Google went from fragmented vim/emacs/IntelliJ usage to 80% of google3 development in Cider V (VSCode frontend + proprietary backend) by 2023.
Key Takeaways
Monorepo scale broke traditional IDE assumptions: local indexing, local builds, and local analysis all fail at google3 size, requiring a dedicated backend language graph updated per commit.
Cider started ~2016 as a browser editor for technical writers, gained traction after adding LSP-based code completion against a pre-indexed backend, then replaced its custom frontend with VSCode in 2020.
Switching to the VSCode frontend gave the team a mature extension ecosystem and unblocked ~100 internal extensions within two years, removing the Cider team from the critical path for every workflow improvement.
Convergence on one platform directly enabled AI features like ML-assisted code review comment resolution and Smart Paste; a fragmented IDE landscape would have made those impractical to ship.
A team of roughly a dozen engineers spent two years building Cider V; the article argues this was a small investment given leverage over all of google3 developer productivity.
Hacker News Comment Review
Xooglers push back on the narrative: large C++/Python populations stayed on vim and emacs throughout, and the IDE fragmentation problem was mostly acute for Java, not the whole company.
Java developers faced a partially forced migration rather than organic adoption; IntelliJ internal plugin support was dropped, generating resistance over missing power-refactoring features in Cider V.
Commenters note Google’s Windsurf acquisition and the removal of AI Ultra from Workspace plans as context for how Google’s internal IDE investment is not translating cleanly into external products.
Notable Comments
@dobx2010: Java teams were pushed to Cider V after Google stopped supporting IntelliJ internal plugins; some still use Android Studio for non-Android Java work.
@StilesCrisis: Android and Chromium workflows remain outside Cider V/Critique, meaning the “all Googlers” framing in the article overstates actual coverage.
