Proof Point Studio
Real software in motion, not slides in a video.
- Agentic orchestration
- Architecture & developer experience
- Data science & evaluation
Internal tooling, not a product. An agent reads a repository’s own documentation, proposes value-ranked demo capabilities, builds each into a real interactive page, records it, and assembles the footage into a short reel.
“The most persuasive content for a proof-first brand is not a typeset claim — it’s real software doing the real thing: an eval scorecard resolving, an agent run executing, a before/after on a live dashboard. That footage is unfakeable, which is exactly why most vendors can’t post it.”
The problem
Marketing a proof-first practice with slideware undercuts the claim. But recording real software is laborious, and the moment it becomes laborious it stops happening.
The approach
Four typed, one-directional stages, so each can be inspected on its own. A scout proposes capabilities with a claim tied to its supporting evidence; a builder renders the chosen one into a real branded page; Playwright records it headful at portrait resolution; an assembler cuts cover, demo, outro and voiceover together.
How it works
- Scout: an agent reads the target repository’s agentic docs and proposes value-ranked capabilities, each with a narrative arc and a claim bound to its evidence
- Builder: renders a scene spec into a real interactive web page, not a mockup
- Recorder: Playwright drives Chromium headful at 1080×1920
- Assembler: cover, demo, outro and voiceover into one file, reusing Content Engine for assembly, captions and the claim-provenance gate
Stack
ESM TypeScript, strict, with pnpm, tsx, vitest and Playwright. Claude for the scout stage; ffmpeg via Content Engine for assembly. Ships its own agent instructions and skills, so the pipeline is agent-driven by design.
Where it stands
- One real reel end to end: a live refine run, a real interactive scorecard demo, recorded at 1080×1920 and assembled