sprAI & prAI
Strategy

The System Prompt Revolution: Why AI Apps Should Let Users Write Their Own Instructions

Why exposing system prompts to users isn't just good UX—it's a competitive advantage that turns your most advanced customers into your R&D team.

The System Prompt Revolution: Why AI Apps Should Let Users Write Their Own Instructions

TL;DR: Most AI apps fail because they hide their system prompts from users. The companies that win will be those that let users customize AI behavior, turning advanced customers into their R&D team while serving mainstream users with better defaults.

The Gmail Problem

Pete Koomen recently wrote about Gmail's AI assistant, and it perfectly captures what's wrong with most AI applications today. Gmail's Gemini integration asks you to describe what email you want written, then generates something that sounds nothing like you. It's formal, wordy, and completely misses your actual communication style.

The problem isn't that Gemini is dumb. The problem is that Gmail won't let you teach it how to write like you.

Gmail's AI assistant interface showing generic email generation

Gmail's AI assistant interface showing generic email generation

This is what Koomen calls the "horseless carriage" problem. Early cars looked like horse-drawn carriages because that's what people knew. Today's AI apps look like traditional software because that's what developers know. But AI isn't traditional software.

What We Learned at MadKudu

We made this exact mistake with our research agent. We built a sophisticated AI system that could analyze companies and generate insights, but we kept the system prompt hidden from users. The result? Endless requests to modify behavior.

"Can you make it focus more on competitive analysis?" "Why does it always mention the same metrics?" "Can it write in a different tone for our board presentations?"

Every request meant engineering time, product meetings, and delayed releases. We were building a horseless carriage.

The breakthrough came when we built an MCP that let enterprise customers create their own research agents using our tools. They could start with our system prompt as inspiration, modify it, or write their own from scratch.

Suddenly, our most advanced customers became our R&D team. They showed us what actually mattered in their workflows, what language resonated with their stakeholders, and what insights they really needed. We learned more about our product in three months than we had in the previous year.

"Can you make it focus more on competitive analysis?"
"Why does it always mention the same metrics?"
"Can it write in a different tone for our board presentations?"

These requests stopped being support tickets and became product insights.

The Two-Tier Strategy

Here's what we discovered: there are two types of AI users, and you need to serve both.

Mainstream users want something that works out of the box. They don't want to think about prompts or system instructions. They want to click a button and get results that are "good enough."

Advanced users want control. They understand their domain better than your product team ever will. They have specific workflows, terminology, and requirements that no generic system prompt can capture.

The companies that win will serve both by exposing system prompts to advanced users while providing excellent defaults for everyone else.

Mainstream users vs advanced users

Mainstream users vs advanced users

Why This Matters for Your Business

If you're building AI products, hiding system prompts is a competitive disadvantage for three reasons:

1. You're fighting your users instead of learning from them

Every request to modify AI behavior is a signal about what your users actually need. When you hide the system prompt, you're forcing users to work around your limitations instead of letting them solve their own problems.

2. You're missing your best product insights

Your most advanced users are your best product researchers. They understand the domain, they know what works, and they're willing to invest time in getting the AI to behave correctly. Let them teach you.

3. You're creating unnecessary support burden

Instead of users customizing the AI to their needs, they're asking your support team to do it. This creates a bottleneck and prevents you from scaling.

The Implementation Playbook

Here's how to implement this at your company:

  • Start with defaults that work for 80% of users Build a system prompt that handles the most common use cases well. This is your baseline for mainstream users.

  • Expose the system prompt to advanced users Give power users access to edit the system prompt. Make it easy to see what the current prompt looks like and modify it.

  • Provide templates and examples Don't make users start from scratch. Give them templates for common modifications and examples of what other users have built.

  • Learn from user modifications Track what users are changing in their system prompts. These modifications are gold for improving your default prompt and understanding user needs.

  • Iterate on defaults based on user insights Use what you learn from advanced users to improve the experience for everyone. Their customizations often become your next set of default features.

The Competitive Advantage

Companies that embrace this approach will have a massive competitive advantage. They'll move faster because their users are doing the R&D. They'll build better products because they're learning from real usage patterns. And they'll create more loyal customers because advanced users become invested in the platform.

The future belongs to companies that treat AI as a platform, not a product. The system prompt is the API that makes that possible.

Want to see this in action? Check out how we're building AI-native tools at MadKudu, or explore the MCP protocol that's making this approach possible across the industry.

Related reading: