The Agentic Marketing Stack: How Greg Isenberg Built a Full Marketing Pipeline With Claude Code

From Raw Idea to Live A/B Test in One Terminal Session

Greg Isenberg recently broke down a marketing workflow on The Startup Ideas Podcast that stopped us in our tracks. Not because it was theoretical — but because he and his guest Amir demonstrated it live, end to end, in a single sitting. They took a raw business idea and turned it into a live, conversion-tracked landing page with an active A/B experiment running on it. No designer. No front-end developer. Just a terminal and five connected tools.

The workflow Greg shared combines context-driven planning, AI-powered design iteration, automated page building with Claude Code, real-time analytics, and — the part most marketers still miss — programmatic social media distribution through tools like Postiz and its MCP server. Here is how every piece fits together, and how you can run the same stack yourself.

Why Context-First Planning Changes Everything

Most people start their AI marketing workflow by typing “build me a landing page for X” into a terminal. The result is always generic. The AI has zero context about the business, the target customer, or the competitive landscape. Greg pointed this out as the single biggest mistake he sees.

The fix is straightforward: start with context, not code. In the workflow Greg demonstrated, Amir used Idea Browser — a tool that connects directly to Claude Code as an MCP (Model Context Protocol) server. That connection lets Claude Code pull project files containing the ICP definition, positioning statements, competitive analysis, and growth strategy before writing a single line of code.

For the demo, they worked on a real project: an AI sparring partner for B2B sales teams, specifically niched into freight software. The project already had an offer definition file covering the target customer profile, the transformation promise, the value proposition, and competitive positioning. When Claude Code connected to Idea Browser through MCP, it ingested all of that context automatically.

The biggest problem right now: everyone can build landing pages and everyone has ideas. The gap is knowing where to get customers and having confidence the idea is worth pursuing. Context-first planning closes that gap.

This approach matters because AI agents are only as good as the context they receive. A landing page built with full business context converts differently than one built from a one-line prompt. The data compounds over time — every experiment result, every customer insight, every pivot gets stored and referenced in future sessions.

Skills-Based Asset Generation: One Prompt, One Complete Deliverable

Idea Browser includes what they call “skills” — specialized prompt templates designed for specific business tasks. Greg watched Amir use two of them live: Lead Magnet Legend (for creating a lead capture offer) and Landing Page Architect (for structuring the page).

The Lead Magnet Legend skill analyzed the freight software niche and produced a specific lead magnet concept: “5 Objections That Kill Freight Software Deals.” It saved that concept as a file inside the project context, meaning every future prompt in that session — and every future session — would have that decision baked in.

The prompt Amir used was a single sentence:

Connect to Idea Browser MCP. Look at my AI B2B sparring partner project.
Pull the right context. Then use the Lead Magnet Legend skill
to create a lead magnet landing page.

One prompt. The agent read the project, selected the skill, generated the asset, and stored the output for the next step. No back-and-forth. No manual file management. This is the pattern that separates a casual AI user from someone running a repeatable marketing operation.

Design Iteration Before Code: Why Paper Changes the Game

Here is where most “vibe-coded” marketing projects fall apart. Developers iterate directly in code, lose track of visual versions, and end up with a page that looks like it was designed at 2 AM. Greg called this out explicitly — the design step is where amateurs and professionals diverge.

Paper solves this by connecting to Claude Code and providing an actual design interface. You create visual iterations — hero sections, testimonial blocks, pricing cards — and refine them before any production code gets written. Think of it as a design tool where your AI agent can read the output and build directly from it.

During the demo, Amir refined hero sections, swapped layouts, and tested different visual hierarchies. The result was a landing page with professional typography, intentional spacing, and real visual hierarchy. It looked designed, not coded.

The workflow is simple:

1. Open Paper and create a component (hero section, CTA block, testimonial grid)
2. Iterate on the visual direction until you are satisfied
3. Claude Code reads the design and builds the production page from it

This step adds maybe 15 minutes to your workflow but completely changes the quality of what ships. Professional design is still a differentiator, even in the age of AI-generated everything.

Analytics and A/B Testing From the Terminal With Humblytics

Once the page went live, Amir showed how Humblytics handles the analytics layer. Standard traffic data, channel attribution, heatmaps for scroll depth and click tracking, and full funnel visualization — all connected and running from the moment the page deployed.

But the part that made Greg react on air was the A/B testing. Amir used Claude Code to hit the Humblytics API and create an A/B experiment directly from the terminal:

Using the Humblytics API, create an A/B experiment (no-code type)
for the headline on this landing page.

Within seconds, the experiment was live. The control headline stayed in place. A variant rotated in automatically for a split of the traffic. Humblytics tracked conversions, clicks, and engagement for both versions in real time. No code deploy. No staging environment. The Humblytics script dynamically updated the page content on the fly.

The variant it generated: “Every lost deal started with an objection your rep wasn’t ready for.”

Greg’s reaction: “Honestly, not bad. Pretty fire.”

Humblytics also has a recommendations endpoint. Point your agent at it, and it scrapes your site, pulls traffic insights, reads your existing analytics data, and returns a detailed report on what to test next. The experiments write themselves.

The Missing Piece: Automated Social Distribution With Postiz

Greg’s workflow covers idea validation through conversion testing. But there is one more piece that turns a tested landing page into actual traffic: distribution. This is where Postiz fits into the agentic marketing stack.

Postiz is an AI-powered social media management platform that connects to 32+ channels — X, LinkedIn, Instagram, TikTok, YouTube, Reddit, Facebook, Pinterest, Threads, Bluesky, Discord, Slack, and more. What makes it relevant to this stack is that Postiz exposes both a public API and an MCP server, meaning Claude Code can schedule and publish posts programmatically — the same way it creates landing pages and runs A/B tests.

Connecting Postiz to Claude Code via MCP

The Postiz MCP server lets any MCP-compatible AI agent manage social media through natural language. The setup takes about 30 seconds. Add this to your Claude Code MCP configuration:

{
  "mcpServers": {
    "postiz": {
      "url": "https://api.postiz.com/mcp/YOUR_API_KEY"
    }
  }
}

Once connected, Claude Code gets access to eight tools: listing your connected social accounts, getting platform-specific rules and character limits, scheduling posts, generating AI images, creating videos, and triggering platform-specific helpers (like fetching Reddit flairs or YouTube playlists).

Scheduling Posts From the Terminal With the Postiz CLI

For developers who prefer direct CLI control, the Postiz CLI (npm install -g postiz) gives you full programmatic access:

# Authenticate
postiz auth:login

# List your connected social accounts
postiz integrations:list

# Schedule a post to X/Twitter
postiz posts:create \
  -c "We just published our freight software objections guide. Here are the 5 deal-killers every rep should know:" \
  -s "2026-04-17T14:00:00Z" \
  -i "your-x-integration-id"

# Schedule a thread
postiz posts:create \
  -c "Thread: 5 objections killing your freight software deals" \
  -c "1/ The pricing objection. Most reps fumble this because..." \
  -c "2/ The integration objection. Your prospect's IT team will ask..." \
  -s "2026-04-17T14:00:00Z" \
  -i "your-x-integration-id"

# Upload media and attach to a post
postiz upload ./landing-page-screenshot.png
postiz posts:create \
  -c "New guide: handling freight software objections" \
  -m "https://uploads.postiz.com/your-image.png" \
  -s "2026-04-17T15:00:00Z" \
  -i "your-linkedin-integration-id"

Every command outputs JSON, making it scriptable with jq or any automation tool. You can batch-schedule an entire week of promotional content for your landing page in a single shell script.

Multi-Platform Distribution in One API Call

The Postiz API supports multi-platform posting in a single request. You can send platform-specific content — a short punchy post for X, a longer thought leadership piece for LinkedIn, a visual-first post for Instagram — all in one POST /public/v1/posts call:

curl -X POST https://api.postiz.com/public/v1/posts \
  -H "Authorization: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "type": "schedule",
    "date": "2026-04-17T14:00:00Z",
    "posts": [
      {
        "integration": {"id": "x-integration-id"},
        "value": [{"content": "Short punchy post for X", "image": [], "delay": 0}]
      },
      {
        "integration": {"id": "linkedin-integration-id"},
        "value": [{"content": "Longer thought leadership piece for LinkedIn...", "image": [], "delay": 0}]
      }
    ]
  }'

This is the distribution layer Greg’s stack was missing. You validate the idea with Idea Browser, design it with Paper, build it with Claude Code, test it with Humblytics, and then push promotional content across 32 channels with Postiz — all from the same terminal session.

The Compound Effect: Why Results Feed Back Into Context

The final layer in Greg’s framework is feeding results back into your project context. After each experiment, you add the performance data — which headline won, what converted, where traffic came from — into Idea Browser’s project files. The next time you prompt for a landing page, a growth strategy, or a new lead magnet, your agent starts from real data instead of guessing.

Amir’s team does this with a master performance log. Every A/B test result, every revenue figure, every optimization gets stored. When they spin up the next campaign, the AI starts from historical performance data.

With Postiz in the mix, the feedback loop extends to social distribution too. The Postiz analytics API (postiz analytics:platform <integration-id>) gives you per-channel performance data — followers gained, impressions, engagement rates. Feed that back into your project context and your AI agent learns which channels drive traffic, which messaging resonates, and where to double down.

“If you have that context compounding over time, you’re just going to get better results. That’s how you compete.” — Amir

The Arbitrage Window Is Open Right Now

Greg made a comparison that stuck: Facebook ads used to cost 5 cents a click. Amir remembers paying $0.05 per click and charging companies $2 per install. The margin was massive because almost nobody knew the platform existed yet. Today, average CPC sits around $2-3. The window closed because everyone caught up.

The same window is open right now with agentic marketing tooling. According to Greg, 99.999% of people have never seen this stack. They are still building landing pages manually, guessing at headlines, and waiting weeks for analytics data. Someone running this workflow can go from raw idea to live A/B experiment in a single session.

There is also a services play here. Amir mentioned customers paying $5,000 to $10,000 per month for his team to run this exact workflow on their behalf: connect ad platforms, get full attribution, optimize landing pages, and report back on where top-of-funnel spend should go.

“99.999% of people don’t know that this stack exists. And there’s an opportunity to create offers and arbitrage right now.” — Greg

Your Action Checklist: Run This Stack Today

Here is the step-by-step process to replicate Greg’s agentic marketing workflow:

1. Set up Idea Browser and connect it to Claude Code as an MCP server. Create a project with your ICP, positioning, and offer definition.
2. Use a skill (Lead Magnet Legend or Landing Page Architect) to generate your first marketing asset from a single prompt.
3. Open Paper and design your landing page components visually before building in code. Iterate on hero sections, CTAs, and layout before a single line of production code gets written.
4. Build the page with Claude Code. Let it read your Paper designs and project context to generate a professional, conversion-focused landing page.
5. Install Humblytics on your live page for traffic analytics, scroll-depth heatmaps, click maps, and funnel tracking.
6. Run your first A/B experiment from the terminal using the Humblytics API. Test headlines, CTAs, or hero images without deploying code.
7. Connect Postiz to Claude Code via MCP or CLI. Schedule promotional content across all your social channels directly from the terminal.
8. Log results back into your project context — both landing page performance and social media analytics — so your AI agent compounds learning over time.

Context Compounds. Tools Multiply. Distribution Scales.

The stack Greg demonstrated works because each tool feeds the next. Idea Browser holds the context. Paper refines the design. Claude Code builds the page. Humblytics runs the experiment. Postiz distributes the content. And the results flow back into Idea Browser for the next cycle.

One session. One terminal. A live, tested, conversion-tracked landing page with social media promotion running across 32+ channels. No designer. No developer. No guessing.

The tools exist today. The arbitrage window is open. The question is whether you start running this workflow now — or wait until everyone else catches up.

Ready to add automated social distribution to your marketing stack? Try Postiz free and connect it to Claude Code in under a minute. Schedule posts across 32+ social channels from your terminal, track analytics per channel, and feed performance data back into your AI workflow. Your agentic marketing stack is not complete without the distribution layer.