Most of us have spent time wrestling with prompts, trying to get just the right balance of clarity and specificity so the AI will deliver what we want. It can be frustrating—especially when you know the system is capable of great output but your prompt falls short.
Dharmesh Shah, founder of HubSpot and author of the simple.ai newsletter, has been experimenting with prompting techniques for years. In his recent issue, he introduced a method he calls meta prompting—and it might just be the simplest way to unlock better results from AI. Even better, he’s built a free tool, Metaprompt.com, that makes it easy for anyone to start using this technique right away.
In this post, I’ll break down what meta prompting is, why it works, and how Dharmesh’s tool—paired with Agent.ai’s agentic workflows—can help you instantly upgrade your AI results.
At its core, meta prompting is a simple idea: Instead of trying to perfect your prompt in one shot, you ask the AI to rewrite your prompt for you—and you give it permission to ask clarifying questions along the way.
Here’s the process:
Dharmesh’s Metaprompt.com tool makes this process frictionless. It presents optimization questions as checkboxes, so you can refine your prompt in seconds and either copy the improved version or run it directly with GPT-5.
You can use this for any prompt, but it’s even more powerful when paired with the prompts you’re using to build agents in Agent.ai, this method goes from useful to transformational. Instead of just crafting better one-off prompts, you can use meta prompting to design high-quality instructions for your agents. That means repeatable, consistent execution—not just once, but every time.
Here’s an example of where I was using AI to try to solve a problem, but my prompt just wasn’t giving the result I needed. I put it in Metaprompt.com, and voila:
Prompt engineering can feel intimidating. There are countless frameworks, rules, and best practices floating around. For newcomers, it can seem overwhelming. Meta prompting lowers the barrier dramatically.
Even if it’s your first day trying AI, you can use this method to:
Then, here’s where Agent.ai makes the leap:
In other words, meta prompting doesn’t just improve your interaction with AI—it lays the groundwork for building reliable agent-powered workflows.
AI is powerful, but only if we know how to communicate with it. Meta prompting gives everyone—from first-time users to seasoned pros—a simple way to bridge the gap between intention and execution.
With Metaprompt.com, you don’t have to memorize frameworks or study guidebooks. You just need a rough idea, a willingness to refine, and a couple of clicks. And with Agent.ai, you can take those improved prompts and turn them into agents that work for you every day—running processes, handling tasks, and scaling your impact.
That’s the real magic: Not just better prompts, but better outcomes.