How I Use AI for Development, Without Delegating Code to It

I don't use AI to go faster.

I use it especially at the beginning and during development, when everything isn't clear yet.

At the start, to lay the foundations: understand the problem, structure the idea, avoid going in the wrong direction. During development, to test the robustness of an existing implementation, reformulate logic, or verify I haven't forgotten anything.

Once the code is in place, I take back control. I adjust, simplify, optimize.

The Most Important Thing: Provide Context

With AI, everything is decided before the first response.

If I simply ask it "do this for me", the result is almost always mediocre. On the other hand, if I take the time to give it all the context, the quality changes completely.

I explain to it:

• the type of project
• the real objective of the feature
• technical constraints
• and above all, my level of requirement

When a response doesn't suit me, I almost always consider it my fault: I wasn't precise enough in the rules I gave it.

Impose Code Rules

I never ask AI to "do its best". I give it clear rules, for example:

• code must be readable above all
• every complexity must have a real justification
• code must remain easy to maintain over time
• prioritize a simple and reliable solution

These rules aren't specific to AI. They're the same ones I try to apply when coding alone.

AI then becomes an extension of my way of working, not a generic code generator.

Decide Rather Than Propose Ten Solutions

I prefer AI to decide and justify, rather than propose multiple solutions.

In most cases, there aren't ten good solutions: there's one that's truly coherent with the context, and others that are "acceptable".

If the proposed solution doesn't suit me, we work together until we arrive at something clean. It's much more effective than comparing a list of theoretical options.

What AI Really Helps Me Do

Where it's most useful for me:

• structure an idea
• reformulate complex logic
• identify edge cases
• analyze an existing implementation

I often use it as a mirror: "Is this approach coherent in the long term?"

It helps me think, not decide.

What I Almost Always Correct

Even when a response "works", I often go back to:

• simplify logic
• optimize certain parts
• remove what seems unnecessary to me

AI doesn't always think like a developer who will maintain their code for years. I do.

Code can work and still be unacceptable in terms of readability or structure. In that case, I correct without hesitation.

AI as Assistant, Never as Pilot

I see AI as a decision-making tool, not an autopilot.

I always keep:

• complete understanding of the code
• responsibility for choices
• the final version

If I don't understand a line, it doesn't stay. Code I don't understand is code I can't maintain.

Conclusion

I use AI especially at the beginning and during development, to think better and better frame my choices. Then, I take back control, adjust, and optimize.

AI is effective when you give it clear rules, context, and real requirements. It becomes useless — even counterproductive — when you delegate responsibility to it.

As long as I keep control, AI remains an excellent tool. But the code, choices, and final quality remain entirely on my side.