diff --git a/usecases/ai-video-editing.md b/usecases/ai-video-editing.md
new file mode 100644
index 0000000..0f7d7f6
--- /dev/null
+++ b/usecases/ai-video-editing.md
@@ -0,0 +1,49 @@
+# AI Video Editing via Chat
+
+Editing videos usually means opening a timeline editor, dragging clips around, and clicking through menus. For repetitive edits — trimming intros, adding subtitles to a batch of clips, adjusting color on 10 videos — that manual loop eats hours.
+
+This use case turns video editing into a conversation. Describe what you want changed, drop the file, and get the result back. No timeline, no GUI.
+
+## What You Can Do
+
+• Trim, cut, and merge clips by describing timestamps in plain language
+• Add background music with automatic audio ducking
+• Generate and burn subtitles from speech (50+ languages)
+• Color grade footage ("make it warmer", "match the look of the first clip")
+• Crop to vertical for TikTok/Reels/Shorts
+• Batch process multiple files with the same edit
+
+## Skills You Need
+
+- [video-editor-ai](https://clawhub.ai/skills/video-editor-ai) — chat-based video editing with BGM, subtitles, and export
+- [ai-subtitle-generator](https://clawhub.ai/skills/ai-subtitle-generator) — auto captions, subtitle burning, SRT export
+
+## How to Set It Up
+
+1. Install the skills:
+```bash
+clawhub install video-editor-ai
+clawhub install ai-subtitle-generator
+```
+
+2. Drop a video file into chat and describe your edit:
+```text
+Trim this video from 0:15 to 1:30, add background music (something upbeat),
+and burn subtitles in English.
+```
+
+3. For batch processing, describe the pattern:
+```text
+I have 5 clips in /videos/raw/. For each one:
+- Crop to 9:16 vertical
+- Add auto-generated captions at the bottom
+- Export as mp4
+```
+
+The agent handles the API calls, polls for completion, and delivers the finished files back to chat.
+
+## Tips
+
+- Be specific about timestamps and output format ("export as mp4 at 1080p")
+- For subtitle work, mention the source language if it's not English
+- Color grading works best with reference descriptions ("warm sunset tones") rather than technical values