Why Does Platform Engineering Keep Recreating the Same Bottleneck?

In this GOTO State of the Art interview, Abby Bangser argues that platform engineering is conceptually well understood but keeps failing the same way in practice: organizations move from DevOps to platform engineering, only to watch their platform team turn into a new bottleneck overwhelmed by the whole organization's requests — precisely the problem DevOps was supposed to solve. She traces this back to an architectural flaw: too many platforms are still built as centralized Terraform machines instead of a marketplace of composable offerings.

Her blunt 'platform as a product' test asks whether the team has ever said no to a feature request or deprecated something. If not, what they have isn't a product but a request queue — a quick way to check if a platform team truly owns its offering.

The most urgent part of the discussion concerns AI. Bangser frames AI agents as the new forcing function pushing platform maturity forward, and calls out the persistent misconception that platform engineering equals infrastructure-as-code — renaming your Terraform team doesn't count. The real goal is building a self-service, compliant, and coherent experience at organizational scale, for human developers and AI agents alike. The Team Topologies interaction modes, ranging from high-collaboration to fully automated on-demand APIs, offer a practical health check for where a platform actually sits on that maturity curve.