Systems Not Slides · AI GTM & Marketing Operating System
Your AI Marketing Department. Built on Systems, Not Slides.
{{N_SPECIALISTS}} specialized AI marketing agents, one coordinated operating system, powered by Claude Code and designed to turn GTM work into repeatable systems.
- {{N_SPECIALISTS}}specialist agents
- {{N_DEPTS}}departments
- {{N_SKILLS}}reusable skills
- {{N_WORKFLOWS}}coordinated workflows
Product overview
Stop building assets. Start building marketing systems.
Most marketing teams ship disconnected decks, docs and campaigns. Systems Not Slides installs a coordinated department into Claude Code. Specialists sense the market, shape positioning, scale messaging, run launches that learn, arm sellers, and measure what actually moves deals. Every result feeds shared memory, so the next piece of work starts smarter.
- ListenSourced market, customer and competitor signals
- LearnSynthesis into implications, not news dumps
- ActPositioning, messaging, launches and plays
- MeasureNarrative Fit → Engagement Quality → Sales Momentum
- ImproveLearnings promoted to canon with human approval
Lesson→Actions→Tools→Metric
Evidence, not assertions
Every claim is labelled: hypothesis, AI recommendation, verified evidence, human decision, executed action or measured result. Decision-stage claims need a valid ProofID from the Proof Vault.
Humans stay in charge
Agents draft and recommend. Workflows stop at human approval gates before anything enters the positioning canon, gets published, gets sent or gets budget.
Memory that compounds
Fifteen shared knowledge stores, including ICPs, personas, the Proof Vault, ESOT, message memory and decision history, give every specialist the same source of truth.
Agent directory
{{N_DEPTS}} departments. {{N_SPECIALISTS}} specialists. One Chief of Staff.
Each specialist is a Claude Code subagent with its own mission, triggers, inputs, approved sources, execution steps, output contract, quality bar, handoffs and performance metric.
AI Marketing Chief of Staff
{{COS_DESCRIPTION}}
How it works
From objective to evidence-backed output
Describe your objective
"Our competitor just launched an autonomous close feature. What do we do?"
The Chief of Staff routes it
A deterministic router picks the workflow and the specialists, then shows you the plan and the approval gates.
Agents execute the workflow
Specialists work in sequence or in parallel, handing off artifacts that follow defined output contracts.
Evidence and quality checks
Contract checks, ProofID validation and human approval gates run before anything moves forward.
Learnings feed the next decision
Approved results update the shared knowledge layer and decision history, so the system keeps improving.
Interactive explorer
Browse departments, agents, skills and workflows
Example use cases
What you can run on day one
Each use case maps to a workflow you can try first in demo mode, on a fictional example company, without connecting any of your own systems.
Methodology
Built on Systems, Not Slides
The operating system turns the frameworks in Ray Beharry's book Systems, Not Slides: An AI Blueprint for Product Marketing (2025) into working agents, skills and workflows.
Core frameworks include the Continuous Market Intelligence Loop, the Adaptive Positioning Loop, Brand Voice Memory, Message Kernels, the Proof Vault and ProofIDs, the Learning Launch Loop, the Enablement Source of Truth, Line → Proof → Next Step, the Predictive CI Loop and the PMM Impact Chain. SEO, demand generation, organic social and community, editorial content, lifecycle email and conversion optimization are new execution capabilities this product adds on top of the book.
Where each capability comes from
- {{N_CORE}} agents implement core book methodology
- {{N_ADAPTED}} agents adapt book concepts into new formal roles
- {{N_EXTENSION}} agents are product extensions beyond the book
A methodology map in the package traces every agent, skill and workflow to its source chapter.
Installation
Runs where you already work: Claude Code
- Claude Code (terminal, desktop app, web or IDE extension) with a Claude plan or API access
- Python 3.10+ with PyYAML and jsonschema for the validators and demo mode
- Optional: exports or MCP connections for GA4, Search Console, HubSpot, Salesforce, LinkedIn, Semrush/Ahrefs, Notion, Airtable, Slack, Gong, n8n/Zapier
No credentials are needed to try it. Demo mode runs every workflow end-to-end on a fictional company with synthetic data.
# 1. Unzip the release and open it in Claude Code
cd systems-not-slides-ai-marketing-department-v{{VERSION}}
python3 -m pip install -r requirements.txt
# 2. Check everything and watch a workflow run
python3 scripts/sns.py validate
python3 scripts/sns.py demo wf06-competitive-response
# 3. Start Claude Code and onboard your company
claude
> /sns-setup
> /wf-market-pulse
Download
Get the AI Marketing Department
Version {{VERSION}}. The release ZIP or repository link will appear here once it has been approved for distribution.
Download link coming soonFAQ
Questions, answered plainly
What do I need to run it?
Claude Code and Python 3.10 or newer. The agents, skills and workflows are plain files that Claude Code loads from the project folder. The Python scripts provide validation, routing, run state, metrics and imports.
Do I have to connect my marketing tools?
No. Every integration is optional. You can start in demo mode, then import CSV or JSON exports from GA4, Search Console, HubSpot, Salesforce, LinkedIn, Semrush, Ahrefs, Notion, Airtable, Slack, Gong, n8n or Zapier. Live API or MCP connections are documented as configuration you set up and verify in your own environment.
Where does my data go?
Your knowledge stores are files in your own project folder. The system adds no servers or tracking. Model calls go through your own Claude Code setup and are governed by your account's terms. Credentials belong in environment variables, never in the repository.
Will the agents publish, send or spend on their own?
No. Agents draft and recommend. Workflows stop at human approval gates, and publishing, sending and spending stay with your team.
Can it make up results or competitor claims?
The system is designed against that. Agents label evidence classes, decision-stage claims must cite a valid ProofID, and governance hooks block knowledge edits that would put unapproved or unproven claims into the canon. As with any AI system, a human still reviews outputs before they are used.
What is it good for, and what isn't it for?
It covers market intelligence, positioning, messaging, launches, sales enablement, competitive response, SEO, demand generation, organic social and community, editorial content and newsletters, lifecycle email, conversion optimization, analytics and marketing operations. It does not replace your CRM, analytics or marketing automation platforms. Its outputs are only as current as the evidence you give it.
How does it relate to the book?
It implements the frameworks from Systems, Not Slides (2025) by Ray Beharry as working agents and workflows, and adds new execution capabilities. The package contains original implementation instructions with attribution, not the book's text.