In February 2025, Andrej Karpathy described a new way to write code. Give direction. Let AI execute. Iterate on results. He called it vibe coding. Collins Dictionary named it Word of the Year.
Within weeks, marketers asked the same question. What if we worked this way too?
James Dickerson coined the phrase. Greg Isenberg and The Boring Marketer gave it structure. Practitioners started building systems around it.
They called it vibe marketing.
The idea is simple. A marketer becomes an orchestrator of AI workflows and agents. The boring manual work disappears. Creative thinking and prompting take the front seat. You set a goal like “acquire 1,000 users” and the bulk of execution happens for you.
686%
Search growth in 12 months.
MarTech
47%
Of Fortune 500 companies experimenting.
MarTech
$7.29B → $183B
AI marketing agent market by 2033. 49.6% CAGR.
Rellify
AI slop.
9x
Surge in “AI slop” mentions. From 461,000 to 2.4 million in one year.
Meltwater
54%
Negative sentiment toward AI-generated content. October 2025.
Meltwater
80–90%
Of AI agent projects fail in production.
RAND study, 2025
+19%
AI-generated ads outperform human-created ads on clickthrough.
NYU/Emory study via Fortune
−33%
Drop in purchase likelihood when consumers know it was made by AI.
NYU/Emory study via Fortune
AI works better. People trust it less.
This is how most people use AI today.
You
ChatGPT
Generic output
Rewrite 60%
Publish
Forgotten
You open ChatGPT. You write a prompt. You get generic output. You rewrite 60% of it. You publish something passable. Tomorrow you do it again. The AI remembers nothing. You start from zero.
Same input. Same output. Every day.
The Old Way
You are the bottleneck.
The New Way
Review & Direct
Agent System
You set the taste. Agents handle execution.
The 5 Levels of
AI Marketing Maturity
Level 5
Autonomous agents
Self-improving, memory-driven, human oversight
Level 4
Agents with skills
Chained workflows, one command
Level 3
Skills + Foundation
Brand voice baked in, consistent output
Level 2
Manual skills
Reusable methodology, still one-shot
Level 1
Custom prompts
Ad-hoc, no memory, start from zero
The tool is not the advantage.
How you use it is.
The difference between AI slop and AI that works is architecture.
Foundation
Brand voice. Memory. Orchestrator. 30 minutes to set up. This is the context layer. Every agent, every skill, every output starts here. Without it, the AI guesses. With it, the AI knows your brand.
Strategy
Keywords. Positioning. Audience research. The AI analyzes your market, identifies gaps, and builds a research layer that informs everything downstream. No more gut-feel targeting.
Execution
Copy. SEO. Email. Creative. Each task runs through a dedicated skill with domain expertise baked in. Not a generic prompt. A methodology file. Frameworks from Ogilvy, Schwartz, and Hopkins built into the instruction set.
Distribution
One input becomes 16 outputs across 8 platforms. A single blog post turns into tweets, LinkedIn posts, email sequences, ad copy, and more. Formatted for each platform. Matched to each audience.
A skill is transferred expertise.
Not a prompt. A methodology file.
You write the methodology once. It runs forever. And it runs on your voice. The AI does not guess. It follows proven frameworks, scores its own output, and recommends a winner.
Portable across 30+ AI tools. Works in Claude, ChatGPT, Cursor, Codex. Bring it wherever you work.
35 open-source marketing skills on GitHub. 10,500 stars. Free.
Corey Haines, github.com/coreyhaines31/marketingskills
01
Skill.md
The methodology. Frameworks, rules, process.
02
References
Brand voice, positioning, audience docs it reads from.
03
Scripts
Step-by-step execution logic. Generate, score, recommend.
04
Tool Calls
What it triggers. Search, write, publish, deploy.
Just a markdown file. Works in Claude, ChatGPT, Cursor, Windsurf, and 30+ other tools.
A Prompt
“Write me an email.”
A Skill
Lines of structured methodology
Schwartz awareness levels
Proof stacking
CTA patterns
Objection handling
Voice calibration from your brand file
Self-scoring and winner recommendation
Reads from your brand voice file. That is why the output sounds like you.
Skills cross-reference each other. The copywriting skill calls page CRO. The cold email skill pulls from sales enablement. One foundation feeds all of them.
An agent is an autonomous executor.
A skill runs when you call it. An agent reads your goal, plans the steps, picks the right skills, and executes. You give it one command. It decides the workflow.
This is not theoretical. One eight-person team runs 38 agents across 8 departments. Every role is a markdown file. Every department is a folder. The agents handle research, drafting, distribution, and reporting. The humans handle strategy and approval.
The pattern repeats everywhere. Solo founders running entire companies with AI employees built on lunch breaks. Billion-dollar startups replacing legacy platforms with autonomous agents. The shift is not coming. It already happened.
Memory is what makes it compound.
Without memory, every session starts from zero. With memory, the system learns.
Run 1
Brand voice → first draft
Run 5
+ performance data → tighter headlines
Run 20
+ audience feedback → your voice, your patterns
Two halves make this work. The system learns from data. What converted. What bounced. What got clicks and what got ignored. You sharpen it with taste. What to keep. What to cut. What sounds right and what sounds like a machine wrote it.
Every decision gets encoded. Every run gets smarter.
What practitioners say. From Reddit’s r/marketing. 4.2 million members.
The Builders
“A single marketer can outpace a 10-person team.”
Building systems. Shipping daily. Compounding results.
The Pragmatists
“The machine gets you to 80%. A human gets you the rest.”
Automate the boring parts. Keep a human in the loop for trust moments.
The Displaced
“I was forced to use AI until the day I was laid off.”
Corporate content writer. Reddit. Via Blood in the Machine.
Works
- Use AI to research. Publish with human hands.
- Build skills from proven frameworks, not vibes.
- A human approves every output before it ships.
- One agent, one job. No Swiss Army knives.
- Start rough. Iterate with real data. Get to 90%.
Doesn’t
- Let AI post without a human reading it first.
- Ship first drafts. They always sound like a machine.
- Build it once and walk away.
- Optimize for polish. Real writing has rough edges.
- Scale volume when you have nothing worth saying.
The uncomfortable truth. Presented with numbers, not opinions.
30%+
Decline in freelance writing job demand since ChatGPT launched.
Harvard Business School
20%
Headcount loss in early-career marketing roles.
Stanford study
$600K → <$10K
One copywriting agency’s revenue collapse. Eight people. No system to differentiate from AI output. The work disappeared.
Marcus Wiesner, Blood in the Machine
This is not theoretical. 7,000+ job cuts linked to AI in 2025 alone.
Rick Rubin can not play instruments.
Can not operate a mixing board.
His entire value is taste.
Paul Graham put it differently. “When anyone can make anything, the big differentiator is what you choose to make.”
Dane Knecht, CTO of Cloudflare, said the same thing from the engineering side. “Building is easy now. Knowing what to build, and what not to, is the hard part.”
The moat is not technical skill. The moat is judgment.
The risk is not AI replacing you.
It is someone using it better.
You review. Agents execute.
The system proposes. You decide.
AI handles research, first drafts, distribution, formatting, data analysis, scheduling. Humans handle strategy, judgment, brand taste, relationships, the ambiguous and emotional.
The teams that treat AI as infrastructure with human oversight will outperform everyone. The ones that treat it as a replacement will flood the internet with garbage.
The technology is ready. The skills exist. The agents exist. The question is whether you build a system or keep copy-pasting into ChatGPT.
aimarketing.sh · 2026