Don't write the prompt. Get interviewed

Writing prompts can be intimidating. Individuals often glance at a lengthy sample complete with constraints, formats, and edge cases, then close the tab, believing they cannot produce something similar. To overcome this challenge, one should consider an alternative approach.

Instead of directly creating a prompt, engage in an interview with the agent. Request them to delve into the stack, architecture, and the limits of the decisions involved. By posing a series of targeted questions, you can gather valuable insights that would have otherwise required extensive prompt writing.

For instance, inquire about the behavior of the system when the portal is open in multiple tabs simultaneously. This approach helps uncover potential issues that might not be evident through a traditional prompt.

Furthermore, be aware of potential mistakes made by the agent. In this case, the agent initially focused on integrating with PaperLedger before examining the code. Recognize that not all questions are equally important, and prioritize those that address critical aspects of the project.

Once you have gathered sufficient information through the interview, it becomes easier to construct the scenario, define boundaries, and establish contracts between the components. This information acts as a foundation for creating the final prompt, eliminating the need to write it from scratch.

To initiate the process, use a simple "get started" command. The agent will then take charge, building the app, transitioning to the Python portal, and establishing the connection between the two sides. By following this method, the decisions you would have made naturally emerge in the written document.

The entire interview process, including the agent's subsequent actions, can be found in the article: [https://hram.github.io/en/articles/mobile-ledger/](https://hram.github.io/en/articles/mobile-ledger/).