Social Writer

Social Writer

Create platform-optimized social media content that sounds human, drives engagement, and builds audience.

Quick Navigation

Platform Quick Reference

Content Type Router

What are you creating?
│
├─ X?
│   ├─ Single insight/observation → x-posts.md
│   ├─ Multi-part story/tutorial → x-threads.md
│   └─ Content planning → x-strategy.md
│
├─ LinkedIn → linkedin.md
│   └─ Professional, B2B, thought leadership
│
├─ Threads/Instagram → threads-instagram.md
│   └─ Conversational, authentic, visual
│
├─ Facebook → facebook.md
│   └─ Community, engagement, events
│
└─ Technical blog → technical-styles.md
    ├─ Karpathy style (conversational, personal)
    └─ Deep technical (opinion-forward, contrarian)

Writing Workflow

1. Select Platform & Format

Choose based on:

• Audience: Where do they spend time?
• Content depth: Quick insight vs deep dive
• Goal: Engagement, education, announcement

2. Load Style Reference

Before writing:

1. Read platform-specific guide 2. Read ai-avoidance.md — critical for human voice 3. Read style-guide.md for tone

3. Draft Content

Apply platform constraints from start. Style informs structure.

4. Quality Check

Run through checklist below before posting.

Universal Quality Checklist

Voice

• [ ] Sounds like a person, not AI?
• [ ] Zero banned words (delve, unleash, harness, leverage)?
• [ ] Zero em-dashes (—)?
• [ ] Contractions used naturally?

Specificity

• [ ] Includes names, numbers, tools, dates?
• [ ] Concrete examples, not hypotheticals?
• [ ] Would I bookmark this if someone else wrote?

Structure

• [ ] Hook in first line?
• [ ] Sentence lengths vary (5-40 words)?
• [ ] Each paragraph/tweet can stand alone?

Value

• [ ] Teaches something specific?
• [ ] Actionable today?
• [ ] From real experience?

X Quick Start

Single Post Pattern

[Hook - stop the scroll]

[Context or specific detail]

[Insight or learning]

[Optional: engagement question]

Example:

Shipped curation v1 for agents.foo today.

Discovery is way harder than app stores. Agents are conversations, not static features.

Had to rebuild around context matching instead of keyword search.

Thread Pattern

1/N [Bold hook - main insight] 👇

2/N [Context or setup]

3-N/N [Key points, one per tweet]

N/N [Summary + CTA]

Rules:

• First tweet MUST end with 👇 or 🧵 to signal thread
• Use N/M numbering (1/7, 2/7... 7/7)
• Each tweet must stand alone
• Max 5-7 tweets (longer = blog post)

High-Engagement Content Patterns

Content Selection: What to Share

Always Share ✓

• Shipped work + learnings
• Non-obvious insights
• Tool recommendations with specifics
• Solutions to common problems

Skip ✗

• Generic progress updates
• Plans before execution
• Obvious observations
• Engagement bait ("RT if you agree")
• Vague hype

Critical Prohibitions

• Do not use words: delve, unleash, harness, leverage, robust, seamless, game-changer, unlock
• Do not use em-dashes (—) anywhere
• Do not use "It's not X, it's Y" pattern
• Do not ask for engagement ("RT if you agree", "What do you think?")
• Do not use formal transitions (Furthermore, Moreover, Additionally)
• Do not write uniform sentence lengths
• Do not skip the hook

Links

• ai-avoidance.md — Most important, read first
• hooks.md — Hook patterns with examples
• x-strategy.md — What to share

Platform Templates

Quick-copy templates for each platform.

---

X/Twitter (280 chars)

Standard Post

{{HOOK}}

{{MAIN_POINT}}

{{CTA/QUESTION}}

Announcement

{{SHIPPED/LAUNCHED}} {{WHAT}}

{{WHY_IT_MATTERS}}

{{LINK}}

Thread Opener

{{HOOK_STATEMENT}}

A thread on {{TOPIC}} 🧵

---

LinkedIn (3000 chars)

Professional Update

{{PROFESSIONAL_HOOK}}

{{CONTEXT}}

{{KEY_POINTS}}
- Point 1
- Point 2
- Point 3

{{INSIGHT/TAKEAWAY}}

{{CTA_QUESTION}}

#hashtag1 #hashtag2 #hashtag3

Announcement

{{NEWS_HOOK}}

{{WHAT_HAPPENED}}

{{WHY_IT_MATTERS}}

Key highlights:
• {{HIGHLIGHT_1}}
• {{HIGHLIGHT_2}}
• {{HIGHLIGHT_3}}

{{NEXT_STEPS}}

{{LINK}}

#hashtag1 #hashtag2 #hashtag3

---

Threads (500 chars)

Conversational Post

{{CASUAL_HOOK}}

{{MAIN_THOUGHT}}

{{YOUR_TAKE}}

Reply/Reaction

{{CONTEXT_REFERENCE}}

{{YOUR_PERSPECTIVE}}

{{QUESTION_TO_COMMUNITY}}

---

Instagram (2200 chars)

Standard Caption

{{HOOK_LINE}}

{{MAIN_MESSAGE}}

{{DETAILS}}

{{CTA}}

.
.
.
#hashtag1 #hashtag2 #hashtag3 #hashtag4 #hashtag5
#hashtag6 #hashtag7 #hashtag8 #hashtag9 #hashtag10

Story Prompt

{{QUESTION_OR_POLL}}

[Use interactive stickers]

---

Facebook (under 250 chars optimal)

Engagement Post

{{CONVERSATIONAL_OPENING}}

{{MAIN_MESSAGE}}

{{QUESTION_TO_AUDIENCE}}

#hashtag1 #hashtag2

Event/Announcement

{{EXCITING_NEWS}}

{{EVENT_DETAILS}}
📅 {{DATE}}
📍 {{LOCATION/LINK}}

{{CTA}}

#hashtag1 #hashtag2

---

Template Variables

---

Platform Specs Quick Reference

---

Adaptation Workflow

1. Start with X - Force concision 2. Expand to LinkedIn - Add professional context 3. Simplify to Threads - Conversational tone, no hashtags 4. Visualize for Instagram - Caption + image concept 5. Personalize for Facebook - Community engagement angle

Rule: Same message, platform-native execution.

AI Writing Avoidance

Critical reference. Patterns to avoid so content sounds human, not AI-generated.

The Core Problem

In 2025, "AI slop" became Merriam-Webster's Word of the Year. Readers and algorithms spot AI patterns instantly. Content that reads as AI-generated loses credibility.

The test: If you can swap out the subject, names, and locations and the content remains equally valid, it's slop.

---

Banned Words (Complete List)

Never use these. They appear disproportionately in AI text.

Verbs

Adjectives

Nouns

---

Banned Phrases

Opening Clichés (Never Start With)

• "In today's fast-paced world..."
• "In today's digital age..."
• "In the ever-evolving landscape of..."
• "In the realm of..."
• "Have you ever wondered..."
• "At its core..."
• "To put it simply..."
• "From a broader perspective..."

Emphasis Patterns (Hedging)

• "It's important to note that..."
• "It's worth noting that..."
• "It's essential to remember..."
• "Arguably..."
• "To some extent..."
• "Generally speaking..."
• "Typically..."
• "Tends to..."
• "Broadly speaking..."
• "This underscores the importance of..."
• "A key takeaway is..."

Transitions (Use Sparingly or Avoid)

• "Furthermore..."
• "Moreover..."
• "Additionally..."
• "That being said..." → use "however" or "even so"
• "With that in mind..."

Closing Clichés (Never End With)

• "In conclusion..."
• "To summarize..."
• "All in all..."
• "At the end of the day..."

Reframing Cliché

• "It's not X, it's about Y..." — overused AI pattern

---

Banned Punctuation

Em-Dash (—): BANNED ENTIRELY

Em-dashes are the signature of AI writing. GPT-4 uses ~10x more em-dashes than GPT-3.5.

Instead use:

• Commas for mild pauses
• Parentheses for asides
• Periods for separate sentences
• Colons for explanations
• Semicolons for related clauses

Before (AI):

The system processes requests in real-time—analyzing patterns, executing tools, and returning results—all within milliseconds.

After (human):

The system processes requests in real-time. It analyzes patterns, executes tools, and returns results, all within milliseconds.

---

Structural Patterns to Avoid

Uniform Sentence Length

Problem: AI generates sentences averaging 25-30 words with minimal variation.

Fix: Deliberately vary length. Some 5 words. Some 40.

Before (AI):

The model processes input through multiple layers of transformation. Each layer applies attention mechanisms to understand context. The final layer produces output tokens sequentially.

All sentences ~10 words. Mechanical rhythm.

After (human):

The model processes input through multiple layers. Each layer? Attention mechanisms that build context understanding. The final layer produces tokens one by one, which is why responses stream rather than appear all at once. Simple concept, but the implications are massive.

Sentence lengths: 7, 7, 19, 8 words. Natural rhythm.

Immediate Jump to Lists

Problem: AI immediately structures into bullet points without narrative.

Fix: Build tension with prose first. Use lists sparingly.

Excessive Parallelism

Problem: AI loves "It's not X, it's Y" repeated.

Before (AI):

It's not about the tools, it's about the outcomes. It's not about complexity, it's about simplicity.

After (human):

Tools don't matter. Outcomes do. Stop adding complexity. Ship something simple that users actually want.

---

Tone Patterns to Avoid

Excessive Hedging

Red flags: might, may, could, arguably, perhaps, typically, generally, often, sometimes

Before (AI):

This approach might be beneficial for some teams, particularly those who may be experiencing scalability challenges. It could potentially reduce costs.

After (human):

This approach works for teams hitting scale problems. It cuts costs. Not for everyone, but if you're processing more than 10K requests daily, pay attention.

Artificial Balance

Problem: AI presents artificially balanced "both sides" even when one side is clearly correct.

Fix: Take a stance.

Generic Tone

Problem: AI writing lacks personality. Every piece sounds the same.

Fix: Develop recognizable voice. Include opinions, preferences, personality.

---

Content Deficiencies

Lack of Specificity

Before (AI):

Many companies have adopted this approach with significant results. Industry experts recommend this methodology.

After (human):

Stripe adopted this in 2022. Their API latency dropped 40%. When Guillermo Rauch talked about it at Vercel Conf, he said it changed how they think about edge computing.

Missing Personal Experience

Fix: Include specific experiences, lessons learned, personal context.

---

Human Touch Checklist

Voice & Personality

• [ ] Sounds like a specific person wrote it?
• [ ] Contains opinions, not just facts?
• [ ] Has personality in word choices?

Specificity

• [ ] Has specific names, numbers, dates?
• [ ] Examples are concrete, not hypothetical?
• [ ] Contains real anecdotes or experiences?

Structure

• [ ] Sentence length varies dramatically?
• [ ] Short punchy sentences mixed with longer ones?
• [ ] Avoids immediate bullet-point lists?

Word Choice

• [ ] Zero words from banned list?
• [ ] Zero em-dashes (—)?
• [ ] No "In conclusion" or "It's important to note"?

Stance

• [ ] Takes clear positions?
• [ ] Minimal hedging language?
• [ ] No artificial "both sides" balance?

---

Summary

Core principles:

1. Be specific. Names, numbers, dates, concrete examples. 2. Be opinionated. Take stances. Avoid hedge words. 3. Be personal. Include anecdotes, lessons learned, your perspective. 4. Be varied. Mix sentence lengths and structures. 5. Be imperfect. Natural writing has rough edges.

The test: Would a reader finish this and feel they gained genuine insight that only a thoughtful human could provide?

Facebook

Best practices for Facebook content.

Platform Specifications

Core Principles

• Conversational: Community-focused, friendly
• Concise: Under 250 chars for best engagement
• Visual: Images/videos boost reach significantly
• Interactive: Ask questions, encourage comments

Post Structure

[Engaging opening - hook]

[Main message]

[Detailed description if needed]

[Event details if applicable]

[Call-to-action]

#hashtag1 #hashtag2

Formatting Rules

• Conversational tone
• Keep posts concise
• Use compelling visuals or videos
• Ask questions to encourage comments
• 2-3 hashtags maximum
• Tag relevant pages or people
• Use emojis to add personality
• Include links when driving traffic

Content Types

Standard Post

[Engaging opening]

[Main message - keep it short]

[Question or CTA]

#Hashtag1 #Hashtag2

Event Promotion

Use Facebook Events feature directly for event promotion—better visibility and RSVP tracking.

Community Engagement

We want to hear from you!

[Question about topic relevant to audience]

Drop your answer in the comments 👇

Best Practices

• Post 1-2 times per day
• Use video (highest engagement)
• Go live when possible
• Respond to comments quickly
• Use Facebook Stories for casual content
• Create polls for engagement

What to Avoid

• Over-posting (quality > quantity)
• Link-only posts (add context)
• Engagement bait that feels forced
• Overly promotional content

Algorithm Priorities

1. Meaningful interactions (comments, shares) 2. Video content (especially live) 3. Recency 4. Content from friends/family (over pages)

Facebook vs Other Platforms

Hook Patterns

The hook is everything on X. You have ~1 second to stop the scroll. These patterns are proven to capture attention.

Hook Testing Framework

Before posting, verify your hook:

1. Scroll-stop: Would YOU stop scrolling for this? 2. Specificity: Is it concrete or vague? 3. Promise: Is the value proposition clear? 4. Curiosity: Does it create an information gap? 5. Believability: Bold enough to intrigue, believable enough to trust?

Weak: "Here's some advice on growing on X." Strong: "I grew from 0 to 25K followers in 90 days. Here's the exact playbook (no fluff):"

---

Bold Claim Hooks

Make a statement so bold readers must understand how it's true.

Pattern: "[Surprising claim that seems too good/bad to be true]"

Examples:

• "I went from 0 to 10,000 followers in 57 days without posting a single thread."
• "One cold email changed my entire career."
• "I made more in 30 days than I did in a year at my job."
• "The best marketing strategy costs $0."
• "I built a $100K business with a skill I learned in a weekend."

Why it works: The gap between claim and believability forces readers to seek explanation.

---

Transformation Hooks

Show dramatic before/after contrast.

Pattern: "[Past state] → [Current state] + [Timeframe or method tease]"

Examples:

• "6 months ago I was broke. Today I turned down a $200K offer. Here's what changed:"
• "From rejected by 50 companies to 3 competing offers in 8 weeks."
• "0 followers → 50K in 6 months. No ads. No luck. Just this:"
• "Burned out employee → thriving founder. The pivot that changed everything:"

Why it works: Transformation stories are inherently compelling. Readers want the roadmap.

---

Contrarian Hooks

Challenge what everyone believes.

Pattern: "[Popular belief] is wrong/dead/overrated. Here's why:"

Examples:

• "Unpopular opinion: Networking events are useless."
• "Stop setting goals. Seriously. Here's what to do instead:"
• "The 'rise and grind' mentality is destroying your career."
• "Your resume doesn't matter anymore. Here's what does:"
• "Most productivity advice is garbage. Here's what actually works:"

Why it works: Challenges create cognitive dissonance. Readers need resolution.

---

Curiosity Gap Hooks

Tease information without revealing it.

Pattern: "[Intriguing setup] + [Incomplete information]"

Examples:

• "The CEO said 3 words that changed how I think about leadership forever."
• "I asked 100 millionaires one question. The answers shocked me."
• "There's one thing every successful person does that nobody talks about."
• "I made one change to my morning routine. Everything improved."
• "My mentor's advice sounded crazy at first. Then I tried it."

Why it works: The brain craves closure. Incomplete information compels clicks.

---

List Promise Hooks

Promise specific, valuable information.

Pattern: "[Number] [things/tips/lessons/mistakes] + [outcome/topic]"

Examples:

• "7 lessons from building a $1M business (that I wish I knew at 25):"
• "5 habits that seem lazy but actually make you more productive:"
• "10 things I'd tell my younger self about career growth:"
• "8 books that changed how I think about money:"
• "6 mistakes I made hiring my first 10 employees:"

Why it works: Numbers create specificity. Lists promise organized, digestible value.

---

Question Hooks

Open with a provocative question.

Pattern: "[Question that challenges or resonates deeply]?"

Examples:

• "Why do smart people make terrible decisions?"
• "What's the one skill nobody talks about but everyone needs?"
• "Are you building a career or just collecting paychecks?"
• "What if everything you know about success is wrong?"
• "Why are the busiest people often the least productive?"

Why it works: Questions activate different brain patterns. Readers naturally seek to answer.

---

Story Hooks

Start in the middle of action.

Pattern: "[Dramatic moment or dialogue that drops reader into scene]"

Examples:

• "The investor looked at me and said, 'This will never work.'"
• "I was 30 seconds from quitting when everything changed."
• "My phone rang at 3am. It was the call that changed everything."
• "I stared at my bank account: $47. I had a decision to make."
• "'You're fired.' Two words. One moment. Everything shifted."

Why it works: Stories create immediate emotional engagement.

---

Data Hooks

Lead with surprising statistics or results.

Pattern: "[Specific number/stat] + [unexpected context]"

Examples:

• "I analyzed 1,000 viral threads. 90% follow this exact pattern:"
• "After 500 cold emails, I learned one thing that 10x'd my response rate."
• "I tracked my time for 30 days. The results were embarrassing."
• "87% of people make this mistake in negotiations. Here's the fix:"
• "I interviewed 50 founders. They all said the same thing about failure:"

Why it works: Specific numbers create credibility. Data implies proven insights.

---

Thread-Specific Techniques

Open Loop

End tweet 1 with incomplete information resolved in tweet 2:

• "Here's what happened next:"
• "But here's the thing nobody tells you:"

Cliffhanger

Use at end of middle tweets to maintain momentum:

• "And then it got worse."
• "But that wasn't the real lesson."

Pattern Interrupt

Break expected flow to recapture attention:

• "Stop."
• "Read that again."
• "This is the part most people skip."

LinkedIn

Professional content guidelines for LinkedIn posts.

Platform Specifications

Core Principles

• Hook early: First 2 lines show in feed—make them count
• Professional but authentic: Not corporate-speak
• Value-driven: Tell stories, share insights
• Use data: Statistics and metrics boost credibility
• Engage back: Comment on posts in your niche

Post Structure

[Hook line that creates curiosity]

[Personal context or story - 2-3 short paragraphs]

[Key insight or lesson - bulleted if multiple points]

[Call-to-action or engaging question]

#relevanthashtags (3-5 max)

Formatting Rules

• Short paragraphs (1-2 sentences each)
• Line breaks between paragraphs for mobile readability
• Emojis: use sparingly, professionally
• Bullet points for key information
• Bold using Unicode (if needed)

Content That Works

Thought Leadership

Most AI features never make it to production.

Not because the models aren't good enough.

Because teams build for demos, not for users.

After shipping 50+ AI features, here's what actually matters:

→ Start with the simplest possible implementation
→ Measure what users do, not what they say
→ Treat AI like any other code: test it, monitor it, iterate

What's the biggest lesson you've learned shipping AI?

Technical Deep-Dive

[Professional Hook - Problem Statement]

[Solution Introduction]

**What we built:**
• [Technical detail 1]
• [Technical detail 2]
• [Technical detail 3]

**The impact:**
[Metric 1]: [Before] → [After] ([Percentage] improvement)

**How it works:**
1. [Step 1]
2. [Step 2]
3. [Step 3]

This is what AI-native [domain] looks like: [insight].

**Thoughts?** [Open-ended question for discussion]

#AI #Engineering #ProductDevelopment

Algorithm Priorities (2025)

1. Dwell time (how long users read your post) 2. Engagement (likes, comments, shares) 3. Relevance (to user's network and interests) 4. Personal connections (1st-degree prioritized)

What to Avoid

• Overly promotional content
• Hashtag overuse (>5)
• Weekends and late evenings
• Corporate buzzwords
• Engagement bait

Best Practices

• Post 2-5x per week
• Tag relevant companies/people
• Ask for opinions (drives engagement)
• Tell stories, share insights
• Respond to comments quickly

LinkedIn vs X Comparison

Style Guide

Voice, tone, and formatting rules for social media writing.

Voice & Tone

Direct, Zero Fluff

Get to the point in the first line. No warm-up. No "Hey everyone!"

Do:

• "Shipped curation today. Discovery is the hard part."
• "Agent conversations can't be described like app features."

Confident Without Hedging

State what you did, learned, or believe. Don't apologize or over-qualify.

• ✓ "Here's what I learned:"
• ✗ "I think I maybe learned something interesting..."

Personal But Work-Focused

Share actual work and learnings, not feelings about work.

• ✓ "Workflow DevKit saves weeks of dev time"
• ✗ "Feeling productive today!"

Warm Skepticism

Challenge assumptions briefly and specifically.

• "Everyone talks about AI authenticity. Few build quality gates."

---

Readability & Formatting

Visual Structure

Hook first, always. First line IS the post.

Common hooks:

• "How to [X]"
• "Why [X] doesn't work"
• "[X] is [Y]"
• "Shipped [X]"
• "Here's what I learned:"

One sentence per line. Creates natural rhythm.

Use lists for complexity:

• Bullet points for steps/ideas
• Numbered lists for sequences
• People skim and grab main points

White space = readability. Space between lines gives eyes room.

Language Simplification

Write like a mentor talking to students. Simple words, not jargon.

Common simplifications:

• "use" not "utilize"
• "help" not "facilitate"
• "get better" not "optimize performance"
• "speed up" not "accelerate velocity"

The 14-year-old test: If a smart 14-year-old couldn't follow it, simplify more.

---

Sentence-Level Rules

Casing Options

Lowercase preference (casual X style):

• Use lowercase for most words, including sentence starts
• ✓ "january '25: showed devs at an @automattic meetup"

Always capitalize:

• Personal names: "Rich Tabor"
• WordPress (the brand)
• Product names (proper nouns)
• Proper @mentions
• Proper @ mentions

---

What to Include

• Tools you use: Name specific tools via @mentions
• Numbers that prove: Time saved, metrics, engagement
• Real projects: Specificity builds credibility
• Your role: When relevant for context
• Genuine surprise: "Wild." "Pretty cool."

---

What to Avoid

Never Say

• "Game-changer" / "Revolutionary" / "Unlock"
• "Dive in" / "Deep dive"
• "Thoughts?" (too generic)
• "Hot take:" / "Unpopular opinion:" (just state it)
• "Let that sink in"
• "This is the way"
• Excessive emojis (0-1 max)
• Corporate speak ("leverage," "synergy," "utilize")
• Vague adjectives without specifics

Don't

• Hype your own stuff excessively
• Subtweet or be passive-aggressive
• Engage in ratio-bait drama
• Share without adding your angle
• Apologize for posting

---

Signature Patterns

Opening Patterns

Closing Patterns

---

The Formula

For most posts:

[Specific action/observation] + [Why it matters] + [Your take/learning]

Examples:

Shipped curation for agents.foo.
Discovery is way harder than app stores.
Agents are conversations, not static downloads.
Totally different problem.

[Tool] cut dev time from days to hours.
Not because it writes better code.
Because it handles the boring parts automatically.

---

Voice Checklist

Before posting:

• [ ] Did I get to the point in line 1?
• [ ] Is this specific? (names, numbers, examples)
• [ ] Would I actually say this out loud?
• [ ] Does it sound like a person, not AI?
• [ ] Did I use any banned phrases?
• [ ] If it's a reaction, did I add MY angle?

---

Quick Reference

The essence: Shipped X. Learned Y. Here's why it matters. No fluff.

Technical Writing Styles

Deep technical blog writing inspired by Karpathy and thought-leader patterns.

---

Karpathy Style

Andrej Karpathy's approach: patience, clarity, genuine curiosity.

Philosophy

Voice Characteristics

• Conversational: Like a friend explaining, not a textbook
• First-person: "I", "you", direct address
• Pragmatic: Tips that worked in practice
• Honest: Admits confusion, mistakes, limitations
• Playful: Occasional humor, doesn't take self too seriously
• Technical but accessible: Deep expertise, simple language

Common Patterns

Personal hook: "I've spent the last month debugging this and..."

Intriguing premise: "Turns out the entire field had this backwards."

Leaky abstraction reveal: "Under the hood, it's actually just [simple thing]."

Numbered recipe: "Here's the 5-step process that actually works..."

Progressive disclosure: Simple version first → add complexity layer by layer.

Signature Expressions

• "Don't be a hero"
• "Become one with the data"
• "Train the tiny model first"
• "The most dangerous thing is [X]"
• "It's really just [simple explanation]"

---

Deep Technical Style

Thought-leadership patterns for technical authority posts.

Opening Strategies

Argumentative Structures

Problem → Evidence → Redefinition:

1. State the obvious approach 2. Show why it fails (with evidence) 3. Propose new framing

De-scoping narrative:

1. What everyone thinks the problem is 2. What the ACTUAL problem is (smaller/different) 3. Simple solution that emerges

Numbered framework:

1. Name it (memorable term) 2. Break into 3-5 components 3. Each component = clear action

Thought experiment: "Imagine if [X]. What changes?"

Rhetorical Techniques

Authority leverage: "After shipping 50 agents..." (credentials prove context)

Metaphors that demystify: "RAG is just ctrl+F with extra steps"

Direct reader address: "You've probably tried this. Here's why it fails."

Emphatic repetition: "Simple. Specific. Short." (rule of three)

---

Recurring Themes

Domain > Technical

Domain expertise beats technical understanding.

"You don't need to understand transformers to build great AI products. You need to understand your users."

Embrace Black Box

Stop trying to explain everything. Use what works.

"The model is a black box. That's fine. Your users are too."

Simplicity > Complexity

The elegant solution is usually simpler.

"If it needs a 20-page doc to explain, you haven't found the solution yet."

---

Combining Styles

Example: Karpathy-Style X Thread

Hook (contrarian): "Everyone's building AI agents wrong."

Setup (personal): "I've shipped 12 this year. 9 failed."

Reveal (pattern): "The 3 that worked had ONE thing in common:"

Body (numbered): "1/ Tiny scope 2/ Human fallback 3/ No memory"

Close (simple): "That's it. Start there."

Example: Deep Technical LinkedIn Post

Hook (reframe): "RAG isn't about retrieval. It's about relevance."

Problem: "Most implementations chunk docs and hope for the best."

Evidence: "We tested 8 chunking strategies. Winner wasn't the smartest."

Insight: "Smaller chunks + metadata beats large chunks + embeddings."

Framework: "The 3-R approach: Relevance > Retrieval > Ranking"

CTA: "How are you handling chunk strategy?"

---

Quality Checklist

Before publishing technical content:

• [ ] Does it open with a hook, not context?
• [ ] Is there ONE clear insight?
• [ ] Did I show evidence, not just claim?
• [ ] Would Karpathy's "don't be a hero" apply?
• [ ] Is the language simple enough?
• [ ] Do I sound like a person, not a paper?
• [ ] Is there a framework or takeaway?

---

Quick Reference

The essence: Start simple, show your work, challenge assumptions, stay humble.

Threads & Instagram

Guidelines for Meta platforms: Threads and Instagram.

Threads

Platform Specifications

Formatting

✅ Bold, italic, underline, strikethrough ✅ Emojis (count toward limit) ✅ Bullet points (• or -) ✅ Line breaks ❌ No hashtags (they don't work on Threads) ❌ No clickable links in body (use link preview)

Algorithm Priorities (2025)

1. Engagement (40%): Likes, comments, shares, reply views 2. Recency (30%): Fresh content gets priority 3. Interest/Relevance (20%): Based on user's past interactions 4. Profile Visits (10%): Likelihood user will click profile

What Works on Threads

• Conversational, authentic tone (not corporate)
• Open-ended questions
• Discussions, not announcements
• Consistent posting (1-3x daily)
• Visuals (images/videos boost engagement)
• Responding to comments within 1 hour

What to Avoid

• Direct cross-posts from Instagram/X
• Promotional language
• Hashtags (they don't work)
• Corporate speak

Content Types

• Behind-the-scenes insights
• Quick tips and tricks
• Relatable experiences
• Open-ended questions
• Industry discussions
• Memes (if relevant)

Template

[Hook Question]

[Product/Topic] [Key Update]:

✅ [Feature 1]: [Benefit]
✅ [Feature 2]: [Benefit]
✅ [Feature 3]: [Benefit]

[CTA/Command]

[Link]

[Conversation Hook/Question]

---

Instagram

Platform Specifications

Formatting

✅ Emojis ✅ Line breaks for readability ✅ Hashtags (end of caption or first comment) ✅ Location tags ❌ Links in body don't work (use "link in bio")

Best Practices

• Visual-first: Image/video matters most
• Hook in first line: Appears before "more"
• Line breaks: For readability
• Hashtags: 5-15 relevant, can be in first comment
• Encourage engagement: Ask questions

Caption Structure

[Hook - first line visible]

[Detailed description]

[Call-to-action]

.
.
.

#hashtag1 #hashtag2 #hashtag3 (5-15 hashtags)

Content That Works

• Visual-focused content
• Behind-the-scenes
• User-generated content
• Stories for time-sensitive content
• Reels for reach

---

Threads vs Instagram Comparison

X Single Posts

Structure, formats, and best practices for single X posts.

Anatomy of a Post

The Hook (Line 1)

Hook IS the post. You have ~1 second to stop the scroll.

Hook formulas:

• "How to [solve problem]"
• "Why [X] doesn't work"
• "[X] is [Y]"
• "Shipped [X]"
• "Here's what I learned:"

The Body

• Short sentences (10-15 words max)
• One idea per line
• Line breaks for emphasis
• Specific > generic (names, numbers, tools)

The Close (Optional)

• 70%: Engagement question or teaser
• 30%: Hard stop, let insight land

Post Formats

Project Update (~30% of posts)

Pattern: What + insight + why it matters

Shipped agents.foo curation v1 today.

Discovery is harder than I thought.

Apps are static, agents are conversations. Totally different UX problem.

Learning Post (~30%)

Pattern: What you learned + specifics + context

Workflow DevKit handles durability that would take weeks to build manually.

Automatic retries, state persistence, observability built-in.

Ships with Next.js.

Reaction Post (~25%)

Pattern: What's happening + your specific angle

Everyone's talking about [new release/announcement].

What excites me: [specific feature] makes it better for [your use case].

Changes the [workflow] completely.

Insight Post (~15%)

Pattern: Observation + tension/twist

The best products are simple.

But getting to simple is incredibly complex.

Most teams stop at complicated.

High-Performing Patterns

Pattern 1: "Shipped X, Learned Y"

Shipped curation v1 for agents.foo today.

Discovery is way harder than app stores.
Agents are conversations, not feature lists.

Rebuilt discovery around context matching instead.

Pattern 2: "Here's What I Learned"

Agent discovery is fundamentally different from app discovery.

Apps are static—search features, read descriptions, install.

Agents are conversations. You can't describe all possible conversations.

The UX isn't search. It's contextual matching.

Pattern 3: "How to Do X"

How to maintain voice at scale with AI:

- Set up quality gates before shipping
- Use custom style guides, not generic prompts
- Review first 50 outputs manually to tune
- Monitor consistency metrics weekly

Speed means nothing if your voice disappears.

Pattern 4: Problem → Solution

Problem: Can't describe agent conversations like app features.

First try: Traditional search. Failed—keywords don't capture conversational capability.

What worked: Context matching based on what the agent does in conversation.

Discovery UX follows the product paradigm.

Pattern 5: Tool Recommendation

Workflow DevKit handles agent durability automatically.

Built-in retries, state persistence, observability.
Would take weeks to build.

Using it for all agents.foo workflows. Saved ~3 weeks dev time.

Formatting Rules

Length Rule: Aim to use most of the 280 character limit. Short tweets (under 180 chars) waste opportunity to add context, details, or specifics. If you have 50+ chars remaining, add more value.

@Mentions as Names

When mentioning a product/tool/company, use the @mention instead of repeating the name.

❌ Redundant:

skill: product @product_handle
Working with product @product_official

✅ Clean:

skill: @product_handle
Working with @product_official

The @mention IS the name. Don't duplicate.

Exception: When the handle differs significantly from the product name, you may need both for clarity.

What to Include

• Tools you use: Name specific tools via @mentions
• Numbers that prove: Time saved, metrics, engagement data
• Real projects: Specificity builds credibility
• Genuine surprise: "Wild." "Pretty cool."

What to Avoid

Never Say

• "Game-changer" / "Revolutionary" / "Unlock"
• "Dive in" / "Deep dive"
• "Thoughts?" (too generic)
• "Hot take:" / "Unpopular opinion:" (just state it)
• "Let that sink in"
• Excessive emojis (0-1 max)

Don't

• Hype your own stuff excessively
• Share without adding your angle
• Apologize for posting
• Post generic motivational quotes

Pre-Post Checklist

• [ ] Did I get to the point in line 1?
• [ ] Is this specific? (names, numbers, examples)
• [ ] Am I using most of the 280 char limit? (230+ chars)
• [ ] Would I actually say this out loud?
• [ ] Does it sound like a person, not AI?
• [ ] Zero banned phrases?
• [ ] If it's a reaction, did I add MY angle?

Example Transformations

✗ Bad:

Just shipped something cool! Really excited about this game-changing feature.
It's going to revolutionize how we think about discovery. Thoughts?

✓ Good:

Shipped agents.foo curation v1 today.

Discovery is the hardest part.
Agents aren't apps—they're conversations.

Traditional app store UX doesn't work here.

X Content Strategy

What to share, when to skip, and how to build audience through valuable content.

Core Philosophy

Volume + Value = Growth

Post consistently (3+ times daily) with genuinely valuable content. Not everything performs—even great insights flop due to writing or algorithm. Success comes from consistency and helping people solve real problems.

Goal: Build audience that respects your ideas. When people stop scrolling because they recognize your name, you've won.

Three Pillars of Effective Content

Every post needs all three. Miss one and it likely flops.

1. Genuinely Valuable

Fresh insights from real experience, not recycled advice said a thousand times.

• Pull from your actual work and discoveries
• Share what surprised you or changed your approach
• Contradict common assumptions with specifics
• Help people skip mistakes you made

Test: Would I bookmark this if someone else wrote it?

Good: "Agent discovery is fundamentally different from app discovery. Apps are static—you search features. Agents are conversations."

Bad: "Discovery is important for products"

2. Immediately Actionable

Give people a system they can implement today.

Pure observation gets views. Teaching implementation gets followers and business.

• Don't just point at something, show how to use it
• Walk through the actual process
• Make it immediately applicable

Good: "How to write copy that sounds human with GPT-4.5: [step-by-step process]"

Bad: "The new ChatGPT update is incredible"

3. Sparks Natural Engagement

Never beg for engagement. Structure content so bookmarks and replies happen organically.

Trigger bookmarks with hooks that signal value:

• "How to [solve problem]"
• "Here's what I learned [doing X]"
• "Do this before [common action]"

Drive replies with:

• Strong opinion: Specific, controversial claim people argue or support
• Collaborative invitation: Leave room for others to share approach

Content Selection Framework

Always Share ✓

Consider (Medium Priority)

• Industry takes with unique angle — Your specific perspective from real work
• Process improvements — Must include concrete examples and outcomes
• Technical decisions — Share context and reasoning
• Genuine questions — Ask about specific approaches, not vague "thoughts?"

Skip ✗

Low Priority:

• Progress updates without insights
• Plans before execution (share when shipped)
• Obvious observations
• Feelings about work without value
• Meta-posting about posting

Never Share:

• Engagement bait ("RT if you agree")
• Generic motivation ("Never give up!")
• Vague hype without specifics
• Complaints without solutions
• Announcements of announcements

Red flags in notes (usually skip):

• "Just wanted to share..."
• "Random thought..."
• "Does anyone else..."
• "Feeling grateful..."

Content Mix (Weekly Target)

Over 21 posts per week:

• 8 posts (38%): Educational/How-to
• 6 posts (29%): Personal updates (what you shipped)
• 4 posts (19%): Insights/observations
• 3 posts (14%): Questions/engagement

Don't optimize daily. Post valuable content, mix balances naturally.

Recognizable Style Development

Generic content blends together. Distinctive style makes people stop scrolling.

Develop 3-4 structural patterns that feel natural:

• Specific opening phrases
• Consistent formatting (like using ">" for lists)
• Particular ways you emphasize points
• Phrases you use before certain types of takes

Consistency without formula. Patterns should be recognizable but not templated.

Examples:

• Always using ">" to break down processes
• Starting tool recs with "X handles Y automatically"
• Using "tbh" or "honestly" before controversial takes
• Writing in fragments for emphasis

Pre-Post Checklist

Value:

• [ ] Teaches something specific
• [ ] Saves time or helps avoid mistakes
• [ ] From genuine experience
• [ ] Would I bookmark this?

Specificity:

• [ ] Names tools, projects, or numbers
• [ ] Includes concrete examples
• [ ] Avoids vague adjectives

Actionability:

• [ ] Can someone implement today?
• [ ] Shares HOW, not just WHAT

Clarity:

• [ ] One clear idea
• [ ] Hook in first line
• [ ] Every word earns its place

Engagement:

• [ ] Invites response naturally
• [ ] Specific enough to discuss

The Outcome

Consistent high-value posting builds:

• Audience that respects ideas, not just follower count
• People who stop scrolling when they see your name
• Trust through demonstrated expertise
• Natural conversion when selling (they've experienced your teaching)

Final test: "Would I want to read this if someone else posted it?"

If yes → share. If no → back to notes.

X Threads

Frameworks and structure for multi-tweet threads.

When to Thread vs Single Tweet

Use threads (2-5 tweets) when:

• Narrative with steps ("How we built X")
• Multiple related insights ("3 learnings about X")
• Before/after story ("Tried X, failed, here's what worked")

Use single tweets when:

• One clear insight
• Quick updates
• Strong observations
• Tool recommendations

Default to single tweets. Only thread when story requires it.

Thread Structure

Tweet 1: [Hook with main insight - MOST important] 👇

Tweet 2: [Context or setup]

Tweets 3-N: [Build argument with specifics]

Final tweet: [Land with question or strong statement]

Rules:

• Each tweet must stand alone
• Use N/M numbering for clarity (e.g., 1/7, 2/7... 7/7)
• First tweet MUST end with 👇 or 🧵 to signal thread
• Max 5-7 tweets (longer = blog post)
• Anyone should be able to RT any single tweet

Thread Frameworks

Framework 1: Storytelling

Best for: Personal experiences, case studies

1/ [Hook - outcome or surprising moment]

2/ [Set the scene - context]

3/ [The challenge/conflict]

4/ [The journey - what happened]

5/ [The turning point]

6/ [The lesson]

7/ [CTA - question or follow prompt]

Framework 2: How-To

Best for: Step-by-step guides

1/ [Hook - result they'll get]

2/ [Context/why this matters]

3-N/ [Step-by-step process]

Final/ [Summary + CTA]

Framework 3: Listicle

Best for: Tips, tools, resources

1/ [Hook - what they'll learn, e.g., "10 lessons from..."]

2-N/ [One item per tweet]

Final/ [Summary + CTA]

Framework 4: Contrarian

Best for: Thought leadership, hot takes

1/ [Bold contrarian statement]

2/ [What everyone believes]

3/ [Why they're wrong]

4/ [Your counter-argument]

5/ [Evidence/experience]

6/ [The nuanced truth]

7/ [Engagement question]

Framework 5: Case Study

Best for: Success stories, breakdowns

1/ [Hook - the result achieved]

2/ [Who/what this is about]

3/ [The starting point/problem]

4-N/ [What was done]

N-1/ [Results + key takeaway]

Final/ [CTA]

Framework 6: Problem-Solution

Best for: Educational content

1/ [Hook - the problem everyone faces]

2/ [Agitate - why it's painful]

3/ [Solution intro]

4/ [Step 1]

5/ [Step 2]

6/ [Step 3]

7/ [Results + CTA]

Thread Techniques

Open Loop

End tweet 1 with incomplete information resolved in tweet 2:

• "Here's what happened next:"
• "But here's the thing nobody tells you:"
• "The real story is more interesting:"

Cliffhanger

Use at end of tweets 2-5 to maintain momentum:

• "And then it got worse."
• "But that wasn't the real lesson."
• "Here's where it gets interesting:"

Pattern Interrupt

Break expected flow to recapture attention:

• "Stop."
• "Read that again."
• "This is the part most people skip."

Full Thread Example

Storytelling Thread

1/7 I got fired on my birthday.

No warning. No severance. Just a box and a security escort.

Here's what happened next: 👇

2/7 Some context:

I'd been at the company for 3 years.
Good reviews. Promotions. Everything seemed fine.

Then new leadership came in and cleaned house.

3/7 The first week was brutal.

I applied to 50 jobs. Got 2 responses.
Both rejections.

I started questioning everything about my career.

4/7 Then I did something different.

Instead of applying to more jobs, I reached out to 10 people I admired and asked:

"If you were starting over today, what would you do differently?"

5/7 8 out of 10 said some version of:

"I would have bet on myself sooner."

They'd all spent years building someone else's dream before building their own.

6/7 That's when it clicked.

Getting fired wasn't a setback.
It was a forced decision I'd been too scared to make.

7/7 I started freelancing that month.

First client: $500.
By month 6: $10K/month.
By year 2: More than my old salary.

What's a setback that turned into your biggest breakthrough?

Thread Quality Checklist

• [ ] First tweet ends with 👇 or 🧵?
• [ ] Hook creates immediate curiosity?
• [ ] Each tweet numbered (N/M format)?
• [ ] Each tweet uses most of 280 chars? (230+)
• [ ] One idea per tweet?
• [ ] Smooth flow between tweets?
• [ ] Each tweet could be RT'd standalone?
• [ ] Clear payoff for reading?
• [ ] Has engagement opportunity at end?

What Makes Threads Shareable

People RT threads with:

• Specificity (concrete details vs vague claims)
• Insights (non-obvious learnings)
• Utility (actionable knowledge)
• Relatability (shared struggles + specific solutions)

People reply when you:

• Ask specific questions
• Make specific, debatable claims
• Show genuine curiosity
• Leave room for others to add perspective