AI Briefing
KOSign in

Dropbox launches AI agent experience built from two engineer side projects

·2026.11.05 19:35

Key point

The system uses isolated VMs and conversation-specific undo to let users direct file tasks while retaining control over permissions and changes.

1 / 2

Details

  • Dropbox introduces a new AI agent experience that allows users to request tasks like organizing folders and renaming files directly within the platform, built from two internal engineer experiments.
  • The agent operates in an isolated Ubuntu workspace running on Dropbox data centers, achieving a 300-millisecond connection time to user files for 75% of workspaces in testing.
  • Security is enforced by a trusted host layer outside the agent sandbox, which manages permissions, blocks general internet access, and routes all service requests through a proxy for credential injection.
  • Changes are traceable to specific conversations, enabling users to undo agent actions via Dropbox Rewind without affecting unrelated edits made by other people or agents.
  • The interface uses the Codex App Server API to display agent activity as structured cards, allowing users to review plans, approve folder access, and monitor progress in real-time.
  • An adapter layer allows the system to support different agent harnesses, such as Pi, by translating their protocols into the standard format understood by the Dropbox interface.
file-managementdropboxai-agentvirtual-machinescodex-app-serverrclone
Read originalShare on Threads

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.

Related

TLDR AIFigma's unreleased editing model, post-trained on Qwen 3.8 27B, beats top third-party model on internal evalsKakaoKakao Releases Open-Source 3B Multimodal Decision-Making Model 'Kanana-2-v-decide'KarrotCarrot Launches Korea's First Public Beta for Advertising MCP, Sharing a Case Study on Expanding Developer RolesMusinsaMUSINSA develops DeciML and NeXA to move beyond prediction-centric personalization toward decision-making and discoveryAWS Tech Blog (Korea)Analysis of Amazon Bedrock AgentCore Runtime's microVM Isolation Principles and Selection CriteriaKakao PayKakaoPay FE Harness: Model Improvements Reduce Agents from 21 to 11