Summary
Brian Casel walks through the complete AI-powered video editing pipeline he built for his YouTube channel — a system he claims has cut his post-production time from over a week down to a single day. Every video on his channel, including this one, is now edited by a Claude Code agent orchestrating a network of custom skills.
The architecture involves several moving parts: a custom “video editing” skill that wraps an FFmpeg engine called “video use,” ElevenLabs for transcription (run multiple times so the agent can self-verify its edits), a custom B-roll indexing app called Tubery, and a motion-graphics skill called Hyperframes. Claude analyzes the script, proposes cuts using an HTML-like tag syntax, selects relevant B-roll from the library, and generates new motion graphics where needed. The process runs in three phases — prep and analysis, first-pass editing, final stitch and export — with Brian reviewing and approving markdown-formatted plans between phases.
What makes the video especially useful is the candid discussion of human-in-the-loop design: Casel describes exactly where he intervenes, how he passes feedback as inline text notes in the transcript, and why certain creative decisions still require human judgment. For developers or creators exploring agentic workflows, this is a practical end-to-end example of what it actually looks like to rebuild a core business process around Claude Code.
📺 Source: Brian Casel · Published July 16, 2026
🏷️ Format: Workflow Case Study
