# Training AI by Recording Human Demonstrations

AI systems increasingly learn tasks by capturing and imitating recorded human activity — screen recordings for digital workflows, worn-camera motion capture for robot manipulation — rather than explicit programming, making harvested human demonstration a core and recurring training input across software agents and robots.

- Conviction: 38 / 100 (forming)
- Horizon: Emerging (watchlist)
- Tracking since: 2026-07-23T00:00:00.000Z
- Last updated: 2026-07-23T05:54:22.032Z
- Canonical: https://polylog.news/ai/trends/learning-from-human-demonstration
- Publisher: Polylog
- Affected regions: Global

## Recent evidence

- [confirms] Anthropic Adds Screen-Recording to Teach Claude New Skills (2026-07-23): Anthropic added screen-recording so users demonstrate a workflow and Claude learns the procedure rather than being programmed step by step, mirroring Tesla's plan to record worker motions for Optimus. Two independent actors adopting demonstration-capture on the same day signals a forming training paradigm.
- [confirms] Tesla to Record Gigafactory Workers' Motions to Train the Optimus Robot (2026-07-23): Tesla's plan to capture Gigafactory workers' manipulation motions to train Optimus is a physical-domain instance of the same demonstration-capture approach, extending imitation-from-recorded-humans from digital workflows into robotics.