Skin deep. You will clone the interface and then meet the wall. Realistically a long weekend.

What stops a copy of Dropbox working?

Sync is a distributed systems problem wearing a folder icon. Two machines edit the same file offline; there is no correct answer, only a policy, and every policy loses somebody's work eventually. Filesystem watching is a swamp of platform-specific APIs, and each one lies to you differently about renames, which are indistinguishable from a delete plus a create unless you track inodes. Case sensitivity, path length limits, reserved filenames and unicode normalisation differ per platform, so the same folder is not representable on all three. Delta sync: uploading a 2GB file because someone changed one byte is the difference between a product and a toy. Doing all of the above without pinning a CPU core or eating a laptop battery.

Can AI copy

Dropbox

A folder that is somehow also on your other computer.

Skin deep

Verdict: Skin deep
Time to copy: a long weekend
Category: ▤ Files & storage
Source: Proprietary

The prompt

Written to be pasted straight into an agent with no editing. It describes the whole job, including the parts that will go wrong.

Build a file sync service. Server stores content-addressed chunks in object storage with a per-file manifest, so uploads and downloads move only changed chunks — use a rolling hash for content-defined chunking rather than fixed offsets. Desktop client watches a directory with the native filesystem API, debounces events, and tracks inodes so renames are detected as renames. Maintain a local database of last-known server state and resolve divergence with a documented policy: last-writer-wins with the loser preserved as a conflicted copy, never silently discarded. Normalise unicode filenames and detect case collisions before writing. Test by editing the same file offline on two machines and reconnecting; that is the only test that matters.

Open inClaude ↗Terminal agents take it from the clipboard.

What copies cleanly

Upload, download, a web file browser, sharing links and version history. All ordinary work against object storage.
A desktop client that watches a directory and pushes changes.
The famous early demo, which was a video, and which is the correct lesson about validating an idea.

Where it stops

Sync is a distributed systems problem wearing a folder icon. Two machines edit the same file offline; there is no correct answer, only a policy, and every policy loses somebody's work eventually.
Filesystem watching is a swamp of platform-specific APIs, and each one lies to you differently about renames, which are indistinguishable from a delete plus a create unless you track inodes.
Case sensitivity, path length limits, reserved filenames and unicode normalisation differ per platform, so the same folder is not representable on all three.
Delta sync: uploading a 2GB file because someone changed one byte is the difference between a product and a toy.
Doing all of the above without pinning a CPU core or eating a laptop battery.

What actually protects them

Not much any more, which is why the category collapsed into a feature of the operating system vendors.
A sync engine refined over fifteen years, which is real but no longer sellable on its own.

Editor's note

Every developer estimates this at a weekend, and the estimate is correct for the version that works on one machine with one user and no conflicts. The remaining 95% of the work is the word 'sync'.

Somebody already did it

Go and look at the original

We would rather you formed your own opinion than took ours. They built the thing; we only measured it.

dropbox.com ↗