Vibeleaderboard
← All Intel
Intel / article

How AI-assisted development helped us build the new Dropbox experience

Source
dropbox.tech
Author
Daniel Reiter Horn,Rich Chan,Pravi Garg,Sourabh Yerfule,Gennadiy Civil
Date

Context

Dropbox says AI-assisted development helped it meet the launch date for its new Dropbox experience largely by handing coding agents defined tasks, including implementation and testing, for engineers to review (Dropbox tech blog). The new experience, which Dropbox began rolling out on November 5, 2026, lets customers ask questions about their files and have Dropbox carry out tasks such as organizing a folder or renaming files, after showing a plan they can edit and approve. A companion post explains how that file-changing is built and kept within the permissions people grant (agent architecture). Shipping it meant bringing features built for the web into the desktop app before a November deadline, while also finishing accessibility and testing work.

The working method is the part Dropbox describes in most detail. Each task started with a written plan that the engineer and the agent agreed on. The agent made the change, ran the relevant builds and tests locally and recorded the results, and an engineer reviewed them before accepting the code. Instructions stored alongside the code told agents which rules to follow, and a working group with champions across teams set shared task formats and review expectations so the approach could be applied consistently across the project.

Dropbox reports that this process delivered 22 accessibility features across Home, All Files, file previews and Photos, including screen-reader announcements and layouts that stay usable at 200% to 400% zoom. It also cleared a backlog item: retiring an older file list had been deferred for two quarters because updating its automated tests required substantial engineering time, and the team split the cleanup into 53 tasks that agents worked through in one week. Agents also added automated tests for five core parts of the web app and, by trying the product during development, found a bug where a deleted Chat history item stayed visible until the page was refreshed. These are Dropbox's own figures for its own project, with no measured before-and-after comparison. The customer evidence it offers is that all eight initial design partners chose to keep the new experience enabled.

Key takeaways · AI-distilled
  • Dropbox says its new experience answers questions about project folders with sources users can open, and for tasks like organizing, renaming, or combining files into a PDF it first shows a plan users can edit and approve; supported file changes can be undone.
  • To bring web features into the desktop app, a May hackathon prototype tested two routes: whole web pages in an Electron web view, or importing individual web components. Dropbox kept both, using full pages for close parity and components where it needed more control.
  • Web and desktop now share React components and data code, while each app keeps its own sign-in, navigation, and networking. Dropbox says AI helped engineers trace dependencies between the pieces and write the adapters that connect shared features to each platform's services.
  • Dropbox's coding agents worked from a written plan agreed with an engineer, ran builds and tests locally, and recorded results for engineer review, with the agents' rules stored alongside the code. The team used this process to ship 22 accessibility features before launch.
  • A file-list retirement deferred for two quarters because of test rewrites was split into 53 tasks that coding agents finished in one week. Agents trying the product during development also caught a bug where deleted chat history stayed visible until a page refresh.
Terms in this piece · Glossary
  • AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
  • embedding — A list of numbers representing a piece of text's meaning, so that similar meanings end up numerically close and can be searched.
Read the source dropbox.tech
Recommended reads
Comments

Checking sign-in…

Loading comments…