Tech Surprises

Over a year after its February 2025 launch, Claude Code is being used beyond software engineering. Anthropic says more than half of public API tool calls now come from outside the developer world (Source: Measuring AI agent autonomy in practice). I am part of that “other 50%”, using Claude Code for market research, copywriting, document creation, and on brand visual storytelling. And I am loving the results.

Here’s a short FAQ with the why and the how I hacked Claude Code to do magic to my marketing workflow.

First, why Claude Code and not the chat version (Claude.ai)?

Claude.ai is great for focused conversations. Claude Code enables an entirely new workflow because it runs inside a folder on your laptop that can be full of relevant artifacts.

Every brief, deck, template, research document, and half finished draft in that folder is available for Claude Code to search, read, and use as needed. It can work across those artifacts and build something new from them.

Instead of opening a new chat, uploading a handful of files, and restating the business context, I open the workspace and start from where the work already lives.

Second, how to set up Claude Code for a marketer user?

Here’s a README file 😉

Step 1: Build a CLAUDE.md file

Describe who you are, what your company does, how you write, what a good output looks like, and what Claude should never do.

Claude Code loads those project instructions at the beginning of each session. No 15 minute warmup to explain your jobs to be done, your company context, your quality bar, and your writing preferences all over again.

You open the workspace and it is ready to work.

Step 2: Give Claude Code a folder on your laptop instead of a question

In 2023, many of us used AI like a search engine: One question. One answer. Move on.

In 2026, the better workflow is giving an agent a well organized set of source materials and a meaningful task. Strategic narratives. Customer research. EBC decks. Analyst notes. Messaging drafts. Product requirements. Sales data. Existing templates.

Claude Code can search, read, and synthesize the relevant materials as it works through a new project.

That mirrors how marketers actually work.

• We switch context constantly, from product roadmaps and PRDs to sales forecasts and campaign copy.
• We synthesize incomplete inputs, deal with messy starting points, and turn them into something useful.

For example, Claude Code can help you:

1. Build a customer facing, 30,000 foot thematic roadmap from product manager inputs, feature descriptions, and strategy documents.
2. Turn messaging documents into frameworks that give creative and campaign teams a stronger starting point.
3. Draft cross functional processes from scattered inputs across Product Marketing, Sales, Product Operations, Sales Operations, Finance, and Sales Enablement.

Step 3: Ask Claude Code to build your marketing artifact for you

Give it a source deck, a template, a spreadsheet, and a clear brief. Then ask it to create a new presentation in the same style.

It can work across the right files, follow the intended structure, match the template, write in your voice, and get you 80% of the way to a finished artifact.

The outcome takes longer than a quick chatbot response. Most of my projects take several minutes. But with the right brief, context, and a few hacks I will share below, you get a real artifact you can open, review, and share with your team after a quick polish.

Three Hacks That Made Claude Code Work for My Marketing Workflow

Hack 1: Turn off the permission prompts – only in the right workspace

Claude Code asks for approval when it is about to take meaningful actions, such as editing files, generating new ones, running commands, or reaching outside the workspace.

That is a good default. But when I am running a contained, end to end workflow in a dedicated project folder, repeated prompts can break the flow.

Fix it with one flag:

claude –permission-mode bypassPermissions

⚠️ Warning: Use bypass permissions only in a dedicated, disposable folder with no sensitive files, no shared drives, no untrusted code, and no untrusted documents. Do not run it from your home folder or an unfamiliar repository.

Hack 2: The CLI looks scary. Yet once you are inside it, it feels like a chat window with access to your files.

The terminal is the first thing that might put marketers off. For those of us who joined marketing from engineering, it may bring the scaries of getting stuck in VIM forever (q!).

But once you’re inside a terminal window, you’re just chatting. You type what you need, it responds, it works on the files in your folder.

If you’ve used Claude in a browser, the experience is almost identical, just without the upload button, because it doesn’t need one. Your files are already there.

Pro tip: you can use the chatbot of your choice, like Claude.ai to help you refine your brief for Claude Code to get better results faster. That’s the only copy-past you’ll need 😉

Hack 3: Experiment often to optimize over time

Anthropic found that experienced Claude Code users approve more actions automatically, while also interrupting more often when they need to redirect the work.

That matches my experience. Give Claude room to work. Watch the output. Step in when the brief needs to change, the quality slips, or the task needs human judgment.

My first artifacts were good. After a few prompt refinements, a stronger CLAUDE.md file, and better source materials, they started to look like something I could plausibly have created myself.

Who’s using Claude Code, ChatGPT Agents, Gemini CLI, or another agentic tool for non-coding work? What’s the most valuable workflow you’ve discovered?

Tags: ai, artificial-intelligence, chatgpt, llm, technology