YouTube Content Pipeline

Daily YouTube creators spend hours each day on inputs: trending topics, X discussions, GitHub releases, news cycles. The actual filming is a fraction of total time. A scouting pipeline makes the input layer agent-driven so the human focus shifts to writing, filming, and editing — the parts only you can do.

What it does

Hourly scout — scans web/X for breaking AI news, picks angles
Topic dedupe — checks your back catalogue (last 90 days) before pitching
Outline drafting — given a topic, drafts a 5-section outline with talking points
Catalogue tracking — when you publish, the agent logs title + view count snapshots so trends are visible
Idea queue — every approved pitch goes to a memory queue you can query

What you'll need

Web fetch + x-tools for sources
Memory for catalogue + queue
Delegate for parallel scouting
Cron for hourly runs
A consistent niche (AI, programming, etc.) — the more specific, the better the scout

Setup

1. Define your niche

Send to Flowly

Remember (tag "channel-niche"): Niche: AI agents and developer tools Audience: technical, intermediate-to-senior devs What I cover: plugin systems, agent frameworks, LLM evals, dev workflows What I skip: general AI hype, model release recaps, beginner tutorials Voice: opinionated, specific examples, contrarian when justified.

2. Hourly scout cron

Send to Flowly

Cron "yt-scout" every hour from 8 AM to 8 PM: Spawn 2 subagents in parallel: Web scout: "Search for AI/dev news from the last hour: - HN front page (filter to AI/dev topics) - arXiv new submissions (CS.AI/CS.LG) - New GitHub repos trending For each item, write: title, source link, why it matters in 1 line. Score 1-10 for video potential (specific, controversial, or novel = high)." X scout: "Search X for hot AI threads from the last hour: - Posts >100 likes in the last 60 min mentioning AI/agents/LLM - Threads from accounts I follow (channel-niche aligned) For each, write: 1-line take + thread link + sentiment (excitement, skepticism, controversy)." After both complete: 1. Combine results, dedupe overlaps (same story across sources) 2. Filter to score >= 7 3. Cross-check against my last 90 days of videos (stored in catalogue): - If we already covered the topic, skip unless angle is genuinely new 4. Take top 3 5. For each, draft a candidate video title + 2-line angle Send to Telegram with reactions: 👍 to queue, 👎 to skip.

3. Outline draft on demand

Send to Flowly

When I say "outline for <topic>": 1. Search my catalogue for related past videos 2. Search current state of the topic via web_fetch 3. Draft a 5-section outline: - Hook (90 seconds — what's the punchline?) - Setup (the standard view) - The twist (your contrarian angle, or what's actually new) - Demo / examples (concrete) - What to do about it (call to action without being cringe) 4. Add: "researched but not in script" notes — interesting tangents I could pull into B-roll 5. Save to memory "outline:<slug>"

4. Catalogue tracking

Send to Flowly

When I publish a video, I'll send: published: <title> — youtube.com/watch?v=<id> You log it in catalogue with date, title, video_id. Nightly cron updates view count snapshots (24h, 7d, 30d). When asked "how did <title> do?", show: views over time, comparison to my channel median for that period, top comments theme (fetch via web).

Tips

Hourly is right for hot niches. AI/dev moves fast. For slower niches (history, finance), every 4 hours.
Dedupe against catalogue is crucial. 2nd-time covering a topic is fine if angle is new; 3rd-time is rarely worth it. The agent should flag this.
Pitch quality > pitch quantity. 3 strong daily pitches beat 15 mediocre ones. Score floor matters.
Don't auto-publish anything. The pipeline gets you to outline. You record and edit. Auto-anything-public ends careers.
Audit weekly. Friday, review the week's scout outputs vs what actually went viral. Adjust the score weights when it consistently misses.