Claude Code for Marketers: What the Same Way of Working Looks Like in Marketing

Sit behind an engineer for ten minutes and the surprising part is how little code gets typed.

She writes two sentences. The export should include the invoice date. Old exports must keep working. Then she waits while the agent reads the project: the files, the conventions, the tests somebody wrote two years ago and forgot about.

The agent proposes a change and shows exactly which lines it wants to touch. She reads it, pushes back on one part, and approves the rest. The tests run. It ships.

Now walk across the office to marketing. Someone is copying an email out of a chat window, pasting it into the email tool, and fixing the line breaks by hand. Both people are using AI. Only one of them has a way of working.

What actually changed for engineers

The headline version is “AI writes code now”. That is true and it misses the point.

Text generation alone would have given engineers a faster way to produce code nobody trusted. What tools like Claude Code, Anthropic’s coding agent, gave them was a loop:

1. Describe the outcome. Not the keystrokes. The result.
2. The agent reads the real context. The actual project, not a summary of it.
3. The agent proposes a change. Specific and bounded.
4. A person reviews the diff. Line by line, with the old version beside the new.
5. Only then does it ship.

Every step matters, but the loop holds together because of two things engineers had long before agents arrived. They had a repository, one place where the truth lives. And they had review, a habit of looking at the change before it goes out.

The agent did not invent the discipline. It plugged into it.

Why marketing never had the loop

Marketing has no repository.

Ask where the brand voice lives and the honest answer is “in somebody’s head, and partly in a deck from last spring”. The current offer is in a chat thread. The launch dates are in a spreadsheet that three people have copies of. The closest thing to version control is a file called final_v3_REAL.

So when a marketer asks an assistant for a launch email, the assistant has nothing to read. It writes from a blank page and a paragraph of pasted instructions. The draft comes back sounding like marketing in general, not like your company in particular.

Then comes the second gap. There is nowhere for the output to land. The draft leaves the chat by clipboard and travels through the email tool, the design tool and the scheduler, losing a little formatting at each border crossing. That journey has its own post: ChatGPT wrote my campaign, now what.

No context going in. No review on the way out. No single place to ship from. That is not a loop. It is a very fast typewriter.

The same loop, translated

Here is the engineer’s way of working next to its marketing equivalent.

Each row is worth a closer look, because each one replaces a habit marketing teams have put up with for years.

Context: the repository you never had

In Marqeable, the things an agent needs to read are kept in one place: your brand voice, your brand look with its colors and assets, your offer, and your marketing calendar.

The calendar is the part people underestimate. It holds events, launches, offers and important dates. It spots events with no campaign linked, and new campaigns are dated around what is already scheduled. An agent that can read the calendar stops proposing a launch email on the same morning as your webinar reminder.

The diff: suggestions beside the passage

Engineers do not review code by reading two whole files and guessing what changed. They read the diff.

The marketing version is the review. An agent review covers language, brand voice, compliance and strategy, and its suggestions appear beside the passage they refer to. You accept, reject or edit each one. Teammates comment on a highlighted passage and reply in a thread.

Compare that with the usual alternative: a rewritten draft arrives, and you spend ten minutes working out what is different. One of those is review. The other is a spot-the-difference puzzle.

The merge: marking it Ready

In engineering, nothing reaches the main branch without approval. In Marqeable, nothing publishes until you mark it Ready.

There is an earlier gate too. Copy is approved before any image is made. Images are decided after the words, and at that gate you pick from your library, describe an image, or ship without one. Engineers would recognize the principle: settle the interface before you build on top of it.

The tests: review and dry run

Marketing has no compiler, which is a shame, because a compiler would have caught a lot of subject lines.

What it does have is two checks. The agent review is the test that runs on the words. The dry run is the test that runs on an automation before launch, so you see what would happen before it happens to a real contact. And after launch, an automation can be paused and edited.

The deploy: send and publish

This is where most AI writing tools end and where the loop has to keep going. Email and text send from Marqeable. LinkedIn posts publish to the company page. Blog posts publish to WordPress. Landing pages are hosted by Marqeable with a lead form, and submissions become leads.

Two limits, stated plainly. Founder-voice LinkedIn posts can be drafted but are not posted to a personal profile. X posts are drafted and sized, and you publish X posts yourself for now.

Where the comparison breaks

An honest analogy admits its weak joint. This one has a big one: engineers can roll back a bad deploy. Nobody can unsend an email.

That changes where the safety has to sit. In software, some of it can live after the release. In marketing, all of it has to live before. It is the reason the approval step is not optional decoration, and the reason an assistant connected to Marqeable can draft but never launch. The longer argument is in the draft-only rule and human in the loop AI marketing.

You can literally use Claude Code

So far this has been an analogy. It is also available as a fact.

Claude Code is Anthropic’s product. Marqeable is a separate company, and it is not made by, endorsed by, or affiliated with Anthropic or OpenAI. What Marqeable offers is a connector that those assistants can use. It works with Claude Code, with Claude in Claude.ai, Claude Desktop and Cowork, with ChatGPT in developer mode, and with Cursor and Codex. Each person signs in as themselves.

Once connected, the assistant can read your brand, voice, offer, calendar, campaigns, content, automations and image library.

It can write only drafts: campaign drafts, content drafts, automation drafts, brief edits, reviewer notes, change requests, image approval and marketing calendar entries.

It cannot send, launch or delete anything. A person reviews and launches in Marqeable.

This post is about the idea, not the installation. For the step-by-step, see the Claude Code setup guide and the Claude Code integration page. If your team prefers another assistant, there are pages for Claude and ChatGPT too.

One loop, start to finish

Here is an example. Say Northwind, a made-up software company, is opening a 30-day pilot.

The marketer describes the outcome: a pilot launch for finance teams, opening on the 8th. The agent reads Northwind’s voice, offer and calendar, and notices a webinar already scheduled that week. It proposes a campaign with an email, a text message, a LinkedIn post and a landing page, dated around the webinar.

The review puts three suggestions beside the email. The marketer accepts two and rewrites the third. A teammate questions the headline in a comment, and they settle it in the thread. The copy is approved, then an image is picked from the library.

The marketer marks the pieces Ready and launches. The email and text go out, the post goes to the company page, and the landing page starts collecting leads.

A week later, replies are sitting in one inbox, and the revenue from the first pilot that turns into a paid account is traced to the exact message that started it.

Nobody copied anything into anything. That is the whole trick.

Frequently asked questions

What is Claude Code?

Claude Code is a coding agent made by Anthropic. An engineer describes an outcome, the agent reads the project, proposes changes, and the engineer reviews them before anything ships. It is Anthropic’s product, not ours.

Can marketers actually use Claude Code with Marqeable?

Yes. Marqeable has a connector that works with Claude Code, Claude, ChatGPT in developer mode, Cursor and Codex. Once connected, the assistant can read your brand, voice, offer, calendar, campaigns, content, automations and image library, and write drafts into Marqeable for a person to review.

Can the assistant send a campaign on its own?

No. Through the connector an assistant writes only drafts: campaign drafts, content drafts, automation drafts, brief edits, reviewer notes, change requests, image approval and marketing calendar entries. It cannot send, launch or delete anything. A person reviews and launches in Marqeable.

Is Marqeable affiliated with Anthropic or OpenAI?

No. Marqeable is an independent company. Claude and Claude Code are Anthropic products, and ChatGPT and Codex are OpenAI products. Marqeable is not made by, endorsed by, or affiliated with either company. It connects to their assistants through its own connector.

The bottom line

Engineers did not get faster because an agent learned to type. They got faster because the agent joined a loop that already had context at one end and review in the middle.

Marketing can have the same loop once it has the same parts: one place for the context, suggestions beside the passage, a Ready mark, and a send button that only a person presses. Agents draft. People approve. The work ships.

And the point of shipping is not more content. It is more leads generated, and more of them won as customers.

See the whole line: this post is part of a series that starts with what Marqeable is and continues with how marketers use Marqeable. Marqeable’s agents draft and review your content, build your campaigns, run automations across text and email, answer visitors with website chat, and tie revenue to the exact message with attribution.

Marqeable runs your campaigns, answers every visitor, text, and email in seconds, and turns them into booked jobs and meetings - even at 9pm on a Saturday. We’re in private beta with a small early cohort. Get early access