Skip to content
RAY BEHARRY

AI Marketing Department · v1.1.0

A complete marketing department, built as a system.

70 specialist agents in 14 departments, coordinated by an AI Chief of Staff, running 14 governed workflows on Claude Code. It is the working implementation of Systems, Not Slides.

agents
71agents
departments
14departments
workflows
14workflows
knowledge stores
15knowledge stores
human approval gates
20human approval gates
output contracts
60output contracts

The business story

A company needs to launch a new product. Here is how the department responds.

Seven questions every launch has to answer, each handled by one of the department's real workflows. Every step shows who does the work, what evidence it relies on, where a person decides, and what gets delivered.

  1. What changed in our market, and does it matter for this launch?

    Workflow 01 · Market Pulse →
    Who does the work
    Market IntelligenceCompetitive IntelligenceMarketing Operations
    Evidence it uses
    Company Context · ICP Library · Market Signals · Buyer Personas · Competitive Intelligence
    Who decides
    Head of Product Marketing (or delegated PMM)
    What you get
    Approved weekly Market Pulse: sourced signal table, ≤3 implications, owned actions, Lesson → Actions → Tools → Metric.
  2. Who is this product for, and why do we win?

    Workflow 02 · Positioning Engine →
    Who does the work
    Market IntelligencePositioning & Strategy
    Evidence it uses
    ICP Library · Company Context · Product Context · Competitive Intelligence · Market Signals
    Who decides
    Head of Product Marketing (with CMO sign-off for category changes)
    What you get
    Approved, versioned positioning canvas in positioning-canon with proof links, validation plan and Lesson → Actions → Tools → Metric.
  3. What do we say, to whom, and on which channel?

    Workflow 03 · Messaging Engine →
    Who does the work
    Messaging & ContentMarketing OperationsPositioning & Strategy
    Evidence it uses
    Positioning Canon · Buyer Personas · Proof Vault · Brand Voice Memory · Message Memory
    Who decides
    Head of Product Marketing (brand lead for voice exceptions)
    What you get
    Approved message kernels plus persona × channel variant grid stored in message-memory, each line tagged with proof, stage and approver.
  4. How do we launch so that we learn fast?

    Workflow 04 · Learning Launch →
    Who does the work
    Launch & GTMSales Enablement
    Evidence it uses
    Positioning Canon · Message Memory · ICP Library · Launch Learnings · Product Context
    Who decides
    Launch owner (PMM lead) with budget holder for any spend; Head of Product Marketing
    What you get
    Approved Launch Learnings Memo: bets confirmed/killed, Day 7/14/30/60 readouts, extend-or-evolve decision and proposed canon updates.
  5. What do sellers say when a buyer pushes back?

    Workflow 05 · Sales Enablement Engine →
    Who does the work
    Sales EnablementCompetitive Intelligence
    Evidence it uses
    Buyer Personas · Competitive Intelligence · Enablement Source of Truth (ESOT) · Message Memory · Proof Vault
    Who decides
    Pricing owner (pricing) or Product lead (roadmap); PMM for other topics; Sales Enablement lead with PMM owner
    What you get
    Approved ESOT changes and the LPNS answer that triggered them, with objection lab evidence and owner confirmations.
  6. Is it working, and what should change?

    Workflow 11 · GTM Performance Engine →
    Who does the work
    Analytics & Learning
    Evidence it uses
    Performance Observations · Campaign Experiments · Positioning Canon · Message Memory · ICP Library
    Who decides
    CMO or VP Marketing (budget holder for Scale moves)
    What you get
    Approved Command Board: narrative fit, engagement quality, sales momentum and revenue outcomes with owned Stop / Start / Scale decisions.
  7. What does leadership need to decide?

    Workflow 12 · Executive Marketing Brief →
    Who does the work
    Market IntelligenceAnalytics & LearningCompetitive IntelligenceSales EnablementMarketing Operations
    Evidence it uses
    Market Signals · Decision History · Performance Observations · Competitive Intelligence · Enablement Source of Truth (ESOT)
    Who decides
    CMO (or executive sponsor)
    What you get
    Executive-approved marketing brief with ≤3 prioritized decisions, each logged to decision-history and routed to an owner.

Departments, evidence, approvers and deliverables come straight from the workflow specs in the v1.1.0 package. Nothing here is a client result.

Executive overview

What the department does for the business.

It runs marketing as one loop: Listen → Learn → Act → Measure → Improve. Every workflow delivers Lesson → Actions → Tools → Metric, and every claim says whether it is a hypothesis, an AI recommendation, verified evidence, a human decision, an executed action or a measured result.

  1. 01

    listen

  2. 02

    learn

  3. 03

    act

  4. 04

    measure

  5. 05

    improve

Departments are placed by the loop stage of the workflow steps their agents run.

Departments

Fourteen departments of five specialists.

Each department owns a part of go-to-market, with its own skills, workflows and the knowledge it is responsible for.

Org chart

Every agent, and how they connect.

Open a department, select an agent and follow its handoffs. The knowledge layer lights up with what that agent reads and writes.

Shipped in v1.1.0 Every agent, workflow and knowledge store shown here comes straight from the downloadable package's own registry. Workflow walkthroughs step through the real specs. They are not live runs and contain no client data.

Human leadership

You: CMO or Head of Marketing

Sets objectives and approves every gate: 20 named approvals across the workflows.

Agent 71 · Orchestration

AI Marketing Chief of Staff

Routes requests, selects specialists, manages handoffs, enforces human gates and validates outputs.

Shared knowledge layer · 15 governed stores

  • Company Context
  • Product Context
  • ICP Library
  • Buyer Personas
  • Market Signals
  • Competitive Intelligence
  • Positioning Canon
  • Brand Voice Memory
  • Message Memory
  • Proof Vault
  • Enablement Source of Truth (ESOT)
  • Campaign Experiments
  • Launch Learnings
  • Performance Observations
  • Decision History

How to explore

Open any of the 14 departments to see its specialists. Select an agent to see who it receives work from, who it hands off to, which knowledge it reads and writes, and where a person approves.

Workflows

Watch the departments work together.

Step through any of the fourteen workflows: which agent runs each step, what it reads, what it writes back, and where a named person approves.

Workflow walkthrough · 14 shipped workflowsStep-through of the real workflow spec, not a live run
Choose a workflow

Market signals → synthesis → strategic implications → recommended actions. A weekly, sourced brief that reduces market noise to at most three implications, each with an owner and the decision it should change.

Ask it: “Run the market pulse” · or run /wf-market-pulse

Step 1 / 7

Listen · Agent

Collect and log signals

Market Research Analyst

/market-pulse

What it does

Gather new signals from market-signals plus public sources; log each with source URL, date and evidence class. Reject anything unsourced.

Reads

Company Context · ICP Library · Market Signals

Writes back to

Market Signals

Produces: signal-log

Each step is taken from the workflow’s machine-readable spec in the v1.1.0 package. A step only completes when its output passes its contract and proof checks.

Shared knowledge

One memory, fifteen governed stores.

Agents never work from blank prompts. They read from and write back to stores with IDs, owners, status, confidence and expiry, so context carries from one step, and one week, to the next.

Human governance

AI expands what marketing can do. People stay accountable for it.

20 human approval gates across the workflows. Nothing decision-stage ships without a named approver.

Human approval gates
Canon changes, spend, sends, publishing and executive decisions stop for a named human approver, with written criteria, before the workflow can continue.
Proof or it doesn't ship
Decision-stage claims must cite an approved, unexpired ProofID from the Proof Vault. The check runs on every step.
Every claim says what it is
Outputs label each statement as hypothesis, AI recommendation, verified evidence, human decision, executed action or measured result.
Governed memory
A guard hook blocks knowledge edits that would put unapproved or unproven claims into canon. Every store records source, owner, status, confidence and expiry.
Truth Pass and Taste Gate
A truth pass and a human taste gate check tone, ethics and accuracy before anything customer-facing is approved.
Insufficient data is an answer
When evidence is thin, agents publish with an ‘insufficient data’ banner rather than padding.
Outcome validation
Narrative Fit, Engagement Quality, Sales Momentum, Launch Learning Rate and Enablement Latency are computed from logged observations, not asserted.

“AI expands the surface area of work; it does not expand judgment.”

Systems, Not Slides, Ch. 10

Get it

Free to use. Open core.

Version 1.1.0. Open core: MIT for code, CC BY-NC-SA 4.0 for content. Commercial resale or bundling needs written permission.