Arqo
co-founder · sole engineer · 2025–present · soft-launched Sep 2026
An AI screenwriting app whose AI analyzes your script and never writes it. Native iOS and macOS, Android, desktop and web.
The Arqo ad, 30 seconds. Sound starts off; turn it on with the control on the slide.
1 / 4Impact
problem
Screenwriting tools are desktop-bound, closed-format and weak on mobile. Writers' rooms need real collaboration, and craft writers don't want a machine writing their scenes.
approach
A cross-platform native suite (iOS, macOS, Android, desktop and web) with the script as the single source of truth and AI as staff, not author. I build it solo, directing Claude Code agent fleets.
outcome
Soft-launched with no marketing: and in the first month, , and who moved an existing script in, with on the waitlist.
How it ships
One engineer, agent fleets. Every agent change clears the same gate.
- 01own worktreeeach agent works in an isolated git worktree, so parallel agents never collide
- 02pull requestevery change arrives as a PR, never a direct push
- 03verify renderuser-facing changes are screenshotted on a preview before anyone calls them done
- 04CItypecheck · row-level security · FDX conformance · golden PDFs · iOS build
- 05I mergenothing merges on an agent's word
| Week of | Commits |
|---|---|
| Apr 13 | 37 |
| Apr 20 | 501 |
| Apr 27 | 183 |
| May 4 | 307 |
| May 11 | 238 |
| May 18 | 177 |
| May 25 | 307 |
| Jun 1 | 674 |
| Jun 8 | 209 |
| Jun 15 | 171 |
| Jun 22 | 145 |
| Jun 29 | 135 |
| Jul 6 | 92 |
| Jul 13 | 131 |
| Jul 20 | 40 |
| Jul 27 | 101 |
| Aug 3 | 6 |
| Aug 10 | 71 |
| Aug 17 | 255 |
| Aug 24 | 59 |
| Aug 31 | 206 |
| Sep 7 | 96 |
| Sep 14 | 112 |
| Sep 21 | 183 |
| Sep 28 | 147 |
| Oct 5 | 117 |
How it's built
6 components · ← → to step through- U13 native editors, one speceditor
- U2Yjs CRDT via Liveblockscollab
- U3deterministic, shared paginatorpaginator
- U4analysis, never ghostwritingAI staff
- U5verified before mergeship gate
- PWRthe stackstack
3 native editors, one spec
The script editor exists three times: ProseMirror on the web, TextKit on macOS, native on iOS. Instead of hoping they agree, one behavior spec (every keystroke, Tab/Enter cycle and page break) runs as a test suite against all three. The web editor started hand-rolled; after regressions I moved it to ProseMirror and wrote down why in an ADR.