Chase Roberts, COO of Northflank, says platform-as-a-service products failed by ignoring varied workloads and bring-your-own-cloud, so teams outgrow them and drop to raw AWS or GCP infrastructure.
An enterprise spent three years building an internal platform only to find developers needed six months to use it, a warning for build-versus-buy decisions.
Roberts says most developer platforms are built as if they fear developers, and that the main blocker to buying instead of building is cultural and must come from the top down.
Why Building Your Own Developer Platform Usually Backfires
Chase Roberts argues that most companies should buy a developer platform instead of building one. Internal platforms rarely raise revenue, take years to build, and carry a lasting maintenance burden without market pressure to improve. Teams asked to do more with less, such as one or two platform engineers serving up to 50 developers, increasingly want to buy. A buyable platform that works across clouds could also weaken provider lock in.
Takeaways
Internal platforms rarely pay for themselves. Building a platform usually does not let a company charge more for its products. Buyers say millions invested bring no pricing power. Roberts frames build versus buy around three tests: pricing power, developer adoption versus complaints, and complexity as a maintenance tax.
Golden paths hide permanent complexity. Platform teams stitch together open source tools, YAML, Terraform and custom code that only the authors understand. One enterprise spent three years building a platform, and developers still needed six months to use it. Roberts's CEO calls such teams a forlorn hope.
Platform teams land in four states. Roberts says teams end up delayed, unadopted, replatforming or in nirvana. He doubts anyone claiming nirvana, meaning minimal maintenance and 90% of workloads on their own platform. An internal platform is a product needing iterations, bug fixes, documentation and upkeep.
Old PaaS missed workload variety and own-cloud needs. Heroku and Cloud Foundry nailed developer experience but handled few workload types and offered little bring your own cloud. Teams adopted them to start fast, then outgrew them and moved to AWS or GCP. Roberts counts databases and scheduled jobs as workloads.
Culture, not technology, blocks adoption. Roberts says the main blocker is cultural and must be solved from the top, since buying software seems less noble than building it. Engineering teams reason that because they can, they should. Internal tools lack market discipline, so mediocrity survives.
Ease of use and flexibility are not a trade-off. Roberts says most platforms act afraid of developers and hide primitives behind abstraction. Northflank aims to make building, deploying, running, scaling and observing feel like one system. Multi cloud is now common, and every deployment change must be repeated per cloud.
Buyable platforms could commoditize the cloud. Kubernetes promised portability, but providers create lock in that gives their infrastructure gravity over workloads. Roberts argues that containers and cloud native standards now allow a bought platform where teams define a deployment path once, making the cloud a commodity.
Tap a name to go there · tap a line to see why they are linked