Case Study: 90/10 Content Pipeline
Your agent does the dishes, you bring the personality
Creating content is a full-time job. Research trends, write posts, reply to comments, track analytics. Or... you let your agent do 90% of it and you handle the 10% that actually requires a human brain. Here's the complete system, broken down by platform.
The 90/10 Split
- • Research what's trending in your niche
- • Draft 3-5 post options per content slot
- • Repurpose one piece across all platforms
- • Handle simple replies (thanks, acknowledgments)
- • Monitor what's performing well
- • Maintain the content calendar
- • Schedule posts at optimal times per platform
- • Pick which draft to post (30 seconds)
- • Add your personality — hot takes, humor
- • Strategic decisions on topics/direction
- • Handle sensitive or controversial replies
- • Record quick voice/video when needed
- • Occasionally say "more of this, less of that"
- ✗3 hours researching trending topics manually
- ✗1 hour writing a single blog post draft
- ✗30 minutes reformatting for each social platform
- ✗Total: 5+ hours for one piece of content
- ✗Posting frequency: 2-3x per week at best
- ✓Agent researches and drafts overnight while you sleep
- ✓15 minutes reviewing and adding your personality
- ✓Agent auto-reformats for Twitter, LinkedIn, newsletter
- ✓Total: 15 minutes of your time per piece
- ✓Posting frequency: daily across all platforms
The Core Workflow
Agent scans Hacker News, Twitter trending topics, specific subreddits, and competitor accounts. Extracts 5 potential topics with engagement data. Saves to content/research.md.
Agent writes 3 draft variations for each topic using your voice (pulled from tacit.md). Adapts format per platform — threads for Twitter, carousels for Instagram, long-form for LinkedIn. Saves to content/drafts.md.
You read the drafts. Delete the weak ones. Tweak the good ones. Reply "post 2" or react with ✅. That's your entire content creation for the day.
Agent posts at optimal times per platform, replies to comments, tracks performance, and reports weekly analytics.
The Research Cron
openclaw cron add \ --name "Content Research" \ --cron "0 7 * * *" \ --tz "America/Chicago" \ --session isolated \ --message "Research today's content opportunities. 1. Search Twitter for trending topics in [AI/SaaS/your niche]. 2. Check Reddit r/[your-subreddit] for top posts (24h). 3. Scan Hacker News front page for relevant discussions. 4. Check what competitors posted in the last 24h. Output 5 topic ideas ranked by: - 🔥 Trending score (is this hot right now?) - 💬 Discussion potential (will people reply?) - 🎯 Audience fit (does our audience care?) For each topic, include: - One-line hook - Key angle/take - Supporting data point or quote Save to content/research.md" \ --model "sonnet"
🔌 Platform-Specific Playbooks
Each platform has its own culture, format, and algorithm. Your agent needs to speak each platform's language fluently.
🐦 Twitter/X — The Engagement Engine
Twitter rewards hot takes, threads, and conversations. Your agent should draft differently here than anywhere else.
- • Format: Single tweets (hot takes), threads (tutorials/stories), quote tweets (commentary)
- • Optimal times: 8-10 AM, 12-1 PM, 5-7 PM (your timezone)
- • What works: Contrarian takes, "Here's what nobody tells you about X", numbered lists, before/after
- • Thread structure: Hook → Problem → Solution → Proof → CTA
- • Reply strategy: Agent replies to comments within 1 hour (algorithm boost). Genuine responses only — no "great point!" spam
openclaw cron add \ --name "Twitter Drafts" \ --cron "30 7 * * *" \ --session isolated \ --message "Read content/research.md. Draft 3 tweets for today: 1. A hot take or observation (single tweet, <280 chars) 2. A mini-thread (3-5 tweets) teaching something 3. A question or poll to drive engagement Use my voice from tacit.md. Be punchy, not corporate. Save to content/drafts/twitter.md" \ --model "sonnet" --announce --channel discord
💼 LinkedIn — The Professional Play
LinkedIn rewards storytelling, lessons learned, and "I was wrong about X" posts. Totally different vibe from Twitter.
- • Format: Long-form posts (1300+ chars perform best), carousel documents, polls
- • Optimal times: Tue-Thu 8-10 AM (business hours)
- • What works: Personal stories, failure lessons, "5 things I learned", industry insights with data
- • Hook formula: Start with a bold statement. Line break. Then the story. LinkedIn shows only first 2 lines before "see more"
- • Engagement hack: End with a question. Comments boost reach 5x vs likes
📸 Instagram/TikTok — The Visual Pipeline
Your agent can't create videos (yet), but it can do everything around them:
- • Agent does: Research trending audio/formats, write scripts, draft captions, generate carousel text, schedule posts
- • You do: Record the 30-second video, take the photo, or approve the carousel
- • Caption formula: Hook line → Value → CTA → Hashtags (agent researches optimal hashtags weekly)
- • Carousel scripts: Agent writes slide-by-slide text. You drop it into Canva. 5 minutes.
- • Reels/TikTok scripts: Agent writes the script + suggests trending audio. You just read it on camera.
📧 Email Newsletter — The Revenue Driver
Email is where the money is. Your agent writes it, you review it, subscribers pay for it.
openclaw cron add \ --name "Newsletter Draft" \ --cron "0 8 * * 1" \ --tz "America/Chicago" \ --session isolated \ --message "Draft this week's newsletter. 1. Review the past 7 days of content/research.md 2. Pick the 3 most interesting topics 3. Write a 500-word newsletter with: - Subject line (A/B test: write 2 options) - Opening hook (personal anecdote or observation) - 3 key insights with commentary - One actionable tip readers can use today - CTA to our product/community Voice: conversational, like emailing a smart friend. Save to content/drafts/newsletter.md" \ --model "opus" --announce --channel discord
💬 Discord/Slack Communities — The Engagement Loop
Your agent can be an active, helpful member of communities where your audience hangs out. Not spamming — genuinely participating.
- • Answer questions in relevant channels (with your knowledge base as context)
- • Share insights from your research (not links to your stuff — actual value)
- • Host weekly threads — "What are you building?" or "Share your wins"
- • Monitor for opportunities — someone asking "is there a tool that does X?" when you built X
- • Rule: 80% value, 20% promotion. Break this and you get banned.
The Content Repurposing Machine
One piece of content should become 5. Your agent handles the transformation:
Take this blog post and create: 1. A Twitter thread (5-7 tweets, punchy, end with link to full post) 2. A LinkedIn post (storytelling format, 1500 chars, end with question) 3. A newsletter snippet (200 words, with one actionable takeaway) 4. An Instagram carousel script (8 slides, one key point per slide) 5. A 30-second video script (hook → problem → solution → CTA) Adapt the tone for each platform. Twitter = punchy. LinkedIn = professional. Newsletter = friendly. Instagram = visual.
Measuring What Works
openclaw cron add \ --name "Content Analytics" \ --cron "0 9 * * 1" \ --session isolated \ --message "Weekly content performance review. Check which posts got the most engagement this week. Compare against last week's performance. Output: - 🏆 Top performer (and WHY it worked) - 📉 Worst performer (and WHY it flopped) - 📊 Engagement trend (up/down/flat) - 🎯 Recommendation for next week's content focus - 💡 One experiment to try this week Save insights to content/analytics.md" \ --model "sonnet" --announce --channel discord
Common Mistakes
Cross-posting the exact same text to Twitter, LinkedIn, and Instagram. Each platform has different culture, format, and audience expectations. Your agent should adapt, not copy-paste.
If every post is "buy my thing," people unfollow. Follow the 80/20 rule: 80% genuinely useful content, 20% promotion. Your agent should know this ratio from tacit.md.
Posting and ghosting kills algorithms. Your agent should reply to comments within 1-2 hours. Genuine replies, not "thanks for sharing!" auto-responses.
Without analytics, you're guessing. Set up the weekly analytics cron. Let data guide your content strategy, not vibes.
How the Agent Thinks Through Content Creation
Here's what's actually happening inside your agent's "brain" when it gets the morning content task:
Common Content Agent Mistakes
If your agent's posts sound like they could be from anyone, your tacit knowledge file needs work. Add more examples of YOUR voice, your hot takes, your specific experiences.
If every post is a "5 tips for X" thread, your audience gets bored. Mix formats: stories, hot takes, tutorials, behind-the-scenes, questions, polls.
Set up a weekly cron that reviews what performed well. Feed those insights back into the content strategy. Your agent should be getting BETTER at content over time, not just producing more.
Share this chapter
Chapter navigation
12 of 36