The Temptation to Overbuild
When a Platform team forms, there is a powerful temptation to build a comprehensive, enterprise-grade platform before anyone uses it. This almost always fails. The team spends 18 months building something nobody asked for, the Stream-aligned teams work around it, and the platform is declared a failure.
Team Topologies advocates for a Thinnest Viable Platform (TVP): the smallest set of APIs, documentation, and tools that meaningfully reduces cognitive load for Stream-aligned teams right now.
What TVP Looks Like in Practice
The TVP is not a product vision — it is a starting point. It evolves based on what Stream-aligned teams actually need.
Stage 1 (Early): The TVP might be nothing more than a wiki page listing approved AWS services, a Terraform module for a standard VPC, and a Slack channel where teams can ask infrastructure questions. That's it. If that reduces cognitive load, it's a valid platform.
Stage 2 (Growing): As patterns emerge, the Platform team adds a self-service database provisioning tool, a standard CI/CD pipeline template, and a Kubernetes namespace provisioning script.
Stage 3 (Mature): The platform has a developer portal, golden paths for common workflows, automated compliance checks, and a self-service catalog. But every feature was added because a Stream-aligned team needed it — not because it looked good on a slide.
Why "Thin" Is a Feature, Not a Limitation
- Avoids bloat: Don't build features nobody needs. Every unused feature is maintenance burden.
- Reduces Platform team cognitive load: The Platform team also has cognitive limits. A monster platform drowns its own builders in maintenance work.
- Ensures agility: A thinner platform is easier to change as requirements evolve. A 50,000-line Terraform monolith is not.
- Faster time to value: Stream-aligned teams get something useful in weeks, not years.
The TVP principle is essentially the MVP (Minimum Viable Product) concept applied to internal platforms. Ship the smallest thing that reduces pain, measure whether it works, and iterate.