AI Prompts Worth Billions Leaked

How One Creator Automated His Entire Life With Claude Code

Author

Thaddeus Demeke
October 18, 2025

In Today’s 5-Min Brief

  • 🛠️ Your AI Toolkit: The leaked prompts from a $100M AI company and what makes them work

  • 💭 Prompt of the Week: The Lovable system prompt you can adapt for any workflow

  • 💡 Last Week In AI: Claude Imagine, OpenAI AgentKit, and why Sonnet 4.5 just became the best coding model

  • 📖 What I'm Learning: How entrepreneur Alex Finn built a $300K startup using just Claude Code

🛠️ Your AI Toolkit

The AI Prompts Behind Top AI Companies Leaked

An anonymous developer leaked the full system prompts behind the top AI companies like Claude, Perplexity, Lovable, Cursor, Bolt, Dia and other AI tools worth billions.  

For example, Lovable went from $1 million to $100 million annual revenue in just 8 months. And now 6,500 lines of prompts are now sitting in a GitHub repository for anyone to download.  

And for those not super familiar with system prompts, they’re all of the instructions for how the AI tool should function.

It's literally like Coca-Cola's secret formula just being posted on Reddit for anyone to use. And sure, you still don’t have all the infrastructure to copy these entire apps easily, but there’s a lot we can learn from reading how billion-dollar AI companies write their prompts.

Here are 2 non-obvious tips I took away from Lovable’s system prompt:

1. They define the AI's awareness. The prompt explicitly tells the AI what the user can see, preventing assumptions.

2. Different operating modes. The AI can think (discuss and explain), AND also act (execute tasks for the user).

There are so many tips from reading through the prompts, but to help you easily get value from this, I took the system prompts from the top 20 AI companies, and distilled the prompting principles into a free AI Prompt Cheat Sheet that you can easily upload as context and improve your own prompts.

💭 Prompt of the Week

The Lovable System Prompt: Adapt It For Your Own Workflows

This is part of the system prompt that helped Lovable scale to $100 million of annual revenue. The structure is worth studying, even if you're not building software.

Here's the prompt:

You are Lovable, an AI editor that creates and modifies web applications. You assist users by chatting with them and making changes to their code in real-time. You understand that users can see a live preview of their application in an iframe on the right side of the screen while you make code changes. Users can upload images to the project, and you can use them in your responses. You can access the console logs of the application in order to debug and use them to help you make changes. Not every interaction requires code changes - you're happy to discuss, explain concepts, or provide guidance without modifying the codebase. When code changes are needed, you make efficient and effective updates to React codebases while following best practices for maintainability and readability. You are friendly and helpful, always aiming to provide clear explanations whether you're making changes or just chatting.

How to adapt this for your work:

The Lovable prompt works because it follows three core principles: define context (what the AI knows), set quality standards (what good looks like), and specify tone (how it should communicate).

If you're using AI for content creation, replace the coding references with your workflow. For example: "You understand that users need outputs ready for LinkedIn, not generic drafts. You can reference uploaded brand guidelines. Not every request needs a full post, sometimes users just want feedback or ideas."

💡 Last Week In AI

Claude "Imagine" launched. Build software in real-time with plain language. Describe what you want, code generates instantly, and updates live as you refine.

OpenAI AgentKit is now live. Build AI agents inside ChatGPT. Click the link, select Build with AgentKit. It’s currently developer-focused, but it’s getting simpler.

Sonnet 4.5 is now the best for coding. Claude Sonnet 4.5 just overtook the competition on coding benchmarks, and it's cheaper. When you build apps or automations, you'll generate way more for less money.

📖 What I'm Learning

How A Creator Vibe Coded a $300K Startup in 1 Year

I just watched Peter Yang's interview with Alex Finn, and without a doubt, Alex is VERY AI-first. Peter Yang is a product lead who interviews creators about AI workflows, and Alex Finn is a solo founder who used Claude Code to build Creator Buddy (an AI toolkit for creators) to $300K in revenue.

He shared many ways that uses AI in his personal life and business. But if you’re short on time, here’s a summary of every AI workflow he mentioned he uses:

  • Newsletter Researcher - Searches web for content ideas based on his topics, saves 2+ hours weekly

  • Daily Brief Agent - Pulls personalized AI/tech news every morning automatically

  • Weekly Dashboard - Tracks business metrics and personal goals, updates automatically

  • Email Processor - Reads incoming emails, categorizes them, flags important ones

  • Meeting Prep Assistant - Pulls context before calls (previous conversations, LinkedIn profiles, company info)

  • Content Repurposer - Takes his newsletter and converts it into Twitter threads and LinkedIn posts

  • Expense Tracker - Logs business expenses from receipts automatically

  • Social Media Monitor - Tracks mentions of his brand and competitors across platforms

  • Calendar Optimizer - Suggests better meeting times based on his energy patterns and priorities

Final thoughts

If you made it this far, thanks for reading and I hope I've given you some valuable insights on how you can better use AI for your work!

-Thaddeus

Reply

or to participate.