Two days, built for engineers.

Micro Frontends: From Savior to Dragon — Our Journey

Amir Eslamdoust

In 2022 we had one feature that needed to live in three applications: our B2B shop, our backoffice, and our merchant center. Micro frontends became our solution, built with Webpack Module Federation. Making it work took more than the remote. Every host had to stay pinned to the same React and Webpack major, and each grew its own plumbing to pass a feature flag across the boundary. Then shared dependencies became a runtime contract between host and remote. Upgrading our design system stopped being a local change, so it mostly stopped happening. Our design system now sits at five different versions across two majors, and the only application on the current major is the one that walked away — in 2024, dropping the remote during an unrelated refactor. The other two still consume it, pinned to two different versions. The last commit in that repository is called "rollback deployment," and in 2026 we are still shipping components named "Legacy" to clean up after it. Nobody ever held a meeting to abandon that architecture. It was abandoned anyway — one host at a time, one skipped upgrade at a time. This year we finally have a plan to remove what is left, with AI-assisted tooling helping us map dependencies and sequence the work. This is that system, told with the commit history open: why it made sense at the time, what it cost to keep, and what removing it takes. The dragon was never micro frontends. The dragon is any architecture that outlives the conditions that justified it.