Skip to main content

Prompt Engineering - Enhancing agent outputs

Prompt engineering involves fine-tuning the text prompts submitted to generative AI. These prompts guide your agent in performing specific tasks and influence the quality of its output. By crafting precise prompts, you can ensure that AI-generated responses align with your expected goals and criteria, reducing the need for extensive post-processing. Prompt engineering is essential for leveraging agents effectively. Whether you are researching for data, creating content, or trying to generate any other type of output, thoughtful prompts make a significant difference.

Where will you use prompts in Devs.ai?​

You'll be using prompts in the following scenarios:

  • Creating an agent - while creating an agent, you can add a description that is used to generate an avatar image and training information. Detailed and specific descriptions will help generate agents that are well-aligned with the purpose you are creating them. You can continue to tweak the instructions to tailor the profile and training information to your requirements.

  • Conversing with an agent - Text-based prompts drive all engagement with AppDirectAIs. Therefore, it is essential to use the most appropriate text to frame these prompts. If you are seeking solutions or answers, phrase your prompts in the form of a question.

How can you improve the effectiveness of prompts?​

Here are some pointers that you can employ to make your prompts more effective:

  • Be specific - add as many specific inputs as possible in your prompts. For instance, if you are using an agent to generate content, input instructions around context, provide examples, and set a desired tone. Move away from generic prompts as these can result in random outputs that may not be related to the context you are looking for.

  • Iterate and revise: getting the most optimal results from an agent is an iterative process and can take several iterations of Q&A. Try to start with basic instructions and build on them to get closer to the output you want.

  • Be mindful of the model's capabilities - consider the strengths and weaknesses of the model that is powering your agent. For example, some models may not be effective for more complex computations or nuanced interactions. Therefore, you may need to break down your objective into simpler prompts to collect the necessary data and build a complete picture.

Example​

Here in this example, the profile image changes dramatically based on the level of detail and context in the input prompt:

Auto-generated prompt text: Dockeeper

Output:

Autoprompt

Revised prompt text: An AI powered assistant app designed to help find answers to questions related to product documentation.

Output:

Detailedprompt

Further reading​

Prompt engineering is a complex subject that requires consideration of various factors to create effective prompts. It is highly recommended to review industry standard guides to gain more detailed insights.