Platform engineering shifts to golden bricks and agent-safe environments

Abby Bangzer of Syntasso argues that platform engineering is moving from rigid "golden paths" to "golden bricks," a composable model that allows teams to swap individual components without rebuilding entire workflows. She contends that traditional golden paths fail because organizations do not document compliance checklists for off-path deviations and build systems assuming static correctness. Bangzer emphasizes that AI agents should be integrated into existing team topologies as "stream aligned" or "platform" team members, rather than replacing human structures. Citing an engineer from Meta at FastFlowConf, she notes that agents work best within bounded, safe environments with clear constraints and least-privileged permissions. She also describes a "marketplace" model for internal platforms, where specialists contribute tools via standardized APIs such as OpenAPI v3 and maturity zones to manage shadow IT and scale expertise without central bottlenecks.

Bangzer characterizes AI agents as "faster consumers" of platform services that lack human empathy or the ability to negotiate roadblocks. She warns that as agent usage scales, the first systems to break are maintenance processes that rely on manual coordination or tickets, as platforms cannot scale if every consumer must be involved in every change. If an AI agent hits a roadblock, it will bypass it rather than wait for human intervention. Bangzer advises organizations to measure business impact by tracking the time required for developers to obtain tools and the time required to maintain resources during security patches. She recommends that platforms expose core business logic via APIs rather than embedding it in specific interfaces like command-line interfaces, graphical user interfaces, or Model Context Protocol servers. This approach allows for flexible consumption by both humans and agents without duplicating logic across channels.