Disclaimer: I am not affiliated with, nor paid by, Antrophic or any other AI company.

For the past 16 months, I have been using Claude AI. I would not even know which model I am on, since Anthropic keeps updating it. I do have a Pro subscription, and if you ask me, it is the best deal out there right now for any starting entrepreneur, freelancer, vibe-coder or tech-savvy person who wants to build something cool.

I personally use it to run my YouTube Channel, which has helped me grow it to over 200 subscribers, 48.000 views and 500.000 impressions in a little under 16 months with minimum effort. You can read more about how I set this up in my previous post.

In this post I want to share how I use prompts to get the most out of Claude. I hope it helps.

How did I get here?

About a year ago I attended a workshop in Athens, Greece, during a Business School conference themed around Marketing & Communication. A professor in Mathematics showed something that silenced the entire room. In roughly 40 minutes, Claude AI had produced work that would normally take a team of marketing professionals several days.

The only input given to Claude was a so-called Master Prompt and the audience profile, such as our job titles and the schools we represented. Lastly it got the context of the conference and a simple instruction to impress us.

The presenter had essentially created a single master prompt that followed this structure:

  • He gave Claude a clear outline of the challenge
  • He provided the core requirements and relevant data, including the audience profile
  • He provided several safeguards to ensure compliance: no hallucinations, no long complicated texts, nothing created without user verification
  • He included an agent-creation instruction, which allowed Claude to guide the user in creating new chats and specialised agents
  • He set the general rules: when in doubt, ask the user, and always explain the reasoning behind each action taken

From there, Claude took over. It built the plan and instructed the user to create several specialised agents to execute the tasks it needed done.

The final result was a fully functional mock-website, a program curriculum for a new fictional MBA programme (Essentials of Business Leadership, that sounds very MBA-like, right?) and a brand guide + strategy outline for how to market the new programme to attract students.

You can imagine the shock of the audience, which consisted mostly of marketing professionals, communications specialist and the occasional Dean or Programme Manager.

I loved it.

So, how does this work?

When the previously mentioned structured is followed and combined into one Master Prompt, you give Claude a clear framework to work within. It does what you want it to do, and it helps you to create the several specialised agents needed to serve the project alongside it.

In order to create a new agent, Claude will instruct you to open a new chat with a specific prompt. You work through that chat, follow the steps, and report the results back to the main conversation. It sounds more complex than it is. Once you have tried it for yourself, it clicks immediately.

It looks a bit like this:

Yes, this is totally AI Generated, but it’s clear!

A Simple Example

Say you want Claude to help you launch a freelance web design service. A Master Prompt for that might look like this:

Goal: Help me launch a freelance web design service targeting small businesses in the Netherlands.

Requirements: I work alone, have a budget of zero, and want my first client within 60 days.

Safeguards: Only suggest free or already available tools. Do not recommend paid ads or anything that requires a team.

Agents needed: A content agent to write my website copy, and an outreach agent to draft my first cold emails.

Rules: Keep everything simple, practical and actionable. No fluff, no jargon. If something requires more than one hour of my time, flag it.

Claude will read this, build a plan around it, and tell you exactly which agents to spin up and in what order. You follow the steps, report back what each agent produced, and the AI stitches it all together. The whole thing runs in one afternoon.

This is a very short prompt, you can imagine that the more complex your request, the more input you’ll be giving to the AI. And that also determines the quality of the work. The more guidance and info you put into it, the better it can understand you and serve your needs. Just like real colleagues, you can’t expect them to read your mind. You need to give guidance and input for them to be successful.

Closing remarks

Agentic AI and well-structured prompts are not here to replace workers or do all the work for you. Managing different chat agents will feel like directing a team of (very fast and very smart) interns. You will need to provide the guidance, the vision and the direction. But within minutes you will have workable results and first drafts to react to.

Entire branding documents ready to skim and correct, not build from scratch. Mock-up pages you can hand off to a designer, or build yourself. That is the shift.

For a one-person operation or a freelancer, the time you save is the real value of AI. Not the output or the content itself, but the hours you get back to spend on the work only you can do.

Good luck with your next project.