How two engineer side projects became the agent taking action inside Dropbox
Daniel Reiter Horn and Rich Chan · 5 November 2026
Read the originalTwo Dropbox engineers turned side projects into the AI agent that changes customers' files, with plan approval, folder permission cards and an undo tied to each conversation.
Daniel Reiter Horn and Rich Chan, engineers at Dropbox, the file storage company, explained how they built the AI agent that organises and renames files in the new Dropbox experience. Both had used coding agents on personal tasks and wanted an agent to change files while they kept control and a way to recover from mistakes. Their two prototypes became one product: an isolated workspace for the agent, with permissions, approval and undo handled outside it.
- For a broad request, the agent must propose a plan, and a person must accept it before the agent may ask to change files.
- When the agent needs to change a folder, the app shows an Allow or Deny card. Permission checks run outside the agent's workspace, so the agent cannot grant itself access.
- Dropbox records which conversation produced each file version, so undoing one conversation leaves unrelated files alone.
- If a person has edited an affected file since, the restored version is saved as a separate renamed copy so that the person's work is kept.
The authors chose to ask for approval of the plan and the folder rather than of each change, because reorganising a photo collection can involve hundreds of changes. A product team letting an AI agent change customer data can weigh the same choice. This is Dropbox's account of its own product.
Store with each file version the AI agent conversation that produced it, and let a person undo one conversation by restoring only the files it changed, saving a renamed copy where someone has edited a file since.
Ask a person to accept the AI agent's plan for a broad request before the agent may request permission to change files.
Derived by Working Surface from the article; more in the Playbook. Source line: Keeping access checks outside the VM means the agent can’t grant itself broader access to files or services.
6 November 2026: teams scaling AI agents are replacing case-by-case checking with one repeatable path that every piece of agent work passes through.