Set it up before you scale it: a workspace that knows your brand, buyers and data rules, a shared library of prompts for real tasks, and a person who reviews anything before it goes out. Connect tools only with the permissions they need, and read each tool's data settings yourself.
What AI does well
Drafting, research, summarising calls, repurposing one piece into ten, and now the setup work too: email lists, tracking, landing pages. Connected through MCP, a model like Claude can carry out tasks inside your marketing tools from a plain-English instruction.
What still needs a person
Claims about customers or competitors, anything with a number in it, approvals, and decisions about customer data. AI drafts. A person signs.
The rules worth writing down
Keep a list of every tool connected to an AI model and what each connection can do. Give each one the narrowest permissions that work. Log what it does. Name the person accountable for its output. These are the same rules a security team would ask of any supplier.
Murfin sets up AI workspaces and trains marketing teams to use them on their own work.