Before the prompt, the problem
Whether it’s a website, an app, or a tool to automate an internal process, artificial intelligence can execute almost anything asked of it very well. The best way to use it is to come to it with clear ideas: know what you want to achieve before describing it. Here are the questions that, in my experience, make the difference between a project that works and one that has to be redone after a few months.
The five questions to ask yourself before starting
- What exactly do I want to happen when someone uses this tool?
- Who will actually use it, and at what point in their day?
- What value does it offer, said in one sentence that needs no explanation?
- Will it have to grow or change over time — and what happens if it does?
- If it will involve multiple languages, markets, or departments, how do I want to manage that complexity from the start?
Answering these questions before opening an editor — or a chat with AI — radically changes the final result, and saves months of corrections. It applies to a showcase website as well as an internal tool or an automation.
AI speeds up execution, it does not replace the questions
Once these answers are clear, artificial intelligence is today the most efficient tool that exists for building quickly: website, app, prototype, automation. The point is not whether to use it — it’s what you bring to it when you start.
Whether you call it design, or writing specifications, the point is not in the technical structure but rather in understanding the problems or the results you want to achieve. It’s similar to choosing a restaurant: we don’t decide where to go to satisfy our hunger, but we normally choose whether to go to a pizzeria, a Chinese restaurant, a Japanese one, or a typical regional restaurant.
We want to satisfy a desire for taste, aromas, and experience. In the same way, before using AI, we need to understand what problem we want to solve and what experience we want to offer ourselves or the end user.
AI is our first end user
It seems curious, but if we want AI to help us, we should explain in chat what experience we want to build, what satisfaction we want to derive or offer our customers. If we don’t do this, AI won’t help us not because it isn’t capable, but because it would have to read our minds. Not being able to do so, it will offer us its average useful solution, not exclusive and even less tailored to our desire.
Technique is important anyway
Once we communicate our desires to AI, we must begin designing the technical structure. In this case too, AI will tend to use the tools it knows best, those that use less energy for implementation, and what it finds best documented online. I myself have realized that when asking an agent to set up the technical structure of a project, the solutions AI proposes to me are strongly conditioned by what I have already used by sharing choices with AI itself, and by what costs less effort in implementation.
To develop a small process that runs only on our computer, it is not necessary to publish the work online and then use it through a browser. Often a small script that runs locally would be enough and requires no online publication. AI, knowing more common solutions better, will tend to propose more complex and expensive solutions than those actually needed. For this reason, it is important to guide AI in choosing the technical structure, explaining to it what the project’s constraints and needs are.
With an eye to the future
Another danger is having AI build an installation that we then cannot adapt to new needs, to the expansion of the project, to its scalability — that is, to the need to make it more powerful and flexible.
In this case it is difficult to tell AI “build something that can change” because indeterminacy is in fact a limit to design. But human beings know, unlike AI, the powerful path of “serendipity” — that is, the path of changing your mind while walking and discovering the context. What to do then? A component-based and modular design helps enormously. Professional developers have used this technique for decades, and it is the only way to be able to implement changes and new features in the project we build.
Anticipating and managing complexity
Anticipating possible growth in terms of complexity and managing this evolution, in my experience, is the key to success even in the smallest projects. Even a small website, which as it evolves might need to implement a complex form, must be able to accommodate this new need with simplicity. If we do not prepare an environment for its evolution, we will have a system that does not respond to our needs.