The argument in one line.
Growing from 1,200 to 14,847 LinkedIn followers in 90 days required moving through four escalating levels of Claude usage, ending in a fully autonomous system that researches, writes, posts, and DMs commenters without human involvement.
Read if. Skip if.
- LinkedIn creators who are posting consistently but growing slowly and suspect they're stuck at a surface-level AI workflow
- Founders and B2B marketers who want to see a full four-level progression from basic Claude prompting to fully autonomous posting and DM sequences
- Non-technical users who want to grow on LinkedIn using Claude but aren't sure where to start or how far it can go
- People not focused on LinkedIn — the entire system is built and calibrated for that platform specifically
- Viewers who already run a fully automated content-to-DM pipeline and are looking for edge optimizations, not the full walkthrough
The full version, fast.
Growing on LinkedIn with AI is not a prompting problem — it is a level problem. Most creators are stuck at level one (basic chat prompting) without knowing that three more levels exist. This walkthrough demonstrates four escalating stages: level one is standard Claude Chat prompting with research enabled, level two is Claude Projects with a tone-of-voice file that makes every post sound like you, level three is Claude Code skills using Playwright to automate browser-based publishing without touching LinkedIn's API, and level four is a fully autonomous Obsidian vault system that generates, schedules, posts, and sends follow-up DMs without manual input. The progression from 1,200 to 14,847 followers in 90 days tracks directly with moving from earlier to later levels.
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.

01 · Hook + gap framing
Result claim (1,200 to 14,847 followers), gap insight (most people stuck at Level 1), promise of all four levels.

02 · Level 1: Basic Claude Chat + Research Mode
Live demo in Claude.ai: prompt for viral LinkedIn post, select angle and audience, enable Research mode to pull hard statistics into the hook.

03 · Sponsor (self-plug: masterclass)
Duncan plugs his own Claude Code + personal brand masterclass — 10-second self-sponsor.

04 · Level 2: Claude Projects + Tone-of-Voice Files
Creates a LinkedIn Post Writer project. Generates a system prompt using Claude. Uploads tone-of-voice (Yoda persona demo), social proof, and personal story docs. Posts now have identity.

05 · Level 3: Claude Code Skill + Playwright Auto-Publish
Builds a Claude Code skill in plan mode. Installs Microsoft Playwright CLI via GitHub. Skill researches Reddit, writes in voice, and browser-automates the LinkedIn publish with zero copy-paste.

06 · Level 3.5: Routines / Daily Schedule
Sets up a Claude Code Routine or /schedule command to run the LinkedIn skill every morning at 8AM, cycling through keyword lists.

07 · Level 4: Obsidian Vault Knowledge Brain + LeadShark
Full autonomous stack: Vault holds audience notes, performance data, recent posts (prevents repetition), coaching transcripts. Reddit-mines for pain. Auto-DMs commenters via LeadShark. CTA to The Build Room community.
Lines worth screenshotting.
- Going from 1,200 to 14,847 LinkedIn followers in 90 days came from upgrading AI usage level, not from posting more frequently.
- Most people using AI for LinkedIn content are stuck at level one — basic chat — without knowing levels two, three, and four exist.
- Posts with a hard statistic in the opening hook consistently outperform posts without one.
- Claude Projects gives you a persistent system prompt and file library so every post follows the same instructions without re-prompting.
- The Microsoft Playwright CLI gives Claude browser access, so it can research, write, and publish a LinkedIn post without you touching anything.
- Saving your cookies after the first login means the browser automation runs fully hands-free on every subsequent session.
- A scheduled Claude skill can post to LinkedIn every morning at 8AM by cycling through a keyword list you define once.
- Auto-DMing every commenter via a tool like LeadShark turns a single LinkedIn post into an automated lead generation loop.
- A knowledge vault containing coaching calls, testimonials, performance data, and personal stories is what separates content that sounds like you from content that sounds like everyone else.
- When no topic is given, a well-built skill can auto-scan Reddit for what your ideal customer is complaining about right now and pick the most relevant angle.
- Knowing your recent post history inside the skill prevents the system from repeating the same content day after day.
- The gap between level one and level four is not technical skill — it is knowing these levels exist.
The ladder format is the playbook.
Four levels of anything — from basic to autonomous — is the most frictionless tutorial format on YouTube right now.
- The gap framing ('most people are stuck at Level 1') beats a benefits list — name the ceiling people are bumping against.
- Self-generating the system prompt on camera (Level 2) is a trust hack — it shows how to get started, not just what the end state looks like.
- The Playwright reveal is the money shot — real browser automation live on screen. Any Claude Code skill that touches the outside world deserves its own demo moment.
- Level 3.5 (routines/scheduling) is easy to miss but worth stealing — a half-step that bridges demo and daily practice.
- The Obsidian Vault as a personal knowledge brain is the most transferable concept — Joe's equivalent is a JoeFlow knowledge vault that feeds dictation sessions, not just LinkedIn posts.
- His skill's one rule ('the hook cannot be mediocre') belongs in every content system — encode it as a constraint, not a suggestion.
Terms worth knowing.
- Claude Projects
- A workspace feature inside Claude that lets users store persistent instructions, tone-of-voice files, and reference documents so the model applies them consistently across every conversation in that project.
- System prompt
- A set of instructions given to an AI model before any user interaction begins, establishing its role, rules, and behavior for the entire session.
- Claude Code skills
- Custom reusable automations built inside Claude Code — similar to scripts — that can perform multi-step tasks like researching a topic, drafting content, and publishing it, all triggered by a single slash command.
- Microsoft Playwright CLI
- A command-line browser automation tool that lets a program control a real web browser — clicking buttons, filling forms, navigating pages — without human interaction.
- Connectors (Claude Code)
- Pre-built integrations in Claude Code that link the agent to external services like GitHub, Notion, Gmail, or Google Drive so it can read from and write to those platforms directly.
- Obsidian Vault
- A local folder of Markdown files managed by the Obsidian note-taking app, used here as a personal knowledge base storing the creator's stories, performance data, audience notes, and proof points.
- ICP (Ideal Customer Profile)
- A detailed description of the specific type of person or business a creator or company most wants to reach, used to filter content topics, messaging, and offers.
- LeadShark
- A LinkedIn automation platform that can automatically send direct messages to users who comment on a post, typically used to deliver lead magnets or start conversations at scale.
- Routine (Claude Code)
- A scheduled, recurring automation in Claude Code that runs a specified skill or set of instructions at a fixed time — for example, publishing a LinkedIn post every morning at 8 AM.
Things they pointed at.
Lines you could clip.
“Most people don't even know that levels two, three, and four even exist. And this gap is costing you followers every single day.”
“Most knowledge workers are one bad quarter away from being replaced, not by a smarter person, by a $20 a month subscription.”
“Quad now has access to your browser, and so anything that you can do on the internet, Claude can do for you.”
“The first line of the post has to stop someone mid-scroll. Everything else can be mediocre. The hook cannot.”
Word for word.
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
The bait, then the rug-pull.
Fourteen thousand, eight hundred and forty-seven. Duncan Rogoff didn't round down — and that specificity is the whole tell. This isn't a LinkedIn guru's boast; it's a timestamp on an experiment, and the video is the methodology. Four levels of Claude. Most creators are stuck at Level 1. This is what the other three look like.
Named ideas worth stealing.
The 4 Levels of Claude
- Level 1: Basic Chat + Research Mode
- Level 2: Claude Projects (system prompt + personal files)
- Level 3: Claude Code Skill + Playwright auto-publish
- Level 4: Obsidian Vault knowledge brain + auto-DM
A progressive skill ladder for Claude-powered LinkedIn content. Each level removes more manual friction until the system runs itself.
LinkedIn Post Recipe (Level 4 skill)
- 1. Name the pain (audience-aware from Vault)
- 2. Show personal proof (how you solved it)
- 3. Point to the community/offer
Three-part LinkedIn post structure baked into his Level 4 skill: pain, proof, CTA.
How they asked for the click.
“If you wanna get access to the exact systems that I use to grow over 200,000 followers and sign multibillion dollar clients to my agency, just check the link in the description.”
Soft close — The Build Room Skool community shown on screen. Also plugs a follow-up Obsidian setup video. Clean double-exit: description link plus related video.









































































