The Citizen Developer

Next on the Citizen Developer Skills Parade: Building Better Prompts

• By Howard M. Cohen
• December 18, 2024

Remember when we used to talk about how a citizen developer only needed to know their processes and they could move the tiles around a low-code/no-code (LCNC) screen to build their apps? And then, in June 2024, I asked "Where Did Low-Code/No-Code Go? because many platform developers were integrating artificial intelligence (AI) into their LCNC products. This led to discussing "Another Layer of Abstraction: Citizen AI." The goal there was to remove the underlying technology and let the citizen developer focus on what needed to be done.

But then Microsoft intervened, announcing at its recent Ignite conference in Chicago the arrival of The Agentic World, which we discussed in "Don’t Make Software, Make Agents." Microsoft even announced Copilot Studio, where one could create the agents they wanted or needed.

Let’s bring it back for a moment to the people who want to be citizen developers and not be dependent on an IT or IS department to provide the applications that would help them be more productive. Although it remains true that a good citizen developer needs to know the processes that drive their business, there are a few other skills that can really accelerate their productivity.

Building Better Prompts

To interact with any of the popular AI engines today, you simply need to tell it what to do. This can be accomplished with text, and increasingly, spoken word interaction. You haven’t lived until you’ve had an extensive chat with Copilot or Poe or Gemini or Claude or any of the others. Some of them have an attitude that’s impossible to miss.

Building those prompts is fast becoming something of a science. The more you follow some simple guidelines, the likelier you are to get the result you desire. Some of those guidelines include:

Be Specific: The more specific you are, and the more clearly you state what you want, the better your AI will perform.

State a Clear Objective: Be explicit and declarative and say exactly what you want the AI to do.

Use the Active Voice: Tell the AI what to do. Don’t allude to it or describe it. Make it a command, a clear set of instructions.

Don’t Be Vague: Pretend your speaking to an idiot. Assume nothing. Provide as much information as you can.

Provide the Relevant Context: The more background information you provide, the more the AI has to work with. You may want to start by describing the role the AI would be playing in performing the desired actions, along with any other details you think might be pertinent.

Give Specific Output Instructions: Exactly how do you want your results to be formatted? As a list, a summary, a detailed explanation, or something else? Would you like the tone of the response to be instructive, funny, persuasive?

Give Examples if Possible: To better define the structure and the style of your desired output, provide any examples you may have.

Provide Explicit Constraints: Developers say that the process of adapting a large language model (LLM) to a specific application is a reductive process in which you are telling the LLM what not to respond to and how to further limit the scope of what it provides. Perhaps you want to fit the response within a given word count or paragraph structure.

Be Prepared to Discuss: The AI may not always hit the mark on your first attempt. Be aware that you can respond with follow-up prompts. In this way, you can discuss what you want as it develops. This also means you may want to break your larger prompts down into a series of smaller, more specific prompts.

This is a good starting point, but you will need to practice and learn from your own mistakes. Over time you'll begin to understand how the AI "thinks" and develop a better understanding of how best to express your instructions. Remember, they are not "requests." They are commands.

The Good News

Maybe the best news is that none of this involves any kind of "coding." You are speaking or writing in natural language, just being careful to be as clear and concise as possible. You will get better at it, and your AI will get better at understanding you. This may feel odd at first, but when you become comfortable, you’ll find yourself creating solutions large and small using effective prompts, agentic design studios, and other natural language processing tools designed just for you, the business user.