Over the past two years, my AI agency Morningside AI has built AI systems for companies of all sizes - from local martial arts gyms to publicly traded companies and even an NBA team. At the core of every one of our client projects are a handful of powerful AI prompts which, in a few hundred carefully chosen words, can automate hundreds or even thousands of hours of manual work.

Being able to write really good prompts like these is a superpower these days, regardless of what career you're in. However, the tricky thing is that writing these kinds of really good prompts usually takes a lot of time and effort. Realizing this, at Morningside AI it became a major priority for us to figure out how to write these kinds of elite, production-grade prompts for our client projects as quickly as possible.

Over the past 2 years, using the latest in AI prompting research and professional-grade prompting software, we created our own rapid prompt engineering system called the PETER process. In this section, I'm going to be revealing our entire system, including the exact prompt engineering tools and software we use, as well as the process we use at Morningside AI to craft hyper-effective prompts in just minutes.

The first step in the Peter process is to plan the prompt. This involves clearly defining the purpose of the prompt, the inputs and outputs, the source of the IP (Intellectual Property), and the examples that will be used to generate the initial draft.

To start, we define the purpose of the prompt - in this case, to create an email order responder that can classify incoming customer emails and provide an appropriate automated response. We gather the necessary inputs (customer email and days since purchase) and the expected output (a plain text email response).

Next, we identify the source of the IP - in this case, the provided project brief that outlines the response categories and guidelines. We also gather 2-3 high-quality examples of input/output pairs to use in the drafting step.

With the planning complete, we can move on to the drafting phase, where we'll use an AI tool to rapidly generate an initial prompt based on the information we've gathered.

First, we'll use the Relevance AI "Perfect Prompt Generator" tool to rapidly generate an initial draft of the prompt. This tool applies the latest research-backed prompt engineering techniques to create a high-quality first draft.

To use the tool:

1. Sign up for a Relevance AI account and clone the "Perfect Prompt Generator" tool into your account.
2. In the tool, paste in the information from the "Prompt Planning" section, including the purpose, inputs, IP, output format, and example input/output pairs.
3. Click "Run" and the tool will generate a draft prompt for you, applying techniques like role prompting, chain-of-thought, and emotion prompting.

The generated prompt will include:

• A role description for the AI system
• The task it needs to perform
• Context about the use case
• Example input/output pairs
• Notes on requirements and constraints

This draft prompt can then be loaded into the Prompt Metheus prompt engineering IDE for further evaluation and refinement.

In this step, we use the prompt engineering software called Prompt Metheus to thoroughly test and evaluate the prompt we generated in the previous step.

First, we load the different components of the prompt (role, task, context, examples, etc.) into Prompt Metheus. This allows us to easily test the prompt against multiple input scenarios.

We then create a dataset of sample inputs to test the prompt against. This includes things like customer emails with varying details about order status, return requests, and technical support inquiries.

With the prompt and test data set up, we can now run the prompt through Prompt Metheus. This allows us to evaluate how well the prompt is performing - is it correctly classifying the email type? Is the response appropriate and in the right tone?

Prompt Metheus provides helpful feedback, allowing us to identify areas that need refinement. We can toggle different sections of the prompt on and off to see their impact, and even test the prompt using different language models to find the most cost-effective option.

The key is to go through an iterative process of evaluating the prompt's performance, making targeted refinements, and re-testing until we're satisfied the prompt is working as expected across a variety of scenarios. This rigorous testing is what allows us to deliver a high-quality, production-ready prompt.

Once we've completed this evaluation phase, we're ready to move on to the final step of refining the prompt based on the insights gained.

In this refining process, it's an iterative cycle of evaluating and refining the prompt until the desired results are achieved. Here are the key steps:

1. Evaluate the Initial Prompt: Run the initial prompt through the test cases and evaluate the outputs. Look for areas that need improvement, such as formatting, tone, accuracy of classification, and completeness of responses.

2. Create Prompt Variants: Based on the evaluation, create new variants of the prompt by modifying specific sections, such as the role, task, context, or examples. This allows you to test the impact of different changes.

3. Test the Variants: Run the prompt variants through the test cases and compare the outputs. Identify the variants that perform the best and meet the requirements.

4. Refine and Iterate: Make targeted refinements to the top-performing variants, such as adjusting the language, tone, or level of detail. Repeat the testing and evaluation process until you're satisfied with the results.

5. Experiment with Model Settings: Try adjusting the model settings, such as temperature and token limit, to see if you can achieve the same or better performance with a less expensive model.

6. Validate the Final Prompt: Once you've identified the optimal prompt, run it through all the test cases to ensure it consistently delivers the desired outputs.

Throughout this process, it's important to make notes and document the changes you've made, as well as the reasoning behind them. This will help you understand what works best and why, which can inform your future prompt engineering efforts.

The goal is to find the right balance between prompt complexity, performance, and cost, ensuring that the final prompt delivers reliable and efficient results for your use case.

After going through the entire prompt engineering process, the final step is to turn these prompts into usable tools that can be really helpful for you, whether you're an employee, a business owner, or an AI agency.

One way to do this is to go back to Relevance AI and create handy internal tools for yourself or for other people in your company, or even for clients if you're an AI agency.

For example, let's say you want to turn the email autoresponder prompt we just created into an actual tool. You can do this in Relevance AI by:

1. Defining the inputs - in this case, the email content and the number of days since the purchase.
2. Building out the prompt again inside Relevance AI, using the finalized version we worked through.
3. Replicating the prompt using the CL 3.5 HiU model, which we determined was the most cost-effective option.
4. Saving and publishing the tool, so that anyone in your company can access it.

Now, anyone can use this tool by simply inputting the email content and the number of days since purchase, and the tool will generate the appropriate response. This allows you to easily leverage the prompt engineering work you've done and turn it into a practical, reusable tool.

The benefit of this approach is that it allows you to extract more value out of the AI models, without having to pay for the tradeoff of spending a lot of time crafting the perfect prompt. By using the prompt engineering process we covered, you can create high-performing prompts quickly and then turn them into standalone tools that can be used across your organization or shared with clients.

This is the exact approach we use at Morningside AI, and it's a powerful way to maximize the value you get from your prompt engineering efforts. By turning your prompts into tools, you can streamline workflows, automate tasks, and deliver better results to your clients or your own business.