Leadership asks whether to build an internal developer platform or buy one. Twelve teams, forty services, growing. How do you frame the decision?
Show the full answer Hide the answer
Reject the framing as a binary
Nobody builds or buys a whole platform. A platform is an assembly: source control, CI, artifact registry, infrastructure provisioning, runtime, secrets, observability, service catalogue, and the glue that makes them one experience.
Most components are bought or adopted from open source. The glue is always built, because it encodes your organisation's conventions, and no product knows them.
So the real question is: which components, and how much glue.
What to build
The golden path — the end-to-end journey from "I need a service" to "it is in production with monitoring, alerting, CI, secrets and an on-call rota". That sequence is specific to your conventions and it is where the value is.
The service templates that encode those conventions.
The catalogue schema — what you record about a service and what you do with it.
What to buy or adopt
Anything that is a solved commodity: CI execution, artifact storage, secret management, observability backends, the container runtime. Building these is a multi-year commitment to a problem someone else has already solved better.
A portal is genuinely borderline. Open-source options exist and are extensible; the cost is that extending them is real engineering. A wiki page plus a command-line tool is a legitimate starting point and defers the decision until the golden path is proven.
The number that decides the scale
Not team count — the ratio of platform engineers to product engineers. At twelve teams you might justify three or four platform engineers. Three engineers cannot build a portal, maintain forty service templates and operate a Kubernetes fleet. Something has to be bought, and the choice of which is a capacity decision rather than a preference.
How to start
With one golden path for the most common service shape, used by two willing teams, measured by time-to-first-deploy. Expand from evidence.
The failure mode to avoid is building a platform for the estate you imagine. Build for the two teams in front of you, and let the third tell you what is missing.