering responds to cognitive overload: it packages infrastructure and delivery practices into self-service capabilities developers can use without expertise in every tool.
- The winning pattern is the golden path, an opinionated, supported route that is the easiest way to do the right thing, not a mandatory gate.
- Platforms succeed when run as products with users, roadmaps and feedback loops, and fail when run as one-off migration projects.
- Build-versus-buy is rarely binary; most teams assemble open-source components such as Backstage with commercial services and thin custom glue.
- Measure adoption and developer outcomes, not the number of features shipped.
01The problem DevOps left behind
In the early 2010s, the DevOps movement broke down walls between development and operations. Developers gained the ability to deploy their own services, and operations engineers gained a reason to automate. The cultural change was real and lasting. But the toolchain exploded alongside it. Today a developer shipping a service might touch a container build, a Kubernetes manifest, a Terraform module, a secrets manager, a service mesh, an observability stack, a feature flag system and a security scanner, each with its own documentation and quirks.
The phrase you build it, you run it turned into an unspoken expectation that every product engineer must also be a part-time infrastructure specialist. The result is cognitive overload: slow onboarding, inconsistent practices from team to team, duplicated effort and a steady trickle of security and reliability mistakes made by people who are simply too busy to learn another system.
Platform engineering emerged as a pragmatic answer. Rather than asking every team to master the whole stack, a dedicated team builds shared capabilities that other engineers consume through self-service. The reasoning echoes the Team Topologies model popularized in 2019, in which platform teams reduce the load on stream-aligned teams so they can focus on delivering value.
A good platform removes decisions that nobody wanted to make and preserves the ones that matter.
02What an internal developer platform really is
An internal developer platform, or IDP, is the sum of the tools, workflows and interfaces that let developers provision environments, deploy, observe and operate their software with minimal friction. It is not a single product you buy. It typically includes a service catalog that records what exists and who owns it, templates that scaffold new services with sensible defaults, automated pipelines, infrastructure provisioning, secrets and access management, and links into monitoring and incident tooling.
Many organizations present these capabilities through a developer portal. Spotify open-sourced Backstage in 2020, and it became a Cloud Native Computing Foundation project and the most recognizable foundation for portals. Commercial portals and cloud provider offerings have followed. The portal is the front door, but the substance lies behind it, in APIs and automation that work whether or not anyone clicks a button.
03Golden paths: the road, not the gate
The central concept is the golden path, a supported, opinionated way to accomplish a common task, such as creating a new web service with logging, metrics, alerts, a CI pipeline and a deployment target already configured. A developer who follows the path gets security patches, upgrades and operational support almost for free. A developer with unusual needs can step off the path, accepting more responsibility in return.
This balance is delicate. If the paved road is too restrictive, teams route around it and the platform becomes shelfware. If it is too loose, it provides little leverage. The best golden paths are built by observing what many teams already do, standardizing the common 80 percent, and leaving clear extension points. Crucially, adoption should come from the path being better than the alternative, not from a mandate. A platform that must force its users has a product problem.
04Treating the platform as a product
The most repeated lesson in platform engineering is the product mindset. Developers are the customers. Their time and attention are the currency. A platform team therefore needs the habits of a product team: user research, a public roadmap, clear ownership, documentation as a first-class deliverable, a support channel with service levels and a way to retire features that no one uses.
The failure mode is the migration project. A central team announces a new platform, spends a year building it in isolation, mandates adoption and declares victory at launch. Then reality arrives: the platform does not fit the needs of the three teams with the oddest workloads, documentation is thin, and nobody owns improvements. Within a year, shadow tooling reappears.
| Dimension | Project mindset | Product mindset |
|---|---|---|
| Goal | Deliver a defined scope by a date | Improve developer outcomes continuously |
| Users | Assumed from requirements | Interviewed, observed and surveyed regularly |
| Adoption | Mandated by leadership | Earned by being the easiest option |
| Roadmap | Fixed plan, change requests | Prioritized backlog informed by feedback |
| Success metric | Features delivered | Usage, satisfaction, lead time, incident rate |
| After launch | Team disbands or moves on | Dedicated team operates and evolves it |
05Build, buy or assemble
Few organizations should build everything from scratch, and few can buy everything off the shelf. Most end up assembling: an open-source portal framework or commercial portal, a managed Kubernetes or serverless runtime from a cloud provider, existing CI systems, infrastructure as code modules, a policy engine and observability tools, joined by internal glue that encodes company-specific rules.
A few questions help. Which capabilities are differentiating for your business, and which are commodity? How much platform engineering capacity can you sustain for years, not months? What will the maintenance burden of an open-source portal look like, including plugin upgrades? How strongly do you need to avoid lock-in? Smaller companies often do best with a thin layer on top of managed services, a philosophy sometimes called the thinnest viable platform, while large enterprises may justify bespoke investment because of scale and compliance needs.
06Measuring what matters
Platform teams are cost centers by default, so they need a credible story about value. The DORA research program has long linked software delivery performance to four metrics: deployment frequency, lead time for changes, change failure rate and time to restore service. Its 2024 report examined platform engineering specifically and found associations with improved developer productivity, along with caveats about throughput and stability when platforms are poorly executed. The lesson is to measure rather than assume.
Pair delivery metrics with experience metrics. Developer surveys capture satisfaction, perceived friction and onboarding time. Platform-specific indicators include the percentage of services on the golden path, time to first deployment for a new hire, provisioning lead time and the volume and age of support requests. Be wary of vanity numbers such as portal page views. A platform that people do not need to visit because automation just works may be the most successful of all.
07The AI factor
AI tooling is reshaping the platform conversation in two ways. First, coding assistants and agents need the same context and guardrails that humans do, so platform teams increasingly provide approved runtimes, scoped credentials and standardized repository instructions for them. Second, platforms can use AI to improve the developer experience itself, with assistants that answer questions about internal documentation, draft service scaffolding or summarize incidents. The caution is familiar: an assistant built on stale or inconsistent documentation will confidently repeat the mess, so the underlying catalog and docs deserve attention first.
08Common pitfalls and how to avoid them
Interview developers and map the slowest, most error-prone journeys before choosing a portal or framework.
Ship one valuable golden path, such as a new service from template to production, and expand by demand.
Make the supported path faster than the alternatives, and show the evidence in lead time and incident data.
Staff these deliberately. A feature nobody understands does not exist.
A platform needs sustained investment, with product management, engineering, and developer relations skills.
09Frequently asked questions
Is platform engineering just DevOps with a new name?
It builds on DevOps rather than replacing it. DevOps is a culture of shared ownership and automation. Platform engineering is an organizational and technical approach that packages that automation into reusable, self-service products.
How big must a company be to need a platform team?
There is no fixed threshold. Teams often feel the pain when many engineering groups repeat the same infrastructure work. Smaller organizations can start with a part-time platform function and managed services.
Do we need Backstage?
No. It is a popular option, but a portal is optional. Many teams start with templates, documentation and automation, adding a portal when discoverability becomes the constraint.
What should the first project be?
Choose the journey with the most friction and the most users, commonly creating and deploying a new service, and make it dramatically easier.
