Is Vibe Marketing as good as they say? Our honest take.

Claude Code's source code was leaked today

Breaking news! Anthropic published Claude Code publicly but accidentally included a file that lets you reverse-engineer the compressed code back into its original, readable form. Someone noticed, used it to reconstruct the full source code, and shared it. Look out for the "10 things I learned from reading Claude Code's source code" posts coming to everywhere soon.

Factory Missions: The Closest Thing to AGI Yet

Factory has launched Missions, long-running AI agents that tackle massive software tasks from start to finish. These aren't your typical chatbots - they'll build entire applications end-to-end without breaking a sweat. Early users describe it as the closest feeling to AGI they've experienced, which is saying something in today's hype-heavy AI landscape.

Google AI Studio Gets a Multiplayer Makeover

Google AI Studio just received a major upgrade that transforms it from a simple AI playground into a proper development platform. The standout feature is multiplayer functionality for real-time collaboration, plus live data connections through Firebase integration and professional UI components. It's Google's answer to making AI app development as social and seamless as a Google Doc.

For as long as we've been builders, we've wished marketing could be fully automated. Now, apparently, it can.

But does it really work? And can solo founders and small teams leverage it?

Strip away the jargon and marketing is basically three things:

• find out who your customers are,
• make stuff they want to see,
• and put it where they'll actually see it.

Vibe marketing means AI handles all three. The human just manages the AI.

Obviously there are loads ways to build AI marketing teams, and many different tools you can use. We're going to focus on the most commonly used ones at this time.

The Architecture of AI Marketing Teams

Building these systems requires four core components working in harmony:

• The Interface: Claude Code or OpenClaw live on your computer, giving AI the ability to create files and execute tasks
• MCPs (Model Context Protocol): Act as hands and eyes for AI to interact with external tools like browsers and analytics platforms
• Skills: Simple text documents that serve as instruction manuals for proven marketing frameworks
• Agents: Specialised AI employees with specific roles and toolsets

The magic happens when these elements combine. Your AI team can pull Google Analytics data through Zapier MCP, analyse competitor websites via Playwright MCP, then generate McKinsey-quality reports without human intervention.

Real-World Success Stories

James Dickerson built an entire marketing agency called "Boring Money" targeting local home service businesses. His process flows seamlessly from competitor research through Perplexity MCP to landing page generation using Anthropic's Front-End Design skill.

The results speak volumes. AI agents read traffic data directly from Google Analytics 4, then produce detailed client reports that rival traditional consulting firms.

Grace Leung created a five-agent AI marketing team for her travel brand:

• Data Analyst pulls performance metrics
• Content Creator writes blog posts and social media content
• Market Researcher identifies trending topics and competitor strategies

Her team operates through a shared Notion Kanban board. She drops tasks in, AI agents execute them, then mark completion. She can even text her AI team from her phone to trigger specific workflows.

The Implementation Blueprint

Start with context loading. Create a "Knowledge Folder" containing brand voice guidelines, product descriptions, audience personas, and high-performing content examples. Command your AI to read these files first in every interaction.

Equip your AI with tools. Essential MCPs include Perplexity for market research, Playwright for competitor website analysis, and Zapier for CRM and analytics integration.

Build your skills playbook. Find content templates you love, ask AI to analyse why they work, then save those insights as text files. Reference specific skills instead of generic prompts.

The architecture matters more than individual tools. Create specialised agents rather than one generalist AI. Your Copywriter Agent gets brand voice files plus direct response skills. Your Design Agent receives visual templates and image generation capabilities.

Advanced Quality Control Systems

AI-generated content often suffers from what practitioners call "AI slop." Combat this through systematic quality controls.

Front-End Design Skills help avoid purple gradients, excessive rounded corners, and emoji overuse. Self-review loops force AI to check for hallucinated brand names and incorrect aspect ratios before submission.

Negative prompting works particularly well. Explicitly state "Do not include logos, brand names, company names, watermarks, or additional text" in your instructions.

The reference-based method provides the strongest foundation. Feed existing branded materials to AI, generate analysis, then create permanent skills from that analysis.

Monetisation Through Distribution

Smart practitioners use programmatic SEO strategies. Scrape structured data through Firecrawl, generate thousands of targeted pages, then capture organic traffic at scale.

The numbers work beautifully. Generate 10,000 pages targeting "Best CRM for Dentists in Miami" style queries. Even 30 visits per page creates 300,000 monthly visitors.

Upwork automation offers immediate revenue opportunities:

• Sub-agents search for RPA jobs worth $500-$5,000
• Build functional mini-demos based on job descriptions
• Send live demonstrations with proposals

MCP servers function as automated sales teams. Build servers that answer product-related questions, publish to MCP registries, then watch AI assistants recommend your product organically.

Human In The Loop

The most successful practitioners maintain human checkpoints. AI builds drafts, humans approve before publication. Work flows through "Draft" folders or Slack channels for approval.

Remote orchestration changes everything. Generate secure links to manage your AI team from anywhere. Your marketing operations become truly location-independent.

The shift from learning complex tools to managing intelligent agents represents a fundamental change in how marketing gets done. The question isn't whether this approach will scale. It's whether you'll adapt before your competitors do.

Here are some of our actual uses so far:

• Kieran has a number of websites which he's using AI to build SEO presences for. Each day a cron job runs which uses his Keywords Everywhere API key to do keyword research, and identify words with high volume and low competition. Then AI will research and write SEO optimised blog posts to target that term. It happens fully on autopilot.
• James is using Claude Code to manage Create With's ad campaigns. It creates all of the ad copy and images, and once approved creates the ad campaigns in our Meta dashboard. It monitors ad performance and uses the information to improve for next time.
• James built an OpenClaw which creates social media posts across multiple platforms and then sends drafts to a Slack thread where he can approve them. He said the flow works fine but the quality of posts is, as always, varied.
• Kieran uses Claude Code with the Sales Navigator MCP to generate lists of potential prospects on LinkedIn. Then he uploads the list to Prosp.ai and lets it run automated outreach.
• We've also shipped a few "tools-as-marketing" for Create With. These are alternatives to blog posts which can be really engaging for site visitors. Here's an example for finding your perfect tool.

In this week's Create with Podcast episode, James and Kieran tackle the often-overlooked world of testing for vibe coders and explore the rise of autonomous AI agents.

Key topics covered:

• Why vibe coders need to start writing tests (and how Claude makes it surprisingly easy)
• Exploring Convex as a superior alternative to Supabase for AI-powered development
• The app store crisis: How AI-generated apps are overwhelming Apple and Shopify's review processes
• Ditching expensive SaaS tools by building custom replacements with AI
• Setting up OpenClaw agents to handle marketing tasks autonomously via Slack
• Claude's new scheduled tasks feature and what it means for automation

Watch Video

Does AI understand if a user write in all capitals they're effectively shouting at it via the keyboard? Or does it just assume its human has accidentally left caps lock on?

David J Phillips

"Make no mistakes DO NOT HALLUCINATE. YOU ARE AN EXPERT SOFTWARE ENGINEER"

— @davj