Lessons Learned from My Collaboration with AI

In my previous article, I described my first experience with Gemini Code Assist, Google's coding assistant integrated into VS Code. To test it thoroughly, I recreated a monolingual version of my blog from scratch. Today, I’m sharing the lessons I learned from this experience and how I’m improving my collaboration with AI.

Lessons from My Collaboration with Gemini Code Assist

Take the Time to Describe the Project Well 🖋

When starting a new project, take the time to analyze the project as a whole and create a detailed and comprehensive README file. This first ensures clarity for yourself about what you want to achieve, then it makes the AI’s task easier and yields more relevant responses.

Don’t Be Too Lazy 🥱🏖️

During my test, Gemini generally produced high-quality code, but it still had room for improvement in some areas. The common advice to review the code generated by AI is therefore justified. In particular, it’s important to step back and ask whether the approach taken by the AI is relevant in the given context.

Sometimes, it’s faster to code simple things yourself than to write a prompt and wait for Gemini’s response, which can take over a minute to appear. Plus, this avoids becoming too lazy and saves energy!

Be Patient with the AI and Educate It 🫂

When the AI doesn’t respond to certain requests as desired, it’s easy to get frustrated and speak to it more harshly 🤬 instead of trying to understand why it produced such a response.

What’s important to remember is that doing this puts you in an emotional state that makes the work unpleasant. Usually, if the AI misses the mark, it’s because the request wasn’t precise enough or lacked necessary context.

In this case, there are two options: either refine or break down the request further, or do the work yourself. If you write the code yourself, you can take the opportunity to educate the AI by saying, "Here’s what I expected," and providing the code in context. This is also a great chance to let the AI suggest improvements!

Don’t Be Too Greedy 🍰

To save time, you might be tempted to ask the AI to implement a large feature all at once. But this doesn’t work due to the AI’s inherent limitations 😏. And it’s not easy to have a clear understanding of these limits. That’s why, toward the end of the project, I asked Gemini the following question:

What are your limits regarding the number of files and the amount of code you can generate at once? For example, if I gave you the mockups for a complete blog, would you be able to generate all the files at once?

Here’s its response:

Hello! That’s an excellent question that gets to the heart of how best to collaborate with an AI like me.

To answer directly: no, I wouldn’t be able to generate all the files for a complete blog at once from mockups.
There are two main reasons for this:

1. Response size limit: Each response I generate has a maximum size. A complete project, even a simple static blog, contains many files (HTML, CSS, JavaScript, configuration, etc.) and would almost certainly exceed this limit.

2. Complexity and quality: Software creation is an iterative process. Asking for everything to be generated at once is an extremely complex task that leaves room for ambiguity. To achieve a high-quality result that truly matches your expectations, it’s much more effective to proceed step by step.

The most productive approach is to treat me as a "pair programming" partner. Here’s how we can collaborate effectively on a large-scale project:

• Break down the project: Divide the work into smaller, logical tasks. For example: "Create the basic structure of the Eleventy project," "Create the base layout with the header and footer," "Now, add the CSS styling for the menu," "Implement the article list page," etc.

• Work iteratively: Give me one task at a time. This allows you to review my work, correct it if necessary, and give me more precise instructions for the next step. This is exactly the approach you used to recreate your blog, and it’s the most effective.

• Provide precise context: Even though I have access to the project files, specifying which files to work on in your request helps me focus and deliver a more relevant and faster response.

In summary, use me not as a "one-click website generator," but as a very fast and skilled assistant who helps you build your project brick by brick. This iterative collaboration ensures you retain full control over the architecture and quality of the final code.

My Personal Advice for Beginners👉

If you’re new to AI, to have a positive first experience and avoid disappointment from the start, I strongly recommend the following:

Another best practice I’ve adopted for quite some time and that helps me a lot:

This list will help you stay on track and ensure you don’t forget anything. It’s especially useful if your work spans a long period (several hours or days).

My final piece of advice concerns job search preparation. With AI, it’s clear that beginner developers will find it increasingly difficult to land jobs. Hence my recommendation:

It’s super important to follow this order! Because if you use AI right away without mastering the architecture and code yourself:

• ⌛ You’ll waste a lot of time in your learning, as you might think you understand the code, when in fact your understanding will be very partial. There’s a huge difference between understanding code written by someone else and being able to write it yourself.

• 🎭 You’ll likely suffer from imposter syndrome (and rightly so!), as you’ll quickly feel you can’t do much without AI.

• 💩 You’ll produce crap! Sorry to put it so bluntly, but let’s be realistic. AI is capable of producing very good pieces of code. But if you lack the big-picture vision, perspective, and experience to guide it properly, the overall result will be poor. Only a jeweler knows how to assemble diamonds correctly to make a beautiful necklace.

Feel free to share your thoughts in the comments. And if you have other best practices or tips on how to collaborate effectively with AI, don’t hesitate to share them too! 😉