Over 60 days. That's how long it took a new Spotify engineer to make their tenth pull request at one point during the company's hiring boom. After Backstage and its golden paths, Spotify reports the figure dropped to 20.
Your numbers will be different, but the shape is probably familiar. Day one is laptops and access requests. Week one is a wiki that contradicts itself. Somewhere in week three the new hire ships a small change, after asking five people how the pipeline, secrets and staging actually work.
Golden paths fix this by giving developers one supported, documented way to do the most common jobs. Most attempts stall because they try to pave everything at once. The ones that work start with one or two painful journeys, not thirty, and measure the result.
What slow onboarding really costs
Every week a new hire spends working out how things fit together is a week of salary with little delivery. It also pulls your most experienced engineers into explaining the same things again. Neither cost shows up on a dashboard, which is why it lingers.
The industry has noticed. Google's announcement of the 2025 DORA report found that 90% of organisations have adopted at least one platform, and linked platform quality to how much value organisations get from AI. DORA's platform engineering guide puts golden paths at the centre of an internal developer platform, and says the platform should absorb infrastructure, networking and security complexity so developers don't need deep expertise in each.
What golden paths are, and what they aren't
Spotify popularised the term. Its 2020 engineering post describes the golden path as the "opinionated and supported" way to build something, such as a backend service, a website or a data pipeline, with a tutorial that walks you through it. The problem it solved had a memorable name: "rumour-driven development", where you only found out how to do things by asking around.
A good golden path is:
- Opinionated. It makes the routine decisions for you: language version, base image, pipeline, logging library, deployment target.
- Supported. A named team maintains it and answers questions about it.
- Optional. Teams can step off it when they have a good reason, and they take on the extra maintenance when they do.
It isn't a mandate. DORA's guide warns against imposing rigid standards on teams with different needs; a data science team and a mobile team won't fit the same template as a backend API. If developers route around the path, treat it as user feedback, not disobedience.
Picking the first one or two journeys
Don't catalogue every workflow. Look for the ones with the highest combined frequency and pain:
- Which tasks do new engineers ask about most in their first month?
- Which tasks generate the most tickets for the platform or DevOps team?
- Which tasks take days of waiting rather than hours of work?
- Which tasks vary most between teams?
- Which tasks carry the most security or privacy risk when done inconsistently?
Two journeys come up in nearly every assessment we run: "get a working dev environment and ship a first change" and "create a new service and deploy it to production". Both touch access, pipelines, environments, secrets and observability, so paving them removes a lot of friction at once.
Get a baseline before you build anything. Ask your last few hires how long it took to reach their first production change and their tenth merged pull request. Then time a real "new service" request from the first message to traffic in production. Write the numbers down; you'll want them in three months.
Golden paths: what a good one includes
A template repository on its own isn't a golden path. Here's what we put into a "new service to production" path.
| Component | What it provides | Typical tooling |
|---|---|---|
| Service template | Repository with build, tests, linting, Dockerfile and health check endpoint | Backstage Software Templates, cookiecutter, GitHub template repositories |
| Pipeline | Build, test, security scans and deployment from the first commit | GitHub Actions, GitLab CI, Azure DevOps, AWS CodePipeline |
| Infrastructure | Pre-approved modules for compute, database and networking | Terraform, Bicep, CloudFormation |
| Secrets and access | Secrets store wired in, least-privilege roles created | AWS Secrets Manager, Azure Key Vault |
| Observability | Logs, metrics and traces with standard dashboards and alerts | OpenTelemetry plus your monitoring tool |
| Documentation | A step-by-step tutorial that ends with a running service | Markdown in the repo, TechDocs |
| Ownership | Service registered with an owner and on-call contact | Software catalogue |
The CNCF platforms white paper lists "golden path" templates and docs as one of 13 platform capability domains, alongside build and test automation, observability, and identity and secrets management. Use that list as a checklist when you design the next path.
One practical gotcha: templates go stale. A service created from last year's template keeps last year's base image and pipeline unless you plan for updates. Keep shared pipeline steps in a central reusable workflow or template that services reference, so a fix lands everywhere, and keep the per-repo copy as thin as you can.
Where Backstage fits in an internal developer platform
Backstage is an open source framework for building developer portals and a CNCF incubation project. Its Software Catalog tracks services, libraries and pipelines, Software Templates create new projects that follow your standards, and TechDocs keeps documentation next to the code.
It's a good front door for golden paths, but it isn't the path. The value is in the templates, pipelines and modules behind it, and Backstage itself is an application someone has to run and upgrade. For a company with a handful of teams, a well-kept template repository plus a clear tutorial often gets most of the benefit for far less effort. We usually suggest adding a portal once you have several golden paths and enough services that finding things has become its own problem.
Measuring developer experience, before and after
Track a small set of measures before and after each path ships:
| Measure | How to collect it |
|---|---|
| Time to first production change for new hires | Ask each new starter, or read it from version control |
| Time to tenth merged pull request | Version control history |
| Time to create a new service in production | Time a real request end to end |
| Golden path adoption | Share of new services created from the template |
| Developer satisfaction | A short quarterly survey |
| Change lead time and change fail rate | DORA's software delivery metrics |
DORA recommends a balanced scorecard of delivery metrics, satisfaction, adoption and task success. It also describes a J-curve: an early gain, a dip as complexity grows, then recovery to a higher level as the platform matures. Tell leadership about the dip before it happens.
How JRC Software builds golden paths
We start with a short platform assessment. In the first week we interview your three or four most recent hires and your team leads, read through the pipelines and infrastructure code, and baseline the measures above. You get a ranked list of developer journeys with a clear recommendation for the first one or two.
Then we build the paths with your engineers, not for them: a service template, a secure pipeline, infrastructure modules and a tutorial a new hire can follow alone. We pair throughout so your team can maintain and extend each path after we've gone. We'll steer you away from starting with a portal, or from paving a journey that only happens twice a year. Golden paths are also the natural place to add guardrails for AI coding tools, which we cover in preparing your platform for AI coding agents.
Still deciding whether you need a platform team at all? Start with platform engineering vs DevOps. For how we work more broadly, see our platform engineering and DevOps services.
If onboarding still takes weeks, request a platform assessment and we'll help you find the one or two golden paths that will make the biggest difference.
Frequently asked questions
What is a golden path in software engineering?
A golden path is an opinionated, supported way to do a common engineering task, such as creating a new service and deploying it to production. It bundles a template, pipeline, infrastructure, documentation and observability so developers don't have to make every decision themselves. Teams can still leave the path when they need to, but then they own the extra maintenance.
Do we need Backstage to build golden paths?
No. Backstage is a popular open source developer portal that makes golden paths easy to find and run, but the path itself is the template, pipeline and infrastructure behind it. Smaller teams can start with template repositories and a clear tutorial. A portal earns its keep once you have several golden paths and many services to catalogue.
How long does it take to build a first golden path?
Usually weeks rather than months, for a focused path on one stack, such as a containerised API on one cloud that reuses existing modules. It depends on how standardised your current pipelines and infrastructure are. The slow part is often agreeing the opinions, like which logging library and deployment target, rather than writing the code.
How do we measure onboarding improvements?
Pick a measure you can collect both before and after. Spotify used the time for a new engineer to make their tenth pull request; time to first production change is another simple option. Pair it with golden path adoption, developer satisfaction and DORA delivery metrics, so you can show that faster onboarding didn't cost you stability.
DevOps & Platforms
Request a platform assessment
Tell us where you're stuck and we'll come back with practical next steps.
Get in touchSources
- Spotify Engineering: How Backstage made our developers more effective
- Google Cloud: Announcing the 2025 DORA report
- DORA: Capabilities: Platform engineering
- Spotify Engineering: How we use golden paths to solve fragmentation in our software ecosystem
- CNCF TAG App Delivery: Platforms white paper
- Backstage: What is Backstage?
- DORA: DORA's software delivery performance metrics