Platform engineering: one shared platform for your teams, and where to start
Platform engineering puts the tools to ship and run software in self-service for all your development teams, with security already built in. The golden path of a web service step by step, a self-assessment to know whether the time has come, and the first brick we lay.
Create the service
A ready-to-use project template
Least-privilege access
Every change
The shared pipeline: build, tests, analysis
Secrets and dependencies checked
Test
An environment created on demand
Described as code, removed after
Release
Progressive rollout, a way back ready
Approved by a person
Run
Logs, metrics, the service’s cost
Alerts and costs tracked
Platform engineering means building and running one internal platform shared by your development teams: the tools to create, test, deliver and run an application, brought together as self-service, with quality and security already built in. A platform team maintains it as a product, whose users are your developers.
This article walks through the golden path of a web service, step by step, gives a grid to tell whether a platform would save you time, and the first brick we lay.
Platform engineering and DevOps: what is the difference?
DevOps is a culture and a set of practices: developers and operators work together, automate and release often. Platform engineering does not replace it, it organises it at scale: one team provides the others, as self-service, with what each one used to build alone. The CNCF (Cloud Native Computing Foundation) white paper defines a platform as an integrated collection of capabilities, presented according to the needs of its users.
The book Team Topologies (2019) makes the platform team one of its four team types: it provides an internal product that speeds up the teams in charge of the products.
A golden path, in practice: the example of a web service
A golden path (a term made popular by Spotify) is the recommended, supported way to build a type of software in your company, from the first commit to production. Here is the one for a typical web service, as we set it up:
Step | What the platform provides | What the developer does |
|---|---|---|
Create the service | A project template: structure, starter tests, build file, starter documentation, an owner declared in the catalogue | Picks the template, names the service |
Every change | The shared pipeline: build, tests, code and dependency analysis, secret scanning, a signed image | Opens a pull request and reads the results in their own tool |
Test | An environment created on demand for the pull request, described as code, removed after the merge | Tests and shows the change |
Release | A progressive rollout, a planned way back, a person’s approval | Approves the release |
Run | Logs, metrics, alerts and the cost of the service, tied to its team from day one | Follows the service, gets its alerts |
A new developer’s first day on this path:
- they get their access, limited to what they need;
- they create a service from the template, or pick up a ticket;
- their first change goes through the same pipeline as their colleagues’, with nothing to install by hand;
- it goes to production behind a person’s approval.
The time it takes them to ship that first change is the measure we track.
A golden path is not an obligation: a team with a specific need leaves it, visibly. If it is bypassed often, the path is what needs improving. The security checks of this pipeline, and those that block a release, are detailed in our article on DevSecOps.
Do you need a platform? The self-assessment
Answer these questions with your teams:
The question | If the answer is yes |
|---|---|
Do several teams each maintain their own delivery pipeline? | A shared pipeline gives them that time back. |
Are the same security rules rewritten project by project, or applied to some only? | Built into the pipeline once, they apply everywhere. |
Does a new developer take long to ship a first change? | A project template and on-demand environments shorten that delay. |
Does nobody know what pipelines and environments cost, or who uses them? | The cost tied to each service and its team shows it. |
Do you have only one or two teams, and nobody to own the platform as a product? | It is probably too early: start with one shared pipeline for your main team. |
Two yeses or more among the first four questions, and someone to own the platform: it is time to lay the first brick. A full portal before a real problem is solved, on the other hand, nearly always comes too early.
The first brick we lay: one shared delivery pipeline
We start with one shared delivery pipeline for a type of service you build often, with its security checks already built in and an environment created on demand. We build it with a pilot team, measure the time it takes a new developer to ship a first change, then open it to the other teams, one at a time.
Why the pipeline rather than the portal: it is what every team rebuilds, and it is where security gets lost from one project to the next. The portal comes later, once the catalogue of services grows. It is the same move as with every wave of technology: do once, for every team, what each one used to redo alone.
To see where you stand, the CNCF maturity model (November 2023, still current) describes four levels, from provisional to optimizing, across five aspects: investment, adoption, interfaces, operations and measurement.
Our offers One safe way to deliver, for all your teams, Test environments your teams create themselves, Flaws caught before release and A cloud bill you understand and control lay these bricks, and the page DevOps & releases describes our work on the subject.
What the research shows
Google Cloud’s 2024 DORA report measured the effect of internal platforms: with a platform, respondents report being 8% more productive, but delivery throughput falls by 8% and change stability by 14%, a result its authors call surprising.
The 2025 report spells out what this changes with AI: when the platform is of poor quality, adopting AI has next to no effect on the organisation’s performance; when it is of good quality, the effect is clear.
The lesson we draw: a platform has an adoption cost, and its quality decides what you will get out of it, AI included. It is built step by step, with its users, and measured, rather than imposed in one go. That is why we start with a pilot team.
Frequently asked questions
What is an internal developer platform?
The set of shared tools and services your developers use as self-service to create, test, deliver and run their applications, often called an IDP.
Does platform engineering replace DevOps?
No. It is a way of scaling it: the DevOps culture stays, and a platform team makes available to every team what each one used to build alone.
What is a golden path?
The recommended, supported way to build a type of software in your company, with quality and security already built in, so the easy way is also the safe way.
Do you need a portal such as Backstage?
Not to start with. Backstage, created at Spotify and open-sourced in 2020, is a portal: a front end for services and their documentation, which becomes useful as the catalogue grows. The first gains come from the shared pipeline, the environments and built-in security.
Sources
Written with an AI assistant from the sources cited, facts checked and reviewed by Mehdi Chafik on 26 October 2026.
- CNCF, Platforms white paper (TAG App Delivery), April 2023.
- CNCF, Platform Engineering Maturity Model, November 2023.
- DORA, Accelerate State of DevOps Report 2024, 22 October 2024.
- DORA, State of AI-assisted Software Development 2025, 23 September 2025.
- Spotify Engineering, “How We Use Golden Paths to Solve Fragmentation”, 17 August 2020.
- Backstage, open-source announcement, 16 March 2020.
- Team Topologies, key concepts.