Grace Leung AI Marketing Team Builder

Last updated: 31 May 2026

Build a fully functional AI marketing team inside Claude Code — with dedicated agents, reusable skills, and a shared task board — that researches, writes, analyzes, and designs campaign deliverables autonomously.

// TL;DR

The Grace Leung AI Marketing Team Builder is a framework for building a persistent, multi-agent marketing team inside Claude Code — with dedicated agents, reusable skills, and a shared task board that researches, writes, analyzes, and designs campaign deliverables autonomously. Use it when a single Claude conversation can no longer handle the breadth of your marketing function, or when you want to systematize recurring workflows instead of relying on one-off prompts. You build it in four layers: map your marketing functions, convert each repeatable task into a skill, group skills into non-overlapping agent roles, then connect everything as a coordinated team.

Framework

// When should you build an AI marketing team in Claude Code?

Use this skill when you need to systematise recurring marketing workflows into an AI-powered team, or when a single Claude conversation is no longer focused enough to handle the breadth of your marketing function. Ideal for marketers ready to move beyond one-off prompts into a persistent, structured agent system.

// What do you need before building your AI marketing team?

Brand context filesrequired
Brand voice guide, style guide, product offerings, and marketing strategy documents to pre-equip agents with brand knowledge.
Branded templatesrequired
Existing on-brand templates (decks, social creatives, landing pages, etc.) that agents will study and replicate in style.
Marketing function maprequired
A list of the repeatable marketing tasks you actually do every week — the raw material for defining skills and agents.
Project folder structurerequired
A root project folder organised into system folders (context, SOPs, templates) and working folders (ads, pages, presentations).
External API keys
API keys for any MCP-connected tools (e.g. image generation services) you plan to integrate into skills.
Task management board
A shared Notion Kanban board (or equivalent) with task title, details, priority, and status fields for team-AI collaboration.

// What principles make an AI marketing team actually work?

Map → Skill → Agent → Team

Every AI marketing team is built in four sequential layers: map your marketing functions, convert each repeatable into a skill, group skills into non-overlapping agent roles, then connect agents and skills as a coordinated team. Skipping layers produces unfocused, unreliable output.

System Folders vs. Working Folders

Separate reusable infrastructure (context, SOPs, templates) from output destinations (ads, pages, presentations). System folders pre-equip agents; working folders capture what agents produce. Mixing them breaks agent orientation.

Reference-Based Method

For any skill that must match a visual or structural standard, give Claude an existing branded example, have it generate a detailed analysis report of that example, then build the skill from the analysis — not from a blank prompt. This anchors output quality to real brand standards.

Style Library as Visual Inspiration, Not Exact Copy

When building creative skills, provide a folder of on-brand templates as a style library. Claude studies the vibe and patterns, not a pixel-perfect copy — this is what produces design output that feels branded rather than templated.

Skills are the shared playbook; Agents are the specialists

Skills are reusable workflow modules any agent can call. Agents are dedicated, focused team members assigned a coherent set of skills matching their role. Piling too many skills into one conversation without agent separation causes Claude to lose focus — just like one person trying to be writer, analyst, and designer all at once.

Agent Routing Rules in CLAUDE.md

Not every task requires a full agent — some executional steps are handled by skills alone. Explicitly encoding routing rules in the CLAUDE.md file tells Claude when to delegate to a sub-agent versus when to call a skill directly, making the system reliably self-directing on complex campaigns.

Context Loading Makes the Whole Difference

Loading brand context files into a dedicated context folder before building any skills or agents is non-negotiable. Agents pre-equipped with brand voice, strategy, and offerings produce fundamentally better work than agents prompted cold.

// How do you build an AI marketing team step by step?

1
Set up your project folder structure
Create a root brand folder. Inside your marketing subfolder, create system folders (context, SOPs, templates) and working folders (ads, pages, presentations). Think of this as how a real business organises their work — the folder architecture is the team's physical office.
2
Load brand context into the context folder
Add your brand voice guide, style guide, product offerings, and marketing strategy before anything else. This is what pre-equips every agent you will build. Do not skip this step — it makes the whole difference in output quality.
3
Initialise the project by creating the CLAUDE.md file
Open the project folder in Claude Code and prompt Claude to scan your folder and draft the CLAUDE.md file for you. This is the custom instructions file for your entire project. Treat it as a living document — you will keep updating it, especially after adding agents.
4
Install official skills via the plugin command
Use the /plugin slash command to add Anthropic's official skills GitHub repository marketplace, then search for and install the document skills pack. Confirm installation by checking that all official skills appear under the slash command menu.
5
Map your marketing function
List every repeatable marketing task you actually do each week. This is your raw material. Each item on this list is a candidate skill. Group naturally related tasks — these groupings will later define your agent roles.
6
Build skills using the Reference-Based Method
For each repeatable workflow: (a) place the relevant branded template or visual reference in the templates folder, (b) prompt Claude to analyse it and produce a detailed analysis report, (c) ask Claude to extend the closest official skill and generate a new custom skill based on the template and analysis report. Claude will create a Skills folder with the new skill's assets inside. Aim for one skill per workflow.
7
Set up MCP connections for skills requiring external tools
For any skill that calls an external service (e.g. image generation), create a .mcp.json file in your project root folder. This file declares which MCP servers Claude can connect to for this project, along with the required API keys. Restart Claude Code and verify connections via the /mcp slash command before building the dependent skill.
8
Build your Style Library for creative skills
Under the templates folder, create a sub-folder (e.g. social-creatives) containing on-brand visual templates and a style guide that describes when each style is appropriate. These serve as visual inspiration — not exact copies — so Claude learns the vibe and applies it generatively.
9
Create dedicated agents using the /agents command
Type /agents in the Claude Code terminal, click 'Create New Agent', select 'for this project only', and let Claude generate the Agent MD file. Define: the agent's name, which skills it should use, its tools, default model, colour, and memory location. Each agent gets its own markdown file in an Agents folder defining its role, skills, and core responsibilities. Keep agent roles non-overlapping — a data analyst thinks in numbers and patterns; a content creator thinks in stories and headlines.
10
Assign skill sets to each agent role
Match skills to agents by cognitive type, not convenience. Suggested roles: Data Analyst (campaign reporting, data visualisation), Content Creator (blog writing, keyword research, lead magnet creation), Market Researcher (audience and trend research), Creative Designer (social creatives, branded decks), Campaign Strategist (campaign briefs, landing pages). Each agent should have a coherent, focused playbook.
11
Update CLAUDE.md with agent routing rules
Before running any complex task, add explicit routing rules to CLAUDE.md telling Claude when to delegate to a sub-agent versus call a skill directly. Rule of thumb: tasks requiring synthesis (research, strategy) benefit from agents; executional, straightforward tasks (single content formats) can be handled by skills alone. These rules make the system self-directing on complex campaigns.
12
Test the team on a full campaign task
Give the team a complex, multi-deliverable brief (e.g. 'Launch a [season/theme] campaign — produce market research, campaign brief, social posts, landing page, and ad creatives'). Observe how Claude routes between agents and skills. Expect roughly 10 minutes for a full package. Assess whether all deliverables are thematically connected — coherent campaign narrative is the quality signal.
13
Connect a shared Notion task board for team-AI collaboration
Set up a Kanban board with columns for To-Do and Complete, and fields for task title, details, and priority. Prompt Claude to scan pending tasks, assign the appropriate agent, execute by priority, and update the task status to Complete with output file paths included. This is the bridge between your human teammates and your AI agents.
14
Activate remote control for mobile access
Type /remote-control in the Claude Code chat to generate a remote control link. Open this link on your mobile device to connect to your local running session and send tasks to your AI team from anywhere. Do not share this link — it grants full control of your session. If context fills up, type 'clear conversation' to reset. Disconnect via the three-dot menu and archive when done. All work saves locally.

// What does the AI marketing team builder look like in practice?

A travel brand wants to launch a seasonal campaign and needs a full marketing package produced consistently and on-brand.

Apply the 4-step team design (map → skill → agent → team) to build five agents (researcher, strategist, content creator, designer, data analyst) each equipped with role-matched skills. Load brand context upfront. Give the team a single complex brief ('Launch a [season] campaign — research, brief, social posts, landing page, creatives'). Claude routes tasks to the right agents, keeps all deliverables thematically connected, and returns the full package in roughly 10 minutes.

A marketing team wants their AI agents to handle tasks collaboratively alongside human teammates without requiring constant manual prompting.

Set up a Notion Kanban board where team members drop tasks with priority and details. Prompt Claude periodically (or via mobile remote control) to scan pending tasks, assign the correct agent based on task type, execute in priority order, and mark tasks Complete with output paths. Human team members see results update in real time; AI agents operate as a 24/7 execution layer.

A marketer needs a presentation skill that reliably produces on-brand slide decks without manual formatting correction.

Use the Reference-Based Method: place an existing branded deck template in the templates folder, prompt Claude to analyse it and produce a detailed analysis report capturing layout rules, colour usage, font hierarchy, and slide types, then ask Claude to extend the official PowerPoint skill into a new branded deck skill using the template and report. The resulting skill produces decks that are 90%+ aligned with brand standards on first run.

// What mistakes should you avoid when building an AI marketing team?

Skipping the context loading step — agents built without brand voice guides, style guides, and strategy documents produce generic output that misses brand standards entirely.
Piling too many skills into one agent or one conversation — this causes Claude to lose focus, the same way one person cannot be writer, analyst, and designer simultaneously. Keep agent roles non-overlapping.
Using the style library as an exact copy directive rather than visual inspiration — this produces templated, rigid output instead of creatively on-brand work.
Forgetting to update CLAUDE.md with agent routing rules before running complex tasks — without explicit routing logic, Claude cannot reliably decide when to use an agent versus a skill, degrading system performance on multi-step campaigns.
Treating CLAUDE.md as a one-off setup file — it must be updated iteratively as you add new agents, skills, and routing rules. It is a living document, not a one-time configuration.
Sharing the remote control link — it grants full control of your local Claude Code session to anyone who has it.
Ignoring context limits during remote sessions — if context fills, the session degrades. Use 'clear conversation' to reset rather than continuing in an overloaded context window.

// What are the key terms in the AI marketing team builder?

Skill: A reusable workflow module scoped to one specific marketing task (e.g. branded deck creation, social creative design). Skills are the shared playbook that agents can call. One skill per workflow is the target.
Agent (Sub-Agent): A dedicated, focused AI team member with its own defined role, assigned skill set, tools, and memory. Agents are used when a task requires synthesis, judgment, or orchestration across multiple skills. Defined by an Agent MD file in the Agents folder.
CLAUDE.md: The master custom instructions file for the entire Claude Code project. It tells Claude who is on the team, what each agent does, and — critically — when to delegate to an agent versus call a skill directly. Must be updated iteratively as the system grows.
Reference-Based Method: A skill-building approach where you provide Claude with an existing branded example, have it generate a detailed analysis report of that example's patterns and rules, then build the skill from the analysis. Produces significantly higher fidelity output than blank-prompt skill creation.
Style Library: A folder of on-brand visual templates and an accompanying style guide placed under the templates folder. Serves as visual inspiration for creative skills — Claude studies the vibe and patterns, not a pixel-perfect copy.
System Folders: Reusable infrastructure folders (context, SOPs, templates) that pre-equip agents with brand knowledge and working patterns. Distinct from working folders where agent outputs are saved.
Working Folders: Output destination folders (e.g. ads, pages, presentations) where agents save their produced deliverables. Kept separate from system folders to maintain clear agent orientation.
MCP Connection (.mcp.json): A configuration file placed in the project root that declares which external MCP servers Claude can connect to for the project, enabling skills to call external tools such as image generation APIs.
Agent Routing Rules: Explicit rules written into CLAUDE.md that define when Claude should delegate a task to a sub-agent versus handle it with a skill call alone. The rule of thumb: synthesis tasks (research, strategy) → agent; executional tasks (single content formats) → skill.
Remote Control: A Claude Code feature activated via /remote-control that generates a link allowing a mobile device to connect to a locally running Claude Code session, enabling task dispatch to AI agents from anywhere.
Context Folder: A system folder containing brand voice guide, style guide, product offerings, and marketing strategy — loaded before any skill or agent is built to pre-equip the entire team with brand knowledge.

// FREQUENTLY ASKED QUESTIONS

What is an AI marketing team in Claude Code?

An AI marketing team in Claude Code is a persistent, multi-agent system where dedicated agents (researcher, strategist, content creator, designer, analyst) share reusable skills and a task board to produce campaign deliverables autonomously. Instead of prompting one conversation repeatedly, you build a structured team that researches, writes, analyzes, and designs on-brand work with minimal supervision.

What is the difference between a skill and an agent in Claude Code?

A skill is a reusable workflow module scoped to one specific task, like branded deck creation or social creative design — it's the shared playbook any agent can call. An agent is a dedicated AI team member with a defined role, assigned skill set, tools, and memory, used when a task requires synthesis or orchestration. Aim for one skill per workflow and keep agent roles non-overlapping.

How do I build an AI marketing team in Claude Code?

Build it in four sequential layers: map your repeatable marketing tasks, convert each into a skill using the Reference-Based Method, group skills into non-overlapping agent roles via the /agents command, then connect agents and skills as a coordinated team. Load brand context first, create a CLAUDE.md with routing rules, and test on a full campaign brief before going live.

How do I make my AI agents produce on-brand content?

Load brand context files — voice guide, style guide, product offerings, and strategy — into a dedicated context folder before building any skills or agents. Then use the Reference-Based Method: give Claude an existing branded example, have it generate a detailed analysis report, and build the skill from that report. For visuals, add a style library of templates as inspiration, not exact copies.

How does an AI marketing team compare to using one-off ChatGPT prompts?

A Claude Code AI marketing team is persistent, structured, and brand-equipped, whereas one-off prompts start cold every time and lose focus as scope grows. The team separates work across specialized agents so no single conversation tries to be writer, analyst, and designer at once. It also integrates a shared task board and reusable skills, making recurring workflows systematic rather than repetitive.

When should I build an AI marketing team instead of just prompting Claude?

Build an AI marketing team when you have recurring weekly marketing tasks worth systematizing, or when a single Claude conversation is no longer focused enough to handle the breadth of your function. If you're producing multi-deliverable campaigns — research, briefs, social posts, landing pages, creatives — a persistent agent system delivers far more consistent, on-brand results than ad-hoc prompting.

What results can I expect from an AI marketing team in Claude Code?

Expect a full multi-deliverable campaign package — market research, campaign brief, social posts, landing page, and ad creatives — produced in roughly 10 minutes, with all deliverables thematically connected. Skills built with the Reference-Based Method typically produce output 90%+ aligned with brand standards on first run. The coherent campaign narrative across deliverables is the key quality signal.

What is the Reference-Based Method for building skills?

The Reference-Based Method is a skill-building approach where you give Claude an existing branded example, have it generate a detailed analysis report of that example's patterns and rules, then build the skill from the analysis rather than a blank prompt. This anchors output quality to real brand standards and produces significantly higher-fidelity results than cold skill creation.

How do I assign skills to the right agent?

Match skills to agents by cognitive type, not convenience. A data analyst thinks in numbers and patterns; a content creator thinks in stories and headlines. Suggested roles include Data Analyst, Content Creator, Market Researcher, Creative Designer, and Campaign Strategist. Each agent should have a coherent, focused playbook — piling too many skills into one role causes Claude to lose focus.

Can I control my AI marketing team from my phone?

Yes. Type /remote-control in Claude Code to generate a link that connects your mobile device to your locally running session, letting you dispatch tasks from anywhere. Never share this link — it grants full control of your session. If context fills up, type 'clear conversation' to reset, and disconnect via the three-dot menu when done. All work saves locally.

What is CLAUDE.md and why does it matter?

CLAUDE.md is the master custom instructions file for your entire Claude Code project. It tells Claude who is on the team, what each agent does, and — critically — when to delegate to an agent versus call a skill directly. It must be updated iteratively as you add agents, skills, and routing rules. Treat it as a living document, not a one-time setup.

// GET THIS SKILL — FREE