Dropbox uses Cursor to index over 550,000 files and build an AI-native SDLC
January 22, 2026 at 12:00 PM
Summary
TL;DR
Dropbox pushed Cursor from grassroots experiments to broad adoption, using monorepo indexing and low-friction access. The result: 90%+ weekly AI use and 1M+ AI-suggested lines accepted per month.
What actually happened
Engineers started experimenting with Cursor and shared learnings via Slack and internal write-ups
The CTO formalized momentum by creating an “AI champions” group
Dropbox reduced signup friction so access felt like “a single click”
A company-wide hackathon in April 2025 pushed leadership to try the tools firsthand
Cursor was rolled out to work against Dropbox’s full monorepo via indexing
Key numbers
More than one million lines of AI-suggested code accepted every month
More than 300,000 requests per second served from Dropbox data centers
Thousands of engineers
Monorepo with more than 550,000 files
More than 90% of engineers use AI tools weekly
April 2025 company-wide hackathon
Why this was hard
Tooling had to work in a large monorepo and demanding production environment
Value depended on reasoning across the entire codebase, not isolated files
Adoption needed to be broad; small pockets of use wouldn’t deliver the velocity goal
Any signup or access friction could stall rollout
How they solved it
Nurtured organic adoption with an internal group of AI champions
Removed barriers to adoption by making access feel like a single click
Used a company-wide hackathon to drive hands-on experience, including executives
Indexed the monorepo by scanning non-ignored files and chunking code into structured pieces
Generated embeddings from chunks to build a semantic index used during code generation/editing
Used the index to follow codebase structure and produce changes that fit existing patterns
What changed
Cursor became part of nearly every development step: writing, review, testing, docs, migrations
PR throughput and cycle time moved into the upper tier of industry benchmarks
Leaders and new hires could navigate and understand the codebase faster via the indexed view
Why this matters beyond this company
Broad adoption can be accelerated by removing access friction and amplifying internal champions
Leadership trying the tools firsthand can change the pace of adoption
Semantic indexing/embeddings are a prerequisite when AI needs full-codebase context
Stealable ideas
Create an “AI champions” group to spread patterns and unblock adoption
Make tool access feel like a single click to avoid rollout drag
Use hackathons to force hands-on evaluation by leaders
Index the whole repo to give AI consistent, codebase-wide context