Overview
Dropbox adopted Cursor as its primary AI coding tool, indexing over 550,000 files across its monorepo to build an AI-native software development lifecycle. With 90% engineering adoption and over 1 million lines of AI-generated code accepted monthly, the article details how organic grassroots adoption, leadership buy-in through hands-on experience, and large-scale codebase indexing drove measurable improvements in engineering velocity and PR throughput.
What You'll Learn
1
How to drive organic AI tool adoption across a large engineering organization through AI champions and friction removal
2
Why leadership hands-on experience with AI coding tools is critical for company-wide adoption
3
How semantic codebase indexing enables AI tools to reason across massive monorepos with hundreds of thousands of files
4
How to measure the impact of AI coding tools on engineering velocity using PR throughput and cycle time metrics
Prerequisites & Requirements
- Basic understanding of software development lifecycle (SDLC) concepts
- Familiarity with monorepo architecture and large-scale codebases(optional)
- Experience with AI-assisted coding tools or IDE extensions(optional)
Key Questions Answered
How does Dropbox use Cursor to manage its massive monorepo with over 550,000 files?
Cursor indexes Dropbox's entire monorepo by scanning each non-ignored file, breaking code into structured chunks, and generating semantic embeddings that capture relationships between code pieces. This semantic index gives AI models the context needed to follow the codebase's structure and generate changes that fit naturally within it, making the codebase more navigable for engineers as well.
What adoption rate has Dropbox achieved with AI coding tools?
More than 90% of Dropbox engineers now use AI tools weekly, with Cursor as the primary driver. Engineers accept over one million lines of AI-suggested code every month. This high adoption was achieved through organic grassroots experimentation, an AI champions program, friction removal in the signup process, and leadership endorsement following a company-wide hackathon.
How did Dropbox drive AI coding tool adoption without a top-down mandate?
Adoption started organically in 2024 when engineers began experimenting with Cursor and sharing findings through Slack and internal write-ups. CTO Ali Dasdan nurtured this by creating a group of AI champions who amplified best practices while removing barriers to access. Making sign-up feel like a single click accelerated adoption further, and a company-wide hackathon in April 2025 brought leadership on board.
Why should engineering leaders personally try AI coding tools before rolling them out?
According to Dropbox CTO Ali Dasdan, firsthand experience is essential because it makes the impact tangible and credible. During a hackathon, Dasdan built an entire project in about two hours using Cursor despite starting the night before it was due. He found that many peer CTOs hadn't tried AI tools at all, and argued that even a single leader testing the tools can immediately see and advocate for their impact.
How does codebase indexing with Cursor work for large-scale monorepos?
Cursor's indexing process scans each non-ignored file, breaks the code into structured chunks, and generates embeddings that capture semantic relationships between code pieces. This creates a searchable semantic index that AI models reference when generating or editing code. At Dropbox's scale of 550,000+ files, this step was critical for enabling context-aware suggestions that fit the existing codebase structure.
What measurable impact has Cursor had on Dropbox's engineering velocity?
Since adopting Cursor, Dropbox's PR throughput and cycle time have moved into the upper tier of industry benchmarks. Engineers accept over one million lines of AI-generated code monthly. The tool appears in nearly every development step including writing, reviewing, testing, documentation, and migrations. Dropbox measures these improvements through an internal framework emphasizing speed, effectiveness, and quality.
Key Statistics & Figures
Files indexed across Dropbox's monorepo
550,000+
Cursor indexed the entire Dropbox monorepo
Lines of Cursor-generated code accepted monthly
1,000,000+
AI-suggested code accepted by Dropbox engineers every month
AI tool adoption across engineering organization
90%+
Percentage of Dropbox engineers using AI tools weekly
Requests per second served by Dropbox infrastructure
300,000+
Dropbox's own data centers serving production traffic
Time for CTO to build hackathon project with Cursor
~2 hours
Ali Dasdan built a 'smart finder' project from scratch the night before it was due
Technologies & Tools
Some links below are affiliate links. We may earn a commission if you make a purchase.
AI Coding Tool
Cursor
Primary AI-assisted coding IDE used across Dropbox's engineering organization for code generation, review, testing, documentation, and migrations
Communication
Slack
Internal channel where engineers shared early experiences and learnings about AI coding tools
Key Actionable Insights
1Create an AI champions program to accelerate organic tool adoption. Identify engineers who are already experimenting with AI tools and empower them to share their workflows through internal write-ups and Slack channels, rather than mandating adoption top-down.Dropbox's CTO noticed early organic activity and nurtured it by creating a formal group of AI champions who amplified their learnings while leadership focused on removing barriers to access.
2Remove all friction from the AI tool signup and onboarding process. Making tool access feel like a single click is critical because even small barriers can significantly slow adoption across a large engineering organization.Ali Dasdan emphasized that speeding up deployment meant removing every point of friction, and as access became easier, more engineers tried the tools and shared what they learned, creating a self-reinforcing adoption loop.
3Engineering leaders should personally use AI coding tools on real projects before making organizational decisions about adoption. A hands-on experience, even on a small hackathon project, provides credibility and conviction that no amount of reading or presentations can replace.Dasdan built a complete 'smart finder' project in about two hours using Cursor during a hackathon, and this personal experience was the catalyst for him to champion the tool more broadly, including to a group of peer CTOs.
4Index your entire codebase when deploying AI coding tools at scale. Semantic indexing allows AI models to understand codebase structure, follow conventions, and generate contextually appropriate code rather than generic suggestions.At Dropbox's scale of 550,000+ files, indexing was critical not just for code generation quality but also for helping engineers themselves navigate and understand the codebase faster, with new hires ramping up more quickly.
5Use company-wide hackathons as a catalyst for AI tool adoption. Hackathons create a low-stakes environment where engineers and leaders can experiment with AI tools on real projects, generating authentic testimonials and use cases.Dropbox's April 2025 hackathon was the turning point that brought leadership on board, with the CTO's personal experience becoming a powerful internal and external advocacy story.
6Measure AI tool impact through established engineering metrics like PR throughput and cycle time rather than relying solely on adoption rates. This provides concrete evidence of velocity improvements and helps justify continued investment.Dropbox uses an internal framework emphasizing speed, effectiveness, and quality, and since adopting Cursor, their metrics moved into the upper tier of industry benchmarks.
Common Pitfalls
1
Waiting too long to adopt AI coding tools out of caution. Some engineering leaders delay adoption because they perceive risks in moving too fast, but Dropbox's CTO argues that being slow to adopt AI is actually a bigger threat than moving too soon.
Ali Dasdan recognized early that speed is the only advantage of any company, and that AI needed to be embedded across the SDLC to achieve the velocity Dropbox was aiming for.
2
Leadership making AI adoption decisions without firsthand experience using the tools. Many CTOs and heads of engineering evaluate AI tools based on reports and demos rather than personal use, which leads to uninformed decisions and lack of conviction when championing adoption.
Dasdan was surprised to learn that many peer CTOs he met with regularly had still not tried AI coding tools, and stressed that even a single leader testing these tools can immediately see and advocate for their impact.
3
Introducing friction in the AI tool onboarding process, such as requiring multiple approval steps or complex setup procedures. Even small barriers can significantly slow adoption across a large engineering organization and prevent the self-reinforcing adoption loop from forming.
Dropbox focused on making the signup process feel like a single click, which was key to accelerating the organic adoption that had already begun among early experimenters.
Related Concepts
Ai-native Sdlc
Monorepo Architecture
Semantic Code Indexing
Code Embeddings
Engineering Velocity Metrics
Pr Throughput
Cycle Time
Developer Productivity
AI Tool Adoption Strategies
Codebase Navigation
Developer Onboarding