AI for Content Workflows and Video Editing Webinar July 9th
A live walkthrough of an AI content workflow: find trending ideas with Claude Code sub-agents and Apify, break long content into reusable pieces, then edit video either with code (cheap) or with generative video (creative but expensive).
Who it's for: Business owners and creators who publish content and want AI to handle research, repurposing and video editing without hiring an editor.
Key takeaways
Research trends with sub-agents and real data
Ask Claude Code to run a dynamic workflow: it splits the job across sub-agents (one per platform or topic), each running an Apify scraper on real social media data instead of a vague web search. Give each agent a budget, such as fifty cents, and ask for content ideas per category. Runs can last 15 minutes to an hour and use a lot of tokens, so I only recommend it on the $100 to $200 Claude plans.
Atomize long content into reusable pieces
A content atomizer takes a long piece of content (a sales call transcript, a meeting, someone else's video plus its comments) and breaks it into small atoms, each tagged as hook, retain or reward and saved to Notion. One video by Nick produced over 20 ideas for blog posts, Shorts, reels and TikToks. Add your own experience before reusing an atom, because that is what adds value.
Teach the AI a framework from a book
I built my atomizer by giving Claude Code the framework from Hormozi's 100M Leads and telling it to atomize content according to that framework and push the results to Notion. I also recommend Influence by Cialdini. Pair this with the yt-dlp Python library, which lets Claude Code pull YouTube transcripts, comments and video stats so you can atomize anything.
Transcripts for timing, Gemini for seeing
Most models cannot watch video, so you give them a timestamped transcript. ElevenLabs gives timestamps precise to the millisecond, which is what editing needs. Gemini is the only model that watches picture and sound together, so use it when what is on screen matters, for example finding when someone wore something or reacted in a sales call. Its transcripts are less precise.
Have Gemini check videos for leaked secrets
After leaking a few API keys early on, I have Gemini watch every video before publishing. It flags API keys, IDs or emails on screen and suggests which time range to blur. I use Gemini 3.1 Flash Lite for this because it is the smallest model and the task does not need much intelligence.
Two ways to edit video with AI
Code-based editing (ElevenLabs transcript read by Claude Code, then HyperFrames) finds keywords and silences and adds cards, icons and captions on top of your footage for about a cent. Generative editing (Gemini, then Gemini Omni) regenerates the video, so you can change your background or look, but costs around seven cents per second and needs several runs. Use code for talking heads, generation for creative effects.
Buy a microphone before a camera
I film with a Sony ZV-E10 and a Sigma lens (about 1.2K for the whole setup) and a Shure SM7B at about 300 dollars. If you have to choose, buy the microphone first. With a budget of about 100 dollars, I suggest the Samson Q2U.
Tools shown
- Claude CodeRunning dynamic workflows with sub-agents for trend research, and editing videos from transcripts with HyperFrames
- CodexRunning the content atomizer skill, to show the workflow works with GPT models too
- ApifyScrapers that pull real trending data from X, YouTube, TikTok and Facebook for each research sub-agent
- NotionStoring content atoms and mapping them to the videos where they are used
- yt-dlpPython library that gives Claude Code YouTube transcripts, comments and video stats
- ElevenLabsMillisecond-accurate transcripts with timestamps for code-based video editing (about 6 dollars per month)
- Google GeminiWatching videos (3.1 Flash Lite), checking for leaked secrets before publishing, and video-to-video editing with Gemini Omni through a Gemini API key
- HyperFramesCode-based editing: adding icons, highlighted captions and a short intro on top of existing footage
- Fal.aiAlternative route to run Gemini Omni when it was blocked (Omni is not available in Europe or the UK without a VPN)
- HeyGenRecommended for creating a realistic avatar and videos with it (same team as HyperFrames)
Questions from the audience
Are dynamic workflows rebuilt every session, or do they reuse tools from before?
The main Claude Code chat decides how to split your request and builds each sub-agent end to end: it picks the model, the tools and the reasoning level. Any tool your main session is connected to, the sub-agent can use too. People call this harness engineering. You can use it for research, UI, UX, back-end or reviewing copy, but it uses a lot of tokens, so I recommend it mainly on the 200 dollar plan.
Why use Gemini Flash Lite instead of a more powerful model?
I always use Flash Lite because it is the smallest model and watching this kind of video does not need much intelligence. You can use 3.1 Pro, but I do not recommend it. One example use: I have Gemini watch every video before publishing to catch leaked API keys, IDs or emails, and tell me which time range to blur.
How much does it cost to edit a short video with each approach?
The code-based approach costs about one cent, because it is all code and you only pay for the transcript. The generative approach costs at least 3 to 5 dollars: usually about seven cents per second, so 70 cents for 10 seconds, and you often need three or four runs to be happy, so a bit under 10 dollars. That is much cheaper than a video editor, but still adds up.
Do I need Claude models, or does GPT work the same?
It works the same. I edit my videos with both GPT models and Claude models and see no difference. I run eight terminals at once, usually four with Claude Code and four with Codex, gives them goals from the hardest task to the smallest, and watches over them.
Watch next
11:55Master Claude Code Dynamic Workflows (49 Agents)Run your own dynamic workflow where agent teams research a topic in parallel.
11:09Turn Any Expert into Your AI Mentor (2026)Turn an expert's book and videos into reusable skills Claude applies to your work.
6:13Stop Wasting Hours: Analyze Any Video with AISet up a Gemini key so an agent can watch your footage and report what it sees.
