Production team has editors narrate their cuts to train future AI editing agents
alexgoughcooper · x · 2026-09-16
A content team bets agentic video editing is near, so editors now record themselves explaining the reasoning behind every cut—pacing, b-roll choices, narrative logic. These transcripts become structured context to feed editing agents as they improve.
The author likens it to adding analytics pixels from day one: build proprietary data infrastructure now so your expertise compounds when AI catches up. Core play: have experts think aloud while working, own the transcripts, and pre-build the taxonomy agents will need.
More from coding & agent
- What benchmarks are worth running locally? Reddit thread on DIY inference evals — JustTooKrul · 2026-09-17
- Billion-param classifiers distilled from frontier models are an order of magnitude cheaper — FrankFelixAI · 2026-09-17
- Workers Delos Raises €10M for 'AI Employees' With Their Own Email and Phone — JafarNajafov · 2026-09-17
- Why Can I Easily Find Code Optimizations That Frontier LLMs Overlook? — oyacaro · 2026-09-17
- Agentic search burns 30-50% of agent tokens: RL-trained model is 20x faster and 100x cheaper — AI Engineer · 2026-09-17
- Unity Ships Official First-Party Plugin for OpenAI Codex, Maintained by Unity Engineers — pvncher · 2026-09-17