DropboxDaniel Reiter Horn,Rich Chan,Pravi Garg,Sourabh Yerfule,Gennadiy CivilExplainer
How AI-assisted development helped us build the new Dropbox experience
Summary
Dropbox is rolling out a new experience that helps users understand files and perform actions like organizing folders or combining documents into PDFs, with changes that can be undone. To meet a November launch deadline, engineers used two integration approaches: embedded web views for web parity and direct React component import for control, with AI helping trace dependencies. Coding agents were delegated defined tasks such as implementing 22 accessibility features, updating tests for five core web components, and retiring an older file list by completing 53 AI-handled tasks in one week. Engineers reviewed all agent output, and an agent discovered a Chat-history bug that was fixed before release. Customer feedback from eight design partners and usage signals helped prioritize improvements, with the agent workflow providing a consistent way to implement fixes.
Context
Dropbox had been working for three years on a new experience to help users understand and work with their files. They faced complex integrations to bring web features into the desktop app, which previously focused on syncing, and needed to meet a November launch deadline. AI-assisted development became a key enabler to complete work that would have been difficult otherwise.
Approach / What changed
Engineers used a prototype during Hackathon to test two integration approaches: embedded web views via Electron for complete page parity, and direct import of web React components for finer control. They established shared React components with platform-specific adapters for networking and services. Coding agents were delegated defined tasks with written plans; agents made changes, ran builds and tests, and engineers reviewed. A working group established shared instructions and task formats. This was used for 22 accessibility features, interface changes, automated test updates, and cleanup tasks such as retiring an older file list via 53 AI-handled tasks completed in one week.
Takeaways
AI coding agents handled 53 tasks to retire an older file list and update its tests in one week, a cleanup deferred for two quarters due to testing effort.
Engineers used two approaches for integrating web features into desktop: embedded web views for complete page parity and direct import of React components for finer control.
An AI agent found that a deleted item in Chat history remained visible until page refresh; engineers fixed the sync issue and added tests to prevent regression.
FAQ
How did Dropbox ensure code quality when using AI agents?
Engineers reviewed all AI-generated code. Each task started with a written plan agreed on by the engineer and agent; agents ran builds and tests locally before engineers accepted the code. Automated tests for five core web app parts were added to catch issues early.
What AI-related accessibility features were delivered?
Engineers delivered 22 accessibility features across Home, All Files, file previews, and Photos. These included screen reader announcements for the Up Next carousel, improved keyboard focus navigation, and layouts that remain usable at 200%–400% zoom.
How did Dropbox reuse code across web and desktop?
They share React components along with the data-access code. Each platform builds and tests those components separately. When a shared feature needs data, it requests it the same way; web handles it via browser, desktop via its own networking, using adapters written with AI help.
Source and freshness
- Source
- dropbox.tech
- Published
- Added to Yomu
Summaries are generated from the original by a language model. Check the original before relying on it.
Keep reading
Related reading
The Good Stuff
Paid picks we use ourselves. Links carry a referral code — a subscription through one earns us a commission at no extra cost to you.
The best engineering reads, once a week
One useful email. Unsubscribe anytime.