Ship the Proof: 90-Second Documentaries

Opportunity AI is not "make a video with a prompt." It is a film assembly line for judgment you already have.

You already own the raw materials: interview clips, screen recordings, before/after stills. In 2026 the bottleneck is directing a repeatable pipeline, not hiring an editor. A solo builder can land a credible 60-90s mini-doc in 2-4 hours for under about A$5-40 once a house style exists. Trust comes from real faces + one specific outcome + visible proof. AI cuts, captions, and fills gaps. It does not fake the customer.

Why short docs beat case-study PDFs

Buyers reduce risk by inspecting three things: why it mattered, what changed, and whether the outcome matches the claim. A 90-second film can hold all three. Text testimonials rarely do.

Psychology that matters in practice:

• Inspectability — voice = stakes; screen = invention; metric card = outcome.
• Peer identity — "someone like me" lands harder than polished brand copy.
• Authenticity > cinema — over-produced or AI-presentered ads often underperform when the stakes are real. Phone A-roll is a feature.

Keep it short enough to finish on LinkedIn or a product page: problem, process, measurable outcome.

The pipeline (compose, don't just clip)

Opus-style clippers extract highlights from long videos. Your job is different: compose mixed assets into one arc.

Steady-state cost: CapCut path A$0-15; Descript + light gen about A$20-60/mo tools; agent factory (Whisper, FFmpeg, Remotion) cheaper per episode after a half-day setup.

Tool ladder

1. Cheap: Phone + CapCut (plus local Whisper/FFmpeg)
2. Prosumer: Descript (dialogue) then CapCut (vertical/captions) then short Runway clips only if you lack B-roll
3. Agent factory: Claude Code or Codex orchestrating Whisper + FFmpeg + Remotion/Hyperframes
4. Enterprise finish: Resolve/Premiere grade, usually overkill for LinkedIn v1

Rule: real subject on camera for trust-critical claims. No AI avatar spokesperson for proof stories.

90-second shot list

Interview prompts (don't script answers): What were you doing the week before this existed? What almost made you quit? Show the moment it clicked. Who should not use this?

Five episodes I could shoot this month

1. RivaFlow, notebook to logged training — app recordings + mat clips
2. Digital team charter, a $50/mo AI company that ships — role cards + weekly review (no confidential spend beyond the public story)
3. Personal CoS, morning brief on film — anonymised UI + calendar blocks without work titles
4. The Room, archive to model to new work — Lightroom grid + clearly labelled AI experiments + human direction
5. BJJ practice loop, closed-loop notes — session logs + consented mat footage

All builder/personal framing. No employer systems, no client PII.

Failure modes

• Vague praise with no number or artefact
• Mid-clause jump cuts from over-aggressive filler removal
• Confidential bleed in screen recordings
• Publishing auto-clips without a human pick
• Believing unit tests mean the cut engine works on real media

Seven-day experiment

1. Pick episode 1 or 3; beat sheet + asset dump
2. Shoot 8-12 min A-roll + 3-5 min screen proof
3. Picture-lock 75-90s (no graphics)
4. Captions, metric card, music; export 9:16 + 16:9
5. Optional: ask Cursor/Codex to brand-kit the pipeline once
6. Publish; ask three peers what the invention and outcome were
7. Retro: time, cash, clarity score; write brand.md for episode 2

Week-one success: strangers name problem + outcome; you'd send it to a prospect; edit under four hours; nothing confidential.

Bottom line

Opportunity AI here is productising judgment as a film assembly line, the same way coding agents productised software assembly, using assets you already create while building.

Efficiency AI drafts the caption. Opportunity AI ships the proof.