• The Inside Track
  • Posts
  • #4: The Vibe Coding Gap, A True AI Collaborator and a Tip For Better Prompting

#4: The Vibe Coding Gap, A True AI Collaborator and a Tip For Better Prompting

Exploring the future of development, a game-changing tool, and a productivity hack you can implement today

Hey, we’re back!

This week writing to you from Las Vegas where I’m attending Cinemacon to launch a new venture of mine; Mojo Works. Not that much focus on AI here so far.

Anyways, let’s get to it.

Today, we’re testing a new, structured format for the newsletter – three sections

  • explore an idea

  • breakdown a case

  • a practical how-to

Let me know if you like this format!

Crossing Vibe Valley

The AI tooling space isn't just evolving—it's exploding.

The barrier to code up a minimum viable product with tools like Databutton, Lovable, or Replit has never been lower. Ideas previously blocked by technical limitations are now flowing into existence.

This approach—creating software by prompting specialized AI agents—has been aptly named vibe coding. And with OpenAI's latest image model updates solving text embedding and style consistency issues, we're now entering the era of vibe marketing too.

Many software engineers dismiss these approaches because "the applications are unfit for large-scale production." I partially agree, but this critique misses a critical point:

The velocity of evolution.

These AI capabilities aren't static. Just two years ago, one-shot prompting a business application's front-end interface was unthinkable. Today? Entirely possible.

Where will we be in two more years? Five?

The vibe-to-production gap will close faster than most anticipate, with significant downstream implications.

This defensive posture isn't new:

  • Painters criticized photography

  • Typesetters dismissed desktop publishing

  • Web developers scoffed at no-code website builders

In each case, the gap closed, accessibility increased, and industries grew exponentially.

History doesn't repeat, but it rhymes. The AI productivity multiplier is following the same pattern—only faster.

Manus: A True AI Collaborator

I recently gained access to Manus—an AI tool that prompted my fastest-ever paid subscription upgrade.

What makes it stand out? While other AI assistants can generate responses, Manus functions as a true general AI agent that breaks complex tasks into actionable steps and works through them to completion.

For instance:

  • Scope a research report → conduct the research → write the paper → build an interactive website displaying key findings

Of all the tools I've tested, this comes closest to feeling like a genuine work collaborator. The standout features:

  • Continuing to send instructions while it works

  • Opening the computing environment to watch it work in real-time

  • Taking manual control when needed (e.g., logging into accounts)

This is what the shift to AI-native workflows actually looks like in practice.

Meta-Prompting: Let AI Build Its Own Instructions

The results you get from ChatGPT or Claude are only as good as your prompts. But what if the AI could write its own instructions?

I call this meta-prompting, and it's become a big part of my workflow:

  1. Give the AI examples of your writing

  2. Ask it to create a detailed instruction set based on your style

  3. Use this custom prompt template for future work

  4. When editing, simply send both your draft and instructions to the AI

This takes 10 minutes to set up but dramatically improves output quality while cutting editing time in half. I use it for everything from newsletter drafts to research and data analysis.

That’s it for today!