In this paper, we explore the application of Large Language Models (LLMs) in the particular context of end-to-end software construction, i.e., in contexts where software developers have a set of requirements and have to design, implement, test, and validate a new software system. Particularly, we report an experiment where we asked three software developers to use ChatGPT to fully implement a Web-based application using mainstream software architectures and technologies. After that, we compare the apps produced by ChatGPT with a reference implementation that we manually implemented for our research. As a result, we document four categories of prompts that can be used by developers in similar contexts, including initialization prompts, feature requests, bug-fixing, and layout prompts. Additionally, we discuss the advantages and disadvantages of two prompt construction approaches: top-down (where we start with a high-level description of the target software, typically in the form of user stories) and bottom-up (where we request the construction of the system feature by feature).