Getting Hooked on My Own Kool-Aid: Why This Side Project Is Finally The "One"

I’ve built side projects before. I’ve launched things that were "neat" or "clever," hoping they would stick. But recently, I crossed a threshold that every indie hacker dreams of but rarely hits: I built something that I literally cannot do my job without.

It’s called PromptKelp, and honestly? I’m addicted to it.

The Insanity of Blind Prompting

If you are building AI agents right now, you know the chaos I’m talking about. You have dozens of agents. You are tweaking system instructions constantly. And frankly, the standard workflow is insane.

Imagine writing complex code in Notepad without a compiler, pushing it to production, and just hoping it works. That is what most of us are doing with system prompts. We are writing these intricate sets of instructions blindly, never knowing if we just introduced a massive ambiguity, a subtle contradiction, or a flaw that will break the agent’s logic.

For a while, I tried to solve this by copy-pasting prompts into ChatGPT and asking for feedback. It was slow. It was manual. It was a nightmare.
I needed a development loop that was rigorous. I needed a way to treat prompts like the critical infrastructure they are. So, I built the workflow I needed.

The Power of Dogfooding

PromptKelp isn't just a tool I sell; it’s the tool I open every single morning. It brings sanity to the chaos by offering a dedicated development loop for system prompts:

• Versioning: No more losing that "magic" phrasing from three versions ago.
• AI Evaluation: Before I ever run an agent, PromptKelp analyzes the prompt for me. It catches the ambiguities and contradictions that human eyes glaze over.
• Real-World Analysis: I can import actual interactions—whether from dogfooding or production—and analyze how the agent actually behaved against what I told it to do.
• Context Awareness: Since so many of my agents use Model Context Protocol (MCP), I can pull in tool lists to see if I’m giving the agent unnecessary or confusing tools.

From "Side Project" to "Essential Tool"

The confidence I feel sharing PromptKelp with the world today is different. It doesn't come from market research or hype; it comes from the fact that my own productivity depends on it. I have refined my agents to be sharper, faster, and more reliable because I finally have a lightning-fast workflow that supports continuous improvement.

I stopped guessing if my tweaks were making things better. Now, I can measure it.

If you are tired of the "tweak and pray" method of prompt engineering, you might want to see what happens when you finally have a dev tool built for the job.